src: add matrix_cubesphere_indexed

drm/matrix.cpp

@@ -1018,7 +1018,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/matrix_cubesphere.cpp

@@ -1105,7 +1105,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/matrix_cubesphere_specular.cpp

@@ -1126,7 +1126,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/matrix_cubesphere_specular_suzanne.cpp

@@ -1188,7 +1188,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/shadertoy.c

@@ -304,18 +304,18 @@ int indirect_buffer(float time)
       | SC_SCISSOR0__YS0(0)
       );
   T0V(SC_SCISSOR1
-      , SC_SCISSOR1__XS1(1600 - 1)
-      | SC_SCISSOR1__YS1(1200 - 1)
+      , SC_SCISSOR1__XS1(800 - 1)
+      | SC_SCISSOR1__YS1(600 - 1)
       );
 
   //////////////////////////////////////////////////////////////////////////////
   // VAP
   //////////////////////////////////////////////////////////////////////////////
 
-  T0Vf(VAP_VPORT_XSCALE,   800.0f);
-  T0Vf(VAP_VPORT_XOFFSET,  800.0f);
-  T0Vf(VAP_VPORT_YSCALE,  -600.0f);
-  T0Vf(VAP_VPORT_YOFFSET,  600.0f);
+  T0Vf(VAP_VPORT_XSCALE,   400.0f);
+  T0Vf(VAP_VPORT_XOFFSET,  400.0f);
+  T0Vf(VAP_VPORT_YSCALE,  -300.0f);
+  T0Vf(VAP_VPORT_YOFFSET,  300.0f);
   T0Vf(VAP_VPORT_ZSCALE,     0.5f);
   T0Vf(VAP_VPORT_ZOFFSET,    0.5f);
 
@@ -409,7 +409,7 @@ int indirect_buffer(float time)
   };
   const int vertex_shader_length = (sizeof (vertex_shader)) / (sizeof (vertex_shader[0]));
   assert(vertex_shader_length % 4 == 0);
-  printf("vs length %d\n", vertex_shader_length);
+  //printf("vs length %d\n", vertex_shader_length);
 
   T0_ONE_REG(VAP_PVS_VECTOR_DATA_REG_128, vertex_shader_length - 1);
   for (int i = 0; i < vertex_shader_length; i++) {
@@ -449,10 +449,17 @@ int indirect_buffer(float time)
 
   // fragment constants
 
+#define PI (3.14159274101257324219f)
+#define PI_2 (PI * 2.0f)
+#define I_PI_2 (1.0f / (PI_2))
+
   const float fragment_consts[] = {
-    time, 0, 0, 0,
+    time, 1.2, 0.01, 0.4,
+    PI_2, I_PI_2, 0, 0,
+    0.25, 0.40625, 0.5625, 0,
   };
   int fragment_consts_length = (sizeof (fragment_consts)) / (sizeof (fragment_consts[0]));
+  assert(fragment_consts_length % 4 == 0);
 
   T0V(GA_US_VECTOR_INDEX
       , GA_US_VECTOR_INDEX__INDEX(0)
@@ -465,16 +472,16 @@ int indirect_buffer(float time)
   // fragment code
 
   const uint32_t fragment_shader[] = {
-    #include "shadertoy_palette.fs.inc"
+    #include "shadertoy_palette_fractal.fs.inc"
   };
   const int fragment_shader_length = (sizeof (fragment_shader)) / (sizeof (fragment_shader[0]));
   assert(fragment_shader_length % 6 == 0);
-  printf("fs length %d\n", fragment_shader_length);
+  //printf("fs length %d\n", fragment_shader_length);
   const int fragment_shader_instructions = fragment_shader_length / 6;
-  printf("fs instructions %d\n", fragment_shader_instructions);
+  //printf("fs instructions %d\n", fragment_shader_instructions);
 
   T0V(US_PIXSIZE
-      , US_PIXSIZE__PIX_SIZE(2) // pixel shader stack frame size
+      , US_PIXSIZE__PIX_SIZE(3) // pixel shader stack frame size
       );
 
   T0V(US_CODE_RANGE
@@ -513,7 +520,7 @@ int indirect_buffer(float time)
     -1.0f,  1.0f, 0.0f
   };
   const int vertices_length = (sizeof (vertices)) / (sizeof (vertices[0]));
-  printf("vtx length %d\n", vertices_length);
+  //printf("vtx length %d\n", vertices_length);
   T3(_3D_DRAW_IMMD_2, (1 + vertices_length) - 1);
   ib[ix++].u32
     = VAP_VF_CNTL__PRIM_TYPE(4)
@@ -708,7 +715,7 @@ int main()
 #define D1GRPH_UPDATE__D1GRPH_SURFACE_UPDATE_PENDING (1 << 2)
 
     uint32_t d1crtc_double_buffer_control = rreg(rmmio, D1CRTC_DOUBLE_BUFFER_CONTROL);
-    printf("D1CRTC_DOUBLE_BUFFER_CONTROL: %08x\n", d1crtc_double_buffer_control);
+    //printf("D1CRTC_DOUBLE_BUFFER_CONTROL: %08x\n", d1crtc_double_buffer_control);
     assert(d1crtc_double_buffer_control == (1 << 8));
 
     // addresses were retrieved from /sys/kernel/debug/radeon_vram_mm

drm/shadertoy_palette_fractal.fs.asm

@@ -0,0 +1,37 @@
+-- CONST[0] = { time, 1.2, 0.01, 0.4 }
+-- CONST[1] = { PI_2, I_PI_2, 0, 0 },
+-- CONST[2] = { 0.25, 0.40625, 0.5625, 0 },
+
+-- temp[0] : { uv0.xy , _, l }
+-- temp[1] : { uv.xy  , _, d }
+-- temp[2] : final_color.xyzw
+-- temp[3] : {col.xyz , i }
+
+-- vec2 uv = uv0; // temp[1]
+src0.rgb = temp[0] : -- uv0
+  temp[1].rg  = MAX src0.rg_ src0.rg_ ;
+
+-- vec4 final_color = vec4(0, 0, 0, 1);
+:
+  temp[2].a   = MAX src0.1 src0.1 ,
+  temp[2].rgb = MAX src0.000 src0.000 ;
+
+-- i = 0;
+:
+  temp[3].a   = MAX src0.0 src0.0 ;
+
+--------------------------------------------------------------------------------
+-- loop start
+--------------------------------------------------------------------------------
+#include "shadertoy_palette_fractal_loop_inner.fs.asm"
+#include "shadertoy_palette_fractal_loop_inner.fs.asm"
+#include "shadertoy_palette_fractal_loop_inner.fs.asm"
+#include "shadertoy_palette_fractal_loop_inner.fs.asm"
+--------------------------------------------------------------------------------
+-- loop end
+--------------------------------------------------------------------------------
+
+OUT TEX_SEM_WAIT
+src0.rgb = temp[2] :
+  out[0].a     = MAX src0.1 src0.1 ,
+  out[0].rgb   = MAX src0.rgb src0.rgb ;

drm/shadertoy_palette_fractal.fs.inc

@@ -0,0 +1,924 @@
+0x00001800,
+0x08020000,
+0x08020080,
+0x00e40720,
+0x00000000,
+0x00000015,
+
+0x00007800,
+0x08020080,
+0x08020080,
+0x00920490,
+0x00c18023,
+0x00000025,
+
+0x00004000,
+0x08020080,
+0x08020080,
+0x00000000,
+0x00810033,
+0x00000000,
+
+0x00001800,
+0x08020001,
+0x080200bc,
+0x00ed8720,
+0x00000000,
+0x00790010,
+
+0x00001800,
+0x08020001,
+0x08020080,
+0x00000720,
+0x00000000,
+0x00000019,
+
+0x00001800,
+0x08020001,
+0x0802c080,
+0x00fb0720,
+0x00000000,
+0x00f6d010,
+
+0x00004000,
+0x08020000,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x00300100,
+0x00000000,
+0x0060e010,
+0x1a000000,
+
+0x00004000,
+0x08020100,
+0x08000500,
+0x00000000,
+0x00600010,
+0x1a000000,
+
+0x00003800,
+0x8802c102,
+0x08020001,
+0x006d86d8,
+0x00000000,
+0x00223030,
+
+0x00003800,
+0x08020003,
+0x08020080,
+0x00000220,
+0x00000000,
+0x00000039,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x004406d8,
+0x00000000,
+0x00a21030,
+
+0x00000800,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000000d,
+0x0000003a,
+
+0x00001000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000400d,
+0x0000003a,
+
+0x00002000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000800d,
+0x0000003a,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x00440221,
+0x00000000,
+0x00221030,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0002c018,
+0x00000000,
+
+0x00004000,
+0x08020001,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08000400,
+0x00000000,
+0x0068c010,
+0x20000000,
+
+0x00004000,
+0x10020080,
+0x08034001,
+0x00000000,
+0x0068c010,
+0x04000000,
+
+0x00004000,
+0x08040480,
+0x0b020001,
+0x00000000,
+0x0028c010,
+0x1c000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c017,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x0802c001,
+0x00000000,
+0x00618010,
+0x5a000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x1800c01c,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0004c01a,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0048c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c019,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0028c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c018,
+0x00000000,
+
+0x00003800,
+0x00220003,
+0x08000480,
+0x006da220,
+0x00000000,
+0x00222020,
+
+0x00004000,
+0x08020080,
+0x08020003,
+0x00000000,
+0x00618030,
+0x30000000,
+
+0x00001800,
+0x08020001,
+0x080200bc,
+0x00ed8720,
+0x00000000,
+0x00790010,
+
+0x00001800,
+0x08020001,
+0x08020080,
+0x00000720,
+0x00000000,
+0x00000019,
+
+0x00001800,
+0x08020001,
+0x0802c080,
+0x00fb0720,
+0x00000000,
+0x00f6d010,
+
+0x00004000,
+0x08020000,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x00300100,
+0x00000000,
+0x0060e010,
+0x1a000000,
+
+0x00004000,
+0x08020100,
+0x08000500,
+0x00000000,
+0x00600010,
+0x1a000000,
+
+0x00003800,
+0x8802c102,
+0x08020001,
+0x006d86d8,
+0x00000000,
+0x00223030,
+
+0x00003800,
+0x08020003,
+0x08020080,
+0x00000220,
+0x00000000,
+0x00000039,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x004406d8,
+0x00000000,
+0x00a21030,
+
+0x00000800,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000000d,
+0x0000003a,
+
+0x00001000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000400d,
+0x0000003a,
+
+0x00002000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000800d,
+0x0000003a,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x00440221,
+0x00000000,
+0x00221030,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0002c018,
+0x00000000,
+
+0x00004000,
+0x08020001,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08000400,
+0x00000000,
+0x0068c010,
+0x20000000,
+
+0x00004000,
+0x10020080,
+0x08034001,
+0x00000000,
+0x0068c010,
+0x04000000,
+
+0x00004000,
+0x08040480,
+0x0b020001,
+0x00000000,
+0x0028c010,
+0x1c000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c017,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x0802c001,
+0x00000000,
+0x00618010,
+0x5a000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x1800c01c,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0004c01a,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0048c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c019,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0028c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c018,
+0x00000000,
+
+0x00003800,
+0x00220003,
+0x08000480,
+0x006da220,
+0x00000000,
+0x00222020,
+
+0x00004000,
+0x08020080,
+0x08020003,
+0x00000000,
+0x00618030,
+0x30000000,
+
+0x00001800,
+0x08020001,
+0x080200bc,
+0x00ed8720,
+0x00000000,
+0x00790010,
+
+0x00001800,
+0x08020001,
+0x08020080,
+0x00000720,
+0x00000000,
+0x00000019,
+
+0x00001800,
+0x08020001,
+0x0802c080,
+0x00fb0720,
+0x00000000,
+0x00f6d010,
+
+0x00004000,
+0x08020000,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x00300100,
+0x00000000,
+0x0060e010,
+0x1a000000,
+
+0x00004000,
+0x08020100,
+0x08000500,
+0x00000000,
+0x00600010,
+0x1a000000,
+
+0x00003800,
+0x8802c102,
+0x08020001,
+0x006d86d8,
+0x00000000,
+0x00223030,
+
+0x00003800,
+0x08020003,
+0x08020080,
+0x00000220,
+0x00000000,
+0x00000039,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x004406d8,
+0x00000000,
+0x00a21030,
+
+0x00000800,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000000d,
+0x0000003a,
+
+0x00001000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000400d,
+0x0000003a,
+
+0x00002000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000800d,
+0x0000003a,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x00440221,
+0x00000000,
+0x00221030,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0002c018,
+0x00000000,
+
+0x00004000,
+0x08020001,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08000400,
+0x00000000,
+0x0068c010,
+0x20000000,
+
+0x00004000,
+0x10020080,
+0x08034001,
+0x00000000,
+0x0068c010,
+0x04000000,
+
+0x00004000,
+0x08040480,
+0x0b020001,
+0x00000000,
+0x0028c010,
+0x1c000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c017,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x0802c001,
+0x00000000,
+0x00618010,
+0x5a000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x1800c01c,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0004c01a,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0048c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c019,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0028c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c018,
+0x00000000,
+
+0x00003800,
+0x00220003,
+0x08000480,
+0x006da220,
+0x00000000,
+0x00222020,
+
+0x00004000,
+0x08020080,
+0x08020003,
+0x00000000,
+0x00618030,
+0x30000000,
+
+0x00001800,
+0x08020001,
+0x080200bc,
+0x00ed8720,
+0x00000000,
+0x00790010,
+
+0x00001800,
+0x08020001,
+0x08020080,
+0x00000720,
+0x00000000,
+0x00000019,
+
+0x00001800,
+0x08020001,
+0x0802c080,
+0x00fb0720,
+0x00000000,
+0x00f6d010,
+
+0x00004000,
+0x08020000,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x00300100,
+0x00000000,
+0x0060e010,
+0x1a000000,
+
+0x00004000,
+0x08020100,
+0x08000500,
+0x00000000,
+0x00600010,
+0x1a000000,
+
+0x00003800,
+0x8802c102,
+0x08020001,
+0x006d86d8,
+0x00000000,
+0x00223030,
+
+0x00003800,
+0x08020003,
+0x08020080,
+0x00000220,
+0x00000000,
+0x00000039,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x004406d8,
+0x00000000,
+0x00a21030,
+
+0x00000800,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000000d,
+0x0000003a,
+
+0x00001000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000400d,
+0x0000003a,
+
+0x00002000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000800d,
+0x0000003a,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x00440221,
+0x00000000,
+0x00221030,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0002c018,
+0x00000000,
+
+0x00004000,
+0x08020001,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08000400,
+0x00000000,
+0x0068c010,
+0x20000000,
+
+0x00004000,
+0x10020080,
+0x08034001,
+0x00000000,
+0x0068c010,
+0x04000000,
+
+0x00004000,
+0x08040480,
+0x0b020001,
+0x00000000,
+0x0028c010,
+0x1c000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c017,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x0802c001,
+0x00000000,
+0x00618010,
+0x5a000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x1800c01c,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0004c01a,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0048c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c019,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0028c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c018,
+0x00000000,
+
+0x00003800,
+0x00220003,
+0x08000480,
+0x006da220,
+0x00000000,
+0x00222020,
+
+0x00004000,
+0x08020080,
+0x08020003,
+0x00000000,
+0x00618030,
+0x30000000,
+
+0x00078005,
+0x08020002,
+0x08020080,
+0x00440220,
+0x00c18003,
+0x00000005,
+

drm/shadertoy_palette_fractal_loop_inner.fs.asm

@@ -0,0 +1,143 @@
+-- uv = uv * 1.5;
+src0.rgb = temp[1] , -- uv
+src0.a = float(60) : -- 1.5
+  temp[1].rg  = MAD src0.rg_ src0.aa_ src0.00_ ;
+
+-- uv = fract(uv);
+src0.rgb = temp[1] : -- uv
+  temp[1].rg  = FRC src0.rg_ ;
+
+-- uv = uv - 0.5;
+src0.rgb = temp[1] ,
+src1.a = float(48) : -- 0.5
+  temp[1].rg  = MAD src0.rg_ src0.11_ -src1.aa_ ;
+
+-- l = length(uv0);
+src0.rgb = temp[0] : -- uv0
+                 DP3 src0.rg0 src0.rg0 ,
+  temp[0].a    = DP ;
+src0.a   = temp[0] :
+  temp[0].a    = RSQ |src0.a| ;
+src0.a   = temp[0] :
+  temp[0].a    = RCP src0.a ;
+
+-- d = i * 0.4 + l;
+src0.a = const[0] , -- 0.4
+src1.a = temp[0]  , -- l
+src2.a = temp[3]  : -- i
+  temp[1].a   = MAD src2.a src0.a src1.a ;
+
+-- d = time * 0.4 + d;
+src0.a = const[0]   , -- 0.4
+src1.a = temp[1]    , -- d
+src0.rgb = const[0] : -- time (r)
+  temp[1].a   = MAD src0.r src0.a src1.a ;
+
+--------------------------------------------------------------------------------
+-- start of 'palette' function
+--------------------------------------------------------------------------------
+
+-- v = d + (vec3(0.25, 0.40625, 0.5625) + 0.5)
+src0.a = temp[1]     , -- d
+src0.rgb = const[2]  , -- vec3(0.25, 0.40625, 0.5625)
+src1.rgb = float(48) , -- 0.5
+srcp.rgb = add       : -- (vec3(0.25, 0.40625, 0.5625) + 0.5)
+  temp[3].rgb = MAD src0.111 src0.aaa srcp.rgb ;
+
+-- v = frac(v)
+src0.rgb = temp[3] : -- v
+  temp[3].rgb = FRC src0.rgb ;
+
+-- v = v - 0.5
+src0.rgb = temp[3]   , -- v
+src1.rgb = float(48) : -- 0.5
+  temp[3].rgb = MAD src0.111 src0.rgb -src1.rgb ;
+
+-- v = cos(v)
+src0.rgb = temp[3] : -- v
+              COS src0.r ,
+  temp[3].r = SOP ;
+src0.rgb = temp[3] : -- v
+              COS src0.g ,
+  temp[3].g = SOP ;
+src0.rgb = temp[3] : -- v
+              COS src0.b ,
+  temp[3].b = SOP ;
+
+-- col = vec3(0.5, 0.5, 0.5) * v + vec3(0.5, 0.5, 0.5)
+src0.rgb = temp[3]   , -- v
+src1.rgb = float(48) : -- 0.5
+  temp[3].rgb = MAD src1.rgb src0.rgb src1.rgb;
+
+--------------------------------------------------------------------------------
+-- end of 'palette' function
+--------------------------------------------------------------------------------
+
+-- d = ex2(-l);
+src0.a = temp[0] : -- l
+  temp[1].a   = EX2 -src0.a ;
+
+-- l = length(uv);
+src0.rgb = temp[1] : -- uv
+                 DP3 src0.rg0 src0.rg0 ,
+  temp[0].a    = DP ;
+src0.a   = temp[0] :
+  temp[0].a    = RSQ |src0.a| ;
+src0.a   = temp[0] :
+  temp[0].a    = RCP src0.a ;
+
+-- d = l * d;
+src0.a = temp[0] , -- l
+src1.a = temp[1] : -- d
+  temp[1].a   = MAD src0.a src1.a src0.0 ;
+
+-- d = d * 8.0 + time;
+src0.a   = temp[1]   , -- d
+src1.a   = float(80) , -- 8.0
+src2.rgb = const[0]  : -- time (r)
+  temp[1].a   = MAD src0.a src1.a src2.r ;
+
+-- d = 0.125 * sin(d); <OMOD>
+-- d = d * 0.159154936671257019043 + 0.5; // 48
+src0.a   = temp[1]   , -- d
+src1.rgb = const[1]  , -- I_PI_2 (g)
+src2.a   = float(48) : -- 0.5
+  temp[1].a  = MAD src0.a src1.g src2.a ;
+-- d = fract(d);
+src0.a   = temp[1] : -- d
+  temp[1].a  = FRC src0.a ;
+-- d = d - 0.5;
+src0.a = temp[1]   , -- d
+src1.a = float(48) : -- 0.5
+  temp[1].a = MAD src0.1 src0.a -src1.a ;
+-- d = 0.125 * sin(d * PI_2);
+src0.a = temp[1] :
+  temp[1].a = 0.125 * SIN src0.a ;
+
+-- d = 1.0 / abs(d);
+src0.a = temp[1] : -- d
+  temp[1].a = RCP |src0.a|;
+
+-- d = 0.01 * d;
+src0.a = temp[1]    , -- d
+src1.rgb = const[0] : -- 0.01 (b)
+  temp[1].a = MAD src0.a src1.b src0.0 ;
+
+-- d = pow(d, 1.2);
+src0.a = temp[1] : -- d
+  temp[1].a = LN2 src0.a ;
+src0.a = temp[1] ,
+src1.rgb = const[0] : -- 1.2 (g)
+  temp[1].a = MAD src0.a src1.g src0.0 ;
+src0.a = temp[1] :
+  temp[1].a = EX2 src0.a ;
+
+-- final_color = col * d + final_color
+src0.rgb = temp[3] , -- col
+src1.a   = temp[1] , -- d
+src2.rgb = temp[2] : -- final_color
+  temp[2].rgb = MAD src0.rgb src1.aaa src2.rgb ;
+
+-- i = i + 1
+src0.a = temp[3] :
+  temp[3].a   = MAD src0.1 src0.a src0.1 ;

drm/shadertoy_palette_fractal_loop_inner.fs.inc

@@ -0,0 +1,224 @@
+0x00001800,
+0x08020001,
+0x080200bc,
+0x00ed8720,
+0x00000000,
+0x00790010,
+
+0x00001800,
+0x08020001,
+0x08020080,
+0x00000720,
+0x00000000,
+0x00000019,
+
+0x00001800,
+0x08020001,
+0x0802c080,
+0x00fb0720,
+0x00000000,
+0x00f6d010,
+
+0x00004000,
+0x08020000,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x00300100,
+0x00000000,
+0x0060e010,
+0x1a000000,
+
+0x00004000,
+0x08020100,
+0x08000500,
+0x00000000,
+0x00600010,
+0x1a000000,
+
+0x00003800,
+0x8802c102,
+0x08020001,
+0x006d86d8,
+0x00000000,
+0x00223030,
+
+0x00003800,
+0x08020003,
+0x08020080,
+0x00000220,
+0x00000000,
+0x00000039,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x004406d8,
+0x00000000,
+0x00a21030,
+
+0x00000800,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000000d,
+0x0000003a,
+
+0x00001000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000400d,
+0x0000003a,
+
+0x00002000,
+0x08020003,
+0x08020080,
+0x00000000,
+0x0000800d,
+0x0000003a,
+
+0x00003800,
+0x0802c003,
+0x08020080,
+0x00440221,
+0x00000000,
+0x00221030,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0002c018,
+0x00000000,
+
+0x00004000,
+0x08020001,
+0x08020080,
+0x00840420,
+0x00000001,
+0x00000001,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0004c00b,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020000,
+0x00000000,
+0x0000c00a,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08000400,
+0x00000000,
+0x0068c010,
+0x20000000,
+
+0x00004000,
+0x10020080,
+0x08034001,
+0x00000000,
+0x0068c010,
+0x04000000,
+
+0x00004000,
+0x08040480,
+0x0b020001,
+0x00000000,
+0x0028c010,
+0x1c000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c017,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x0802c001,
+0x00000000,
+0x00618010,
+0x5a000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x1800c01c,
+0x00000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0004c01a,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0048c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c019,
+0x00000000,
+
+0x00004000,
+0x08040080,
+0x08020001,
+0x00000000,
+0x0028c010,
+0x20000000,
+
+0x00004000,
+0x08020080,
+0x08020001,
+0x00000000,
+0x0000c018,
+0x00000000,
+
+0x00003800,
+0x00220003,
+0x08000480,
+0x006da220,
+0x00000000,
+0x00222020,
+
+0x00004000,
+0x08020080,
+0x08020003,
+0x00000000,
+0x00618030,
+0x30000000,
+

drm/texture.c

@@ -646,7 +646,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_blur.c

@@ -663,7 +663,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_128x128_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_128x128_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_blur_combined.c

@@ -707,7 +707,7 @@ int main()
   intermediate_handle[1] = create_colorbuffer(fd, texture_size, NULL);
 
   {
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
     for (int i = 0; i < texture_size / 4; i++) {
       ((uint32_t*)texturebuffer_ptr)[i] = ((uint32_t*)texture_buf)[i];

drm/texture_cube.c

@@ -765,7 +765,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_cube_clear.c

@@ -935,7 +935,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_cube_clear_zwrite.c

@@ -991,7 +991,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_cube_clear_zwrite_vertex_shader.c

@@ -981,7 +981,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_cube_clear_zwrite_vertex_shader_aos.c

@@ -1040,7 +1040,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_cube_clear_zwrite_vertex_shader_optimize.c

@@ -981,7 +981,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_cube_clear_zwrite_vertex_shader_optimize_zscale.c

@@ -990,7 +990,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_cube_warping.c

@@ -996,7 +996,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

drm/texture_dual.c

@@ -699,10 +699,10 @@ int main()
 
   const int texture_size = 1024 * 1024 * 4;
   texturebuffer_handle[0] = load_texture(fd,
-                                         "../texture/butterfly_1024x1024_argb8888.data",
+                                         "../texture/butterfly_1024x1024_rgba8888.data",
                                          texture_size);
   texturebuffer_handle[1] = load_texture(fd,
-                                         "../texture/bird_1024x1024_argb8888.data",
+                                         "../texture/bird_1024x1024_rgba8888.data",
                                          texture_size);
 
   { // clear colorbuffer

drm/zbuffer_test.c

@@ -972,7 +972,7 @@ int main()
     assert(texturebuffer_ptr != MAP_FAILED);
 
     // copy texture
-    void * texture_buf = read_file("../texture/butterfly_1024x1024_argb8888.data");
+    void * texture_buf = read_file("../texture/butterfly_1024x1024_rgba8888.data");
     assert(texture_buf != NULL);
 
     for (int i = 0; i < texture_size / 4; i++) {

model/cube_indexed.h

@@ -0,0 +1,48 @@
+#pragma once
+
+const int cube_Cube_triangles[] = {
+    0,   1,   2,
+    3,   4,   5,
+    6,   7,   8,
+    9,  10,  11,
+   12,  13,  14,
+   15,  16,  17,
+    0,  18,   1,
+    3,  19,   4,
+    6,  20,   7,
+    9,  21,  10,
+   12,  22,  13,
+   15,  23,  16,
+};
+
+const int cube_Cube_triangles_length = (sizeof (cube_Cube_triangles)) / (sizeof (cube_Cube_triangles[0]));
+
+const float cube_vertices[] = {
+  -1.000000f,  1.000000f, -1.000000f,  0.875000f,  0.500000f, -0.0000f,  1.0000f, -0.0000f,
+   1.000000f,  1.000000f,  1.000000f,  0.625000f,  0.750000f, -0.0000f,  1.0000f, -0.0000f,
+   1.000000f,  1.000000f, -1.000000f,  0.625000f,  0.500000f, -0.0000f,  1.0000f, -0.0000f,
+   1.000000f,  1.000000f,  1.000000f,  0.625000f,  0.750000f, -0.0000f, -0.0000f,  1.0000f,
+  -1.000000f, -1.000000f,  1.000000f,  0.375000f,  1.000000f, -0.0000f, -0.0000f,  1.0000f,
+   1.000000f, -1.000000f,  1.000000f,  0.375000f,  0.750000f, -0.0000f, -0.0000f,  1.0000f,
+  -1.000000f,  1.000000f,  1.000000f,  0.625000f,  0.000000f, -1.0000f, -0.0000f, -0.0000f,
+  -1.000000f, -1.000000f, -1.000000f,  0.375000f,  0.250000f, -1.0000f, -0.0000f, -0.0000f,
+  -1.000000f, -1.000000f,  1.000000f,  0.375000f,  0.000000f, -1.0000f, -0.0000f, -0.0000f,
+   1.000000f, -1.000000f, -1.000000f,  0.375000f,  0.500000f, -0.0000f, -1.0000f, -0.0000f,
+  -1.000000f, -1.000000f,  1.000000f,  0.125000f,  0.750000f, -0.0000f, -1.0000f, -0.0000f,
+  -1.000000f, -1.000000f, -1.000000f,  0.125000f,  0.500000f, -0.0000f, -1.0000f, -0.0000f,
+   1.000000f,  1.000000f, -1.000000f,  0.625000f,  0.500000f,  1.0000f, -0.0000f, -0.0000f,
+   1.000000f, -1.000000f,  1.000000f,  0.375000f,  0.750000f,  1.0000f, -0.0000f, -0.0000f,
+   1.000000f, -1.000000f, -1.000000f,  0.375000f,  0.500000f,  1.0000f, -0.0000f, -0.0000f,
+  -1.000000f,  1.000000f, -1.000000f,  0.625000f,  0.250000f, -0.0000f, -0.0000f, -1.0000f,
+   1.000000f, -1.000000f, -1.000000f,  0.375000f,  0.500000f, -0.0000f, -0.0000f, -1.0000f,
+  -1.000000f, -1.000000f, -1.000000f,  0.375000f,  0.250000f, -0.0000f, -0.0000f, -1.0000f,
+  -1.000000f,  1.000000f,  1.000000f,  0.875000f,  0.750000f, -0.0000f,  1.0000f, -0.0000f,
+  -1.000000f,  1.000000f,  1.000000f,  0.625000f,  1.000000f, -0.0000f, -0.0000f,  1.0000f,
+  -1.000000f,  1.000000f, -1.000000f,  0.625000f,  0.250000f, -1.0000f, -0.0000f, -0.0000f,
+   1.000000f, -1.000000f,  1.000000f,  0.375000f,  0.750000f, -0.0000f, -1.0000f, -0.0000f,
+   1.000000f,  1.000000f,  1.000000f,  0.625000f,  0.750000f,  1.0000f, -0.0000f, -0.0000f,
+   1.000000f,  1.000000f, -1.000000f,  0.625000f,  0.500000f, -0.0000f, -0.0000f, -1.0000f,
+};
+
+const int cube_vertices_length = (sizeof (cube_vertices)) / (sizeof (cube_vertices[0]));
+

model/cubesphere_indexed.h

@@ -0,0 +1,329 @@
+#pragma once
+
+const int cubesphere_Cube_triangles[] = {
+    0,   1,   2,
+    3,   1,   4,
+    1,   5,   2,
+    1,   6,   7,
+    8,   0,   9,
+   10,   4,   0,
+   11,   4,  12,
+   13,   3,  11,
+   14,  15,   3,
+   15,  16,   6,
+    6,  17,  18,
+    7,  18,  19,
+    7,  20,   5,
+    5,  21,  22,
+    2,  22,  23,
+    9,   2,  23,
+   24,  25,  26,
+   27,  25,  28,
+   25,  29,  26,
+   25,  30,  31,
+   32,  24,  33,
+   34,  28,  24,
+   35,  28,  36,
+   21,  27,  35,
+   37,  38,  27,
+   38,  39,  30,
+   30,  40,  41,
+   31,  41,  42,
+   31,  43,  29,
+   29,  44,  45,
+   26,  45,  46,
+   33,  26,  46,
+   47,  48,  49,
+   50,  48,  51,
+   48,  52,  49,
+   48,  53,  54,
+   55,  47,  56,
+   57,  51,  47,
+   58,  51,  59,
+   60,  50,  58,
+   61,  62,  50,
+   62,  63,  53,
+   53,  64,  65,
+   54,  65,  66,
+   54,  67,  52,
+   52,  68,  69,
+   49,  69,  70,
+   56,  49,  70,
+   71,  72,  73,
+   74,  72,  75,
+   72,  76,  73,
+   72,  77,  78,
+   79,  71,  80,
+   81,  75,  71,
+   82,  75,  83,
+   84,  74,  82,
+   85,  86,  74,
+   86,  87,  77,
+   77,  32,  88,
+   78,  88,  89,
+   78,  90,  76,
+   76,  91,  92,
+   73,  92,  93,
+   80,  73,  93,
+   94,  95,  96,
+   97,  95,  98,
+   95,  99,  96,
+   95, 100, 101,
+   84,  94,  85,
+  102,  98,  94,
+  103,  98, 104,
+    8,  97, 103,
+    9, 105,  97,
+  105,  22, 100,
+  100,  21,  35,
+  101,  35,  36,
+  101,  34,  99,
+   99,  32,  87,
+   96,  87, 106,
+   85,  96, 106,
+  107, 108, 109,
+  110, 108, 111,
+  108, 112, 109,
+  108, 113, 114,
+   68, 107, 115,
+   67, 111, 107,
+   65, 111,  66,
+   64, 110,  65,
+  116, 117, 110,
+  117, 118, 113,
+  113,   8, 103,
+  114, 103, 104,
+  114, 102, 112,
+  112,  84, 119,
+  109, 119, 120,
+  115, 109, 120,
+    0,   4,   1,
+    3,  15,   1,
+    1,   7,   5,
+    1,  15,   6,
+    8,  10,   0,
+   10,  12,   4,
+   11,   3,   4,
+   13,  14,   3,
+   14, 121,  15,
+   15, 121,  16,
+    6,  16,  17,
+    7,   6,  18,
+    7,  19,  20,
+    5,  20,  21,
+    2,   5,  22,
+    9,   0,   2,
+   24,  28,  25,
+   27,  38,  25,
+   25,  31,  29,
+   25,  38,  30,
+   32,  34,  24,
+   34,  36,  28,
+   35,  27,  28,
+   21,  37,  27,
+   37, 122,  38,
+   38, 122,  39,
+   30,  39,  40,
+   31,  30,  41,
+   31,  42,  43,
+   29,  43,  44,
+   26,  29,  45,
+   33,  24,  26,
+   47,  51,  48,
+   50,  62,  48,
+   48,  54,  52,
+   48,  62,  53,
+   55,  57,  47,
+   57,  59,  51,
+   58,  50,  51,
+   60,  61,  50,
+   61, 123,  62,
+   62, 123,  63,
+   53,  63,  64,
+   54,  53,  65,
+   54,  66,  67,
+   52,  67,  68,
+   49,  52,  69,
+   56,  47,  49,
+   71,  75,  72,
+   74,  86,  72,
+   72,  78,  76,
+   72,  86,  77,
+   79,  81,  71,
+   81,  83,  75,
+   82,  74,  75,
+   84,  85,  74,
+   85, 106,  86,
+   86, 106,  87,
+   77,  87,  32,
+   78,  77,  88,
+   78,  89,  90,
+   76,  90,  91,
+   73,  76,  92,
+   80,  71,  73,
+   94,  98,  95,
+   97, 105,  95,
+   95, 101,  99,
+   95, 105, 100,
+   84, 102,  94,
+  102, 104,  98,
+  103,  97,  98,
+    8,   9,  97,
+    9,  23, 105,
+  105,  23,  22,
+  100,  22,  21,
+  101, 100,  35,
+  101,  36,  34,
+   99,  34,  32,
+   96,  99,  87,
+   85,  94,  96,
+  107, 111, 108,
+  110, 117, 108,
+  108, 114, 112,
+  108, 117, 113,
+   68,  67, 107,
+   67,  66, 111,
+   65, 110, 111,
+   64, 116, 110,
+  116, 124, 117,
+  117, 124, 118,
+  113, 118,   8,
+  114, 113, 103,
+  114, 104, 102,
+  112, 102,  84,
+  109, 112, 119,
+  115, 107, 109,
+};
+
+const int cubesphere_Cube_triangles_length = (sizeof (cubesphere_Cube_triangles)) / (sizeof (cubesphere_Cube_triangles[0]));
+
+const float cubesphere_vertices[] = {
+   0.316157f,  0.728990f, -0.316157f,  0.687500f,  0.562500f,  0.3362f,  0.8797f, -0.3362f,
+  -0.000000f,  0.839506f, -0.000000f,  0.750000f,  0.625000f, -0.0000f,  1.0000f, -0.0000f,
+   0.333140f,  0.781829f,  0.000000f,  0.687500f,  0.625000f,  0.3553f,  0.9348f, -0.0000f,
+  -0.316157f,  0.728990f, -0.316157f,  0.812500f,  0.562500f, -0.3362f,  0.8797f, -0.3362f,
+   0.000000f,  0.781829f, -0.333140f,  0.750000f,  0.562500f, -0.0000f,  0.9348f, -0.3553f,
+   0.316157f,  0.728990f,  0.316157f,  0.687500f,  0.687500f,  0.3362f,  0.8797f,  0.3362f,
+  -0.316157f,  0.728990f,  0.316157f,  0.812500f,  0.687500f, -0.3362f,  0.8797f,  0.3362f,
+  -0.000000f,  0.781829f,  0.333140f,  0.750000f,  0.687500f, -0.0000f,  0.9348f,  0.3553f,
+   0.500000f,  0.500000f, -0.500000f,  0.625000f,  0.500000f,  0.5774f,  0.5774f, -0.5774f,
+   0.572933f,  0.572933f, -0.296650f,  0.625000f,  0.562500f,  0.6737f,  0.6737f, -0.3038f,
+   0.296650f,  0.572933f, -0.572933f,  0.687500f,  0.500000f,  0.3038f,  0.6737f, -0.6737f,
+  -0.296650f,  0.572933f, -0.572933f,  0.812500f,  0.500000f, -0.3038f,  0.6737f, -0.6737f,
+  -0.000000f,  0.609568f, -0.609568f,  0.750000f,  0.500000f, -0.0000f,  0.7071f, -0.7071f,
+  -0.500000f,  0.500000f, -0.500000f,  0.875000f,  0.500000f, -0.5774f,  0.5774f, -0.5774f,
+  -0.572933f,  0.572933f, -0.296650f,  0.875000f,  0.562500f, -0.6737f,  0.6737f, -0.3038f,
+  -0.333140f,  0.781829f, -0.000000f,  0.812500f,  0.625000f, -0.3553f,  0.9348f, -0.0000f,
+  -0.572933f,  0.572933f,  0.296650f,  0.875000f,  0.687500f, -0.6737f,  0.6737f,  0.3038f,
+  -0.500000f,  0.500000f,  0.500000f,  0.875000f,  0.750000f, -0.5774f,  0.5774f,  0.5774f,
+  -0.296650f,  0.572933f,  0.572933f,  0.812500f,  0.750000f, -0.3038f,  0.6737f,  0.6737f,
+   0.000000f,  0.609568f,  0.609568f,  0.750000f,  0.750000f, -0.0000f,  0.7071f,  0.7071f,
+   0.296650f,  0.572933f,  0.572933f,  0.687500f,  0.750000f,  0.3038f,  0.6737f,  0.6737f,
+   0.500000f,  0.500000f,  0.500000f,  0.625000f,  0.750000f,  0.5774f,  0.5774f,  0.5774f,
+   0.572933f,  0.572933f,  0.296650f,  0.625000f,  0.687500f,  0.6737f,  0.6737f,  0.3038f,
+   0.609568f,  0.609568f, -0.000000f,  0.625000f,  0.625000f,  0.7071f,  0.7071f, -0.0000f,
+   0.316157f, -0.316157f,  0.728990f,  0.437500f,  0.812500f,  0.3362f, -0.3362f,  0.8797f,
+   0.000000f,  0.000000f,  0.839506f,  0.500000f,  0.875000f, -0.0000f, -0.0000f,  1.0000f,
+  -0.000000f, -0.333140f,  0.781829f,  0.437500f,  0.875000f, -0.0000f, -0.3553f,  0.9348f,
+   0.316157f,  0.316157f,  0.728990f,  0.562500f,  0.812500f,  0.3362f,  0.3362f,  0.8797f,
+   0.333140f, -0.000000f,  0.781829f,  0.500000f,  0.812500f,  0.3553f, -0.0000f,  0.9348f,
+  -0.316157f, -0.316157f,  0.728990f,  0.437500f,  0.937500f, -0.3362f, -0.3362f,  0.8797f,
+  -0.316157f,  0.316157f,  0.728990f,  0.562500f,  0.937500f, -0.3362f,  0.3362f,  0.8797f,
+  -0.333140f,  0.000000f,  0.781829f,  0.500000f,  0.937500f, -0.3553f, -0.0000f,  0.9348f,
+   0.500000f, -0.500000f,  0.500000f,  0.375000f,  0.750000f,  0.5774f, -0.5774f,  0.5774f,
+   0.296650f, -0.572933f,  0.572933f,  0.375000f,  0.812500f,  0.3038f, -0.6737f,  0.6737f,
+   0.572933f, -0.296650f,  0.572933f,  0.437500f,  0.750000f,  0.6737f, -0.3038f,  0.6737f,
+   0.572933f,  0.296650f,  0.572933f,  0.562500f,  0.750000f,  0.6737f,  0.3038f,  0.6737f,
+   0.609568f,  0.000000f,  0.609568f,  0.500000f,  0.750000f,  0.7071f, -0.0000f,  0.7071f,
+   0.296650f,  0.572933f,  0.572933f,  0.625000f,  0.812500f,  0.3038f,  0.6737f,  0.6737f,
+   0.000000f,  0.333140f,  0.781829f,  0.562500f,  0.875000f, -0.0000f,  0.3553f,  0.9348f,
+  -0.296650f,  0.572933f,  0.572933f,  0.625000f,  0.937500f, -0.3038f,  0.6737f,  0.6737f,
+  -0.500000f,  0.500000f,  0.500000f,  0.625000f,  1.000000f, -0.5774f,  0.5774f,  0.5774f,
+  -0.572933f,  0.296650f,  0.572933f,  0.562500f,  1.000000f, -0.6737f,  0.3038f,  0.6737f,
+  -0.609568f, -0.000000f,  0.609568f,  0.500000f,  1.000000f, -0.7071f, -0.0000f,  0.7071f,
+  -0.572933f, -0.296650f,  0.572933f,  0.437500f,  1.000000f, -0.6737f, -0.3038f,  0.6737f,
+  -0.500000f, -0.500000f,  0.500000f,  0.375000f,  1.000000f, -0.5774f, -0.5774f,  0.5774f,
+  -0.296650f, -0.572933f,  0.572933f,  0.375000f,  0.937500f, -0.3038f, -0.6737f,  0.6737f,
+  -0.000000f, -0.609568f,  0.609568f,  0.375000f,  0.875000f, -0.0000f, -0.7071f,  0.7071f,
+  -0.728990f, -0.316157f,  0.316157f,  0.437500f,  0.062500f, -0.8797f, -0.3362f,  0.3362f,
+  -0.839506f,  0.000000f, -0.000000f,  0.500000f,  0.125000f, -1.0000f, -0.0000f, -0.0000f,
+  -0.781829f, -0.333140f, -0.000000f,  0.437500f,  0.125000f, -0.9348f, -0.3553f, -0.0000f,
+  -0.728990f,  0.316157f,  0.316157f,  0.562500f,  0.062500f, -0.8797f,  0.3362f,  0.3362f,
+  -0.781829f, -0.000000f,  0.333140f,  0.500000f,  0.062500f, -0.9348f, -0.0000f,  0.3553f,
+  -0.728990f, -0.316157f, -0.316157f,  0.437500f,  0.187500f, -0.8797f, -0.3362f, -0.3362f,
+  -0.728990f,  0.316157f, -0.316157f,  0.562500f,  0.187500f, -0.8797f,  0.3362f, -0.3362f,
+  -0.781829f,  0.000000f, -0.333140f,  0.500000f,  0.187500f, -0.9348f, -0.0000f, -0.3553f,
+  -0.500000f, -0.500000f,  0.500000f,  0.375000f,  0.000000f, -0.5774f, -0.5774f,  0.5774f,
+  -0.572933f, -0.572933f,  0.296650f,  0.375000f,  0.062500f, -0.6737f, -0.6737f,  0.3038f,
+  -0.572933f, -0.296650f,  0.572933f,  0.437500f,  0.000000f, -0.6737f, -0.3038f,  0.6737f,
+  -0.572933f,  0.296650f,  0.572933f,  0.562500f,  0.000000f, -0.6737f,  0.3038f,  0.6737f,
+  -0.609568f, -0.000000f,  0.609568f,  0.500000f,  0.000000f, -0.7071f, -0.0000f,  0.7071f,
+  -0.500000f,  0.500000f,  0.500000f,  0.625000f,  0.000000f, -0.5774f,  0.5774f,  0.5774f,
+  -0.572933f,  0.572933f,  0.296650f,  0.625000f,  0.062500f, -0.6737f,  0.6737f,  0.3038f,
+  -0.781829f,  0.333140f, -0.000000f,  0.562500f,  0.125000f, -0.9348f,  0.3553f, -0.0000f,
+  -0.572933f,  0.572933f, -0.296650f,  0.625000f,  0.187500f, -0.6737f,  0.6737f, -0.3038f,
+  -0.500000f,  0.500000f, -0.500000f,  0.625000f,  0.250000f, -0.5774f,  0.5774f, -0.5774f,
+  -0.572933f,  0.296650f, -0.572933f,  0.562500f,  0.250000f, -0.6737f,  0.3038f, -0.6737f,
+  -0.609568f, -0.000000f, -0.609568f,  0.500000f,  0.250000f, -0.7071f, -0.0000f, -0.7071f,
+  -0.572933f, -0.296650f, -0.572933f,  0.437500f,  0.250000f, -0.6737f, -0.3038f, -0.6737f,
+  -0.500000f, -0.500000f, -0.500000f,  0.375000f,  0.250000f, -0.5774f, -0.5774f, -0.5774f,
+  -0.572933f, -0.572933f, -0.296650f,  0.375000f,  0.187500f, -0.6737f, -0.6737f, -0.3038f,
+  -0.609568f, -0.609568f, -0.000000f,  0.375000f,  0.125000f, -0.7071f, -0.7071f, -0.0000f,
+  -0.316157f, -0.728990f, -0.316157f,  0.187500f,  0.562500f, -0.3362f, -0.8797f, -0.3362f,
+  -0.000000f, -0.839506f,  0.000000f,  0.250000f,  0.625000f, -0.0000f, -1.0000f, -0.0000f,
+  -0.333140f, -0.781829f,  0.000000f,  0.187500f,  0.625000f, -0.3553f, -0.9348f, -0.0000f,
+   0.316157f, -0.728990f, -0.316157f,  0.312500f,  0.562500f,  0.3362f, -0.8797f, -0.3362f,
+  -0.000000f, -0.781829f, -0.333140f,  0.250000f,  0.562500f, -0.0000f, -0.9348f, -0.3553f,
+  -0.316157f, -0.728990f,  0.316157f,  0.187500f,  0.687500f, -0.3362f, -0.8797f,  0.3362f,
+   0.316157f, -0.728990f,  0.316157f,  0.312500f,  0.687500f,  0.3362f, -0.8797f,  0.3362f,
+   0.000000f, -0.781829f,  0.333140f,  0.250000f,  0.687500f, -0.0000f, -0.9348f,  0.3553f,
+  -0.500000f, -0.500000f, -0.500000f,  0.125000f,  0.500000f, -0.5774f, -0.5774f, -0.5774f,
+  -0.572933f, -0.572933f, -0.296650f,  0.125000f,  0.562500f, -0.6737f, -0.6737f, -0.3038f,
+  -0.296650f, -0.572933f, -0.572933f,  0.187500f,  0.500000f, -0.3038f, -0.6737f, -0.6737f,
+   0.296650f, -0.572933f, -0.572933f,  0.312500f,  0.500000f,  0.3038f, -0.6737f, -0.6737f,
+   0.000000f, -0.609568f, -0.609568f,  0.250000f,  0.500000f, -0.0000f, -0.7071f, -0.7071f,
+   0.500000f, -0.500000f, -0.500000f,  0.375000f,  0.500000f,  0.5774f, -0.5774f, -0.5774f,
+   0.572933f, -0.572933f, -0.296650f,  0.375000f,  0.562500f,  0.6737f, -0.6737f, -0.3038f,
+   0.333140f, -0.781829f, -0.000000f,  0.312500f,  0.625000f,  0.3553f, -0.9348f, -0.0000f,
+   0.572933f, -0.572933f,  0.296650f,  0.375000f,  0.687500f,  0.6737f, -0.6737f,  0.3038f,
+   0.296650f, -0.572933f,  0.572933f,  0.312500f,  0.750000f,  0.3038f, -0.6737f,  0.6737f,
+  -0.000000f, -0.609568f,  0.609568f,  0.250000f,  0.750000f, -0.0000f, -0.7071f,  0.7071f,
+  -0.296650f, -0.572933f,  0.572933f,  0.187500f,  0.750000f, -0.3038f, -0.6737f,  0.6737f,
+  -0.500000f, -0.500000f,  0.500000f,  0.125000f,  0.750000f, -0.5774f, -0.5774f,  0.5774f,
+  -0.572933f, -0.572933f,  0.296650f,  0.125000f,  0.687500f, -0.6737f, -0.6737f,  0.3038f,
+  -0.609568f, -0.609568f, -0.000000f,  0.125000f,  0.625000f, -0.7071f, -0.7071f, -0.0000f,
+   0.728990f, -0.316157f, -0.316157f,  0.437500f,  0.562500f,  0.8797f, -0.3362f, -0.3362f,
+   0.839506f, -0.000000f, -0.000000f,  0.500000f,  0.625000f,  1.0000f, -0.0000f, -0.0000f,
+   0.781829f, -0.333140f,  0.000000f,  0.437500f,  0.625000f,  0.9348f, -0.3553f, -0.0000f,
+   0.728990f,  0.316157f, -0.316157f,  0.562500f,  0.562500f,  0.8797f,  0.3362f, -0.3362f,
+   0.781829f, -0.000000f, -0.333140f,  0.500000f,  0.562500f,  0.9348f, -0.0000f, -0.3553f,
+   0.728990f, -0.316157f,  0.316157f,  0.437500f,  0.687500f,  0.8797f, -0.3362f,  0.3362f,
+   0.728990f,  0.316157f,  0.316157f,  0.562500f,  0.687500f,  0.8797f,  0.3362f,  0.3362f,
+   0.781829f,  0.000000f,  0.333140f,  0.500000f,  0.687500f,  0.9348f, -0.0000f,  0.3553f,
+   0.572933f, -0.296650f, -0.572933f,  0.437500f,  0.500000f,  0.6737f, -0.3038f, -0.6737f,
+   0.572933f,  0.296650f, -0.572933f,  0.562500f,  0.500000f,  0.6737f,  0.3038f, -0.6737f,
+   0.609568f,  0.000000f, -0.609568f,  0.500000f,  0.500000f,  0.7071f, -0.0000f, -0.7071f,
+   0.781829f,  0.333140f, -0.000000f,  0.562500f,  0.625000f,  0.9348f,  0.3553f, -0.0000f,
+   0.609568f, -0.609568f, -0.000000f,  0.375000f,  0.625000f,  0.7071f, -0.7071f, -0.0000f,
+  -0.316157f, -0.316157f, -0.728990f,  0.437500f,  0.312500f, -0.3362f, -0.3362f, -0.8797f,
+   0.000000f,  0.000000f, -0.839506f,  0.500000f,  0.375000f, -0.0000f, -0.0000f, -1.0000f,
+   0.000000f, -0.333140f, -0.781829f,  0.437500f,  0.375000f, -0.0000f, -0.3553f, -0.9348f,
+  -0.316157f,  0.316157f, -0.728990f,  0.562500f,  0.312500f, -0.3362f,  0.3362f, -0.8797f,
+  -0.333140f, -0.000000f, -0.781829f,  0.500000f,  0.312500f, -0.3553f, -0.0000f, -0.9348f,
+   0.316157f, -0.316157f, -0.728990f,  0.437500f,  0.437500f,  0.3362f, -0.3362f, -0.8797f,
+   0.316157f,  0.316157f, -0.728990f,  0.562500f,  0.437500f,  0.3362f,  0.3362f, -0.8797f,
+   0.333140f,  0.000000f, -0.781829f,  0.500000f,  0.437500f,  0.3553f, -0.0000f, -0.9348f,
+  -0.296650f, -0.572933f, -0.572933f,  0.375000f,  0.312500f, -0.3038f, -0.6737f, -0.6737f,
+  -0.296650f,  0.572933f, -0.572933f,  0.625000f,  0.312500f, -0.3038f,  0.6737f, -0.6737f,
+  -0.000000f,  0.333140f, -0.781829f,  0.562500f,  0.375000f, -0.0000f,  0.3553f, -0.9348f,
+   0.296650f,  0.572933f, -0.572933f,  0.625000f,  0.437500f,  0.3038f,  0.6737f, -0.6737f,
+   0.296650f, -0.572933f, -0.572933f,  0.375000f,  0.437500f,  0.3038f, -0.6737f, -0.6737f,
+   0.000000f, -0.609568f, -0.609568f,  0.375000f,  0.375000f, -0.0000f, -0.7071f, -0.7071f,
+  -0.609568f,  0.609568f, -0.000000f,  0.875000f,  0.625000f, -0.7071f,  0.7071f, -0.0000f,
+   0.000000f,  0.609568f,  0.609568f,  0.625000f,  0.875000f, -0.0000f,  0.7071f,  0.7071f,
+  -0.609568f,  0.609568f, -0.000000f,  0.625000f,  0.125000f, -0.7071f,  0.7071f, -0.0000f,
+  -0.000000f,  0.609568f, -0.609568f,  0.625000f,  0.375000f, -0.0000f,  0.7071f, -0.7071f,
+};
+
+const int cubesphere_vertices_length = (sizeof (cubesphere_vertices)) / (sizeof (cubesphere_vertices[0]));
+

model/plane.h

@@ -0,0 +1,50 @@
+#pragma once
+
+const vec3 plane_position[] = {
+    {1.000000f, 1.000000f, -0.000000f},
+    {1.000000f, -1.000000f, -0.000000f},
+    {-1.000000f, 1.000000f, 0.000000f},
+    {-1.000000f, -1.000000f, 0.000000f},
+};
+
+const vec2 plane_texture[] = {
+    {1.000000f, 0.000000f},
+    {0.000000f, 1.000000f},
+    {0.000000f, 0.000000f},
+    {1.000000f, 1.000000f},
+};
+
+const vec3 plane_normal[] = {
+    {-0.0000f, -0.0000f, -1.0000f},
+};
+
+const triangle_t plane_Plane_triangle[] = {
+    {
+        {1, 0, 0},
+        {2, 1, 0},
+        {0, 2, 0},
+    },
+    {
+        {1, 0, 0},
+        {3, 3, 0},
+        {2, 1, 0},
+    },
+};
+
+const object plane_Plane = {
+    .triangle = &plane_Plane_triangle[0],
+    .triangle_count = 2,
+};
+
+const object * plane_object[] = {
+    &plane_Plane,
+};
+
+const model plane_model = {
+    .position = plane_position,
+    .texture = plane_texture,
+    .normal = plane_normal,
+    .object = plane_object,
+    .object_count = 1
+};
+

pci_user/main.c

@@ -21,6 +21,41 @@ static inline void wreg(void * rmmio, uint32_t offset, uint32_t value)
   asm volatile ("" ::: "memory");
 }
 
+static inline void wreg_slow(void * rmmio, uint32_t offset, uint32_t value)
+{
+  #define MM_INDEX 0x0
+  #define MM_DATA 0x4
+  wreg(rmmio, MM_INDEX, offset);
+  wreg(rmmio, MM_DATA, value);
+}
+
+static inline uint32_t rreg_slow(void * rmmio, uint32_t offset)
+{
+  wreg(rmmio, MM_INDEX, offset);
+  uint32_t value = rreg(rmmio, MM_DATA);
+  return value;
+}
+
+struct name_address {
+  const char * name;
+  const int address;
+};
+
+const struct name_address display_addresses[] = {
+  #include "../regs/display_registers.inc"
+};
+const int display_addresses_length = (sizeof (display_addresses)) / (sizeof (display_addresses[0]));
+
+const struct name_address memory_controller_addresses[] = {
+  #include "../regs/memory_controller_registers.inc"
+};
+const int memory_controller_addresses_length = (sizeof (memory_controller_addresses)) / (sizeof (memory_controller_addresses[0]));
+
+const struct name_address pcie_addresses[] = {
+  #include "../regs/pcie_registers.inc"
+};
+const int pcie_addresses_length = (sizeof (pcie_addresses)) / (sizeof (pcie_addresses[0]));
+
 int main()
 {
   ////////////////////////////////////////////////////////////////////////
@@ -36,16 +71,47 @@ int main()
 
   void * rmmio = resource2_base;
 
-  uint32_t value1 = rreg(rmmio, 0x6110);
-  printf("[r500] D1GRPH_PRIMARY_SURFACE_ADDRESS %08x\n", value1);
-  uint32_t value2 = rreg(rmmio, 0x6110 + 0x800);
-  printf("[r500] D2GRPH_PRIMARY_SURFACE_ADDRESS %08x\n", value2);
-
-  uint32_t value3 = rreg(rmmio, 0x6118);
-  printf("[r500] D1GRPH_SECONDARY_SURFACE_ADDRESS %08x\n", value3);
-  uint32_t value4 = rreg(rmmio, 0x6118 + 0x800);
-  printf("[r500] D2GRPH_SECONDARY_SURFACE_ADDRESS %08x\n", value4);
-
-  wreg(rmmio, 0x6110, 0x813000);
-  wreg(rmmio, 0x6118, 0x813000);
+  if (1) {
+    for (int i = 0; i < display_addresses_length; i++) {
+      uint32_t value = rreg(rmmio, display_addresses[i].address);
+      printf("%s %x %08x\n", display_addresses[i].name, display_addresses[i].address, value);
+    }
+  }
+
+  if (0) {
+#define MC_IND_INDEX 0x70
+#define MC_IND_INDEX__MC_IND_ADDR(x) (((x) & 0xffff) << 0)
+#define MC_IND_INDEX__MC_IND_ADDR__CLEAR (~0xffff)
+#define MC_IND_DATA 0x74
+    // skip MC_IND_INDEX/MC_IND_DATA
+    const int masks[] = {
+      (1 << 16),
+      (1 << 17),
+      (1 << 20),
+      (1 << 21),
+      (1 << 22),
+    };
+
+    for (int i = 2; i < memory_controller_addresses_length; i++) {
+      const char * name = memory_controller_addresses[i].name;
+      int address = memory_controller_addresses[i].address;
+      int mask = (1 << 16) | (1 << 17) | (1 << 20) | (1 << 21) | (1 << 22);
+      wreg(rmmio, MC_IND_INDEX, MC_IND_INDEX__MC_IND_ADDR(address) | mask);
+      uint32_t value = rreg(rmmio, MC_IND_DATA);
+      wreg(rmmio, MC_IND_INDEX, MC_IND_INDEX__MC_IND_ADDR__CLEAR);
+      printf("%s %x %08x\n", name, address, value);
+    }
+  }
+
+  if (0) {
+#define PCIE_INDEX 0x30
+#define PCIE_DATA 0x38
+    for (int i = 0; i < pcie_addresses_length; i++) {
+      const char * name = pcie_addresses[i].name;
+      int address = pcie_addresses[i].address;
+      wreg_slow(rmmio, PCIE_INDEX, address);
+      uint32_t value = rreg_slow(rmmio, PCIE_DATA);
+      printf("%s %x %08x\n", name, address, value);
+    }
+  }
 }

regs/assembler/error.py

@@ -1,8 +1,10 @@
 import sys
 
-def print_error(filename, buf, e):
+def print_error(e):
     assert len(e.args) == 2, e
     message, token = e.args
+    with open(token.filename, 'rb') as f:
+        buf = f.read()
     lines = buf.splitlines()
     line = lines[token.line - 1]
 
@@ -11,7 +13,7 @@ def print_error(filename, buf, e):
     col_pointer = '^' * len(token.lexeme)
     RED = "\033[0;31m"
     DEFAULT = "\033[0;0m"
-    print(f'File: "{filename}", line {token.line}, column {token.col}\n', file=sys.stderr)
+    print(f'File: "{token.filename}", line {token.line}, column {token.col}\n', file=sys.stderr)
     sys.stderr.write('    ')
     wrote_default = False
     for i, c in enumerate(line.decode('utf-8')):

regs/assembler/fs/__main__.py

@@ -10,8 +10,8 @@ from assembler.fs.validator import validate_instruction
 from assembler.fs.emitter import emit_instruction
 from assembler.error import print_error
 
-def frontend_inner(buf):
-    lexer = Lexer(buf, find_keyword, emit_newlines=False, minus_is_token=True)
+def frontend_inner(filename, buf):
+    lexer = Lexer(filename, buf, find_keyword, emit_newlines=False, minus_is_token=True)
     tokens = list(lexer.lex_tokens())
     parser = Parser(tokens)
     for ins_ast in parser.instructions():
@@ -22,15 +22,15 @@ def frontend_inner(buf):
 
 def frontend(filename, buf):
     try:
-        yield from frontend_inner(buf)
+        yield from frontend_inner(filename, buf)
     except LexerError as e:
-        print_error(filename, buf, e)
+        print_error(e)
         raise
     except ParserError as e:
-        print_error(filename, buf, e)
+        print_error(e)
         raise
     except ValidatorError as e:
-        print_error(filename, buf, e)
+        print_error(e)
         raise
 
 if __name__ == "__main__":

regs/assembler/fs/alu_emitter.py

@@ -17,6 +17,9 @@ def emit_alpha_op(code, alpha_op):
     if alpha_op.dest.omask is not None:
         US_CMN_INST.ALPHA_OMASK(code, alpha_op.dest.omask.value)
 
+    # omod
+    US_ALU_ALPHA_INST.OMOD(code, alpha_op.omod.value)
+
     # opcode
     US_ALU_ALPHA_INST.ALPHA_OP(code, alpha_op.opcode.value)
 
@@ -59,6 +62,9 @@ def emit_rgb_op(code, rgb_op):
     if rgb_op.dest.omask is not None:
         US_CMN_INST.RGB_OMASK(code, rgb_op.dest.omask.value)
 
+    # omod
+    US_ALU_RGB_INST.OMOD(code, rgb_op.omod.value)
+
     # opcode
     US_ALU_RGBA_INST.RGB_OP(code, rgb_op.opcode.value)
 

regs/assembler/fs/alu_validator.py

@@ -110,6 +110,26 @@ class Swizzle(IntEnum):
     one = 6
     unused = 7
 
+class Omod(IntEnum):
+    mul_1 = 0
+    mul_2 = 1
+    mul_4 = 2
+    mul_8 = 3
+    div_2 = 4
+    div_4 = 5
+    div_8 = 6
+    disable = 7
+
+omod_lexemes = OrderedDict([
+    ((b"1", b"0"), Omod.mul_1),
+    ((b"2", b"0"), Omod.mul_2),
+    ((b"4", b"0"), Omod.mul_4),
+    ((b"8", b"0"), Omod.mul_8),
+    ((b"0", b"5"), Omod.div_2),
+    ((b"0", b"25"), Omod.div_4),
+    ((b"0", b"125"), Omod.div_8),
+])
+
 @dataclass
 class SwizzleSel:
     src: SwizzleSelSrc
@@ -119,12 +139,14 @@ class SwizzleSel:
 @dataclass
 class AlphaOperation:
     dest: AlphaDest
+    omod: Omod
     opcode: AlphaOp
     sels: list[SwizzleSel]
 
 @dataclass
 class RGBOperation:
     dest: RGBDest
+    omod: Omod
     opcode: RGBOp
     sels: list[SwizzleSel]
 
@@ -413,14 +435,27 @@ def validate_instruction_operation_sels(swizzle_sels, is_alpha):
         sels.append(SwizzleSel(src, swizzle, mod))
     return sels
 
+def validate_omod_operation(operation):
+    omod = Omod.mul_1
+    if operation.omod != None:
+        integer, decimal = operation.omod
+        key = (integer.lexeme, decimal.lexeme)
+        if key not in omod_lexemes:
+            valid_omods = b", ".join(b".".join(key) for key in omod_lexemes.keys()).decode('utf-8')
+            raise ValidatorError(f"invalid omod, expected one of [{valid_omods}]", integer)
+        omod = omod_lexemes[key]
+    return omod
+
 def validate_alpha_instruction_operation(operation):
     dest = validate_instruction_operation_dest(operation.dest_addr_swizzles,
                                                mask_lookup=alpha_masks,
                                                type_cls=AlphaDest)
+    omod = validate_omod_operation(operation)
     opcode = alpha_op_kws[operation.opcode_keyword.keyword]
     sels = validate_instruction_operation_sels(operation.swizzle_sels, is_alpha=True)
     return AlphaOperation(
         dest,
+        omod,
         opcode,
         sels
     )
@@ -429,10 +464,12 @@ def validate_rgb_instruction_operation(operation):
     dest = validate_instruction_operation_dest(operation.dest_addr_swizzles,
                                                mask_lookup=rgb_masks,
                                                type_cls=RGBDest)
+    omod = validate_omod_operation(operation)
     opcode = rgb_op_kws[operation.opcode_keyword.keyword]
     sels = validate_instruction_operation_sels(operation.swizzle_sels, is_alpha=False)
     return RGBOperation(
         dest,
+        omod,
         opcode,
         sels
     )

regs/assembler/fs/parser.py

@@ -35,6 +35,7 @@ class ALUSwizzleSel:
 @dataclass
 class ALUOperation:
     dest_addr_swizzles: list[DestAddrSwizzle]
+    omod: tuple[Token, Token]
     opcode_keyword: Token
     swizzle_sels: list[ALUSwizzleSel]
 
@@ -115,6 +116,15 @@ class Parser(BaseParser):
             return token.keyword in opcode_keywords
         return False
 
+    def alu_is_omod(self):
+        is_omod = (
+            self.match(TT.identifier, offset=0)
+            and self.match(TT.dot, offset=1)
+            and self.match(TT.identifier, offset=2)
+            and self.match(TT.star, offset=3)
+        )
+        return is_omod
+
     def alu_is_neg(self):
         result = self.match(TT.minus)
         if result:
@@ -154,9 +164,17 @@ class Parser(BaseParser):
 
     def alu_operation(self):
         dest_addr_swizzles = []
-        while not self.alu_is_opcode():
+        while not (self.alu_is_opcode() or self.alu_is_omod()):
             dest_addr_swizzles.append(self.dest_addr_swizzle())
 
+        omod = None
+        if self.alu_is_omod():
+            omod_integer = self.consume(TT.identifier, "expected omod decimal identifier")
+            self.consume(TT.dot, "expected omod decimal dot")
+            omod_decimal = self.consume(TT.identifier, "expected omod decimal identifier")
+            self.consume(TT.star, "expected omod star")
+            omod = (omod_integer, omod_decimal)
+
         opcode_keyword = self.consume(TT.keyword, "expected opcode keyword")
 
         swizzle_sels = []
@@ -165,6 +183,7 @@ class Parser(BaseParser):
 
         return ALUOperation(
             dest_addr_swizzles,
+            omod,
             opcode_keyword,
             swizzle_sels
         )

regs/assembler/fs/tex_validator.py

@@ -55,7 +55,7 @@ def validate_swizzle(token):
     return tuple(swizzles[c] for c in token.lexeme)
 
 def validate_mask_swizzle(token) -> tuple[AlphaMask, RGBMask]:
-    argb_masks = OrderedDict([
+    rgba_masks = OrderedDict([
         (b"none" , (AlphaMask.NONE, RGBMask.NONE)),
         (b"r"    , (AlphaMask.NONE, RGBMask.R)),
         (b"g"    , (AlphaMask.NONE, RGBMask.G)),
@@ -72,9 +72,9 @@ def validate_mask_swizzle(token) -> tuple[AlphaMask, RGBMask]:
         (b"gba"  , (AlphaMask.A, RGBMask.GB)),
         (b"rgba" , (AlphaMask.A, RGBMask.RGB)),
     ])
-    if token.lexeme not in argb_masks:
+    if token.lexeme not in rgba_masks:
         raise ValidatorError("invalid destination mask", token)
-    return argb_masks[token.lexeme]
+    return rgba_masks[token.lexeme]
 
 def validate_masks(ins_ast: parser.TEXInstruction):
     addresses = set()

regs/assembler/lexer.py

@@ -2,6 +2,7 @@ from dataclasses import dataclass
 from enum import Enum, auto
 from itertools import chain
 from typing import Union, Any
+from os import path
 
 DEBUG = True
 
@@ -21,9 +22,11 @@ class TT(Enum):
     bar = auto()
     comma = auto()
     minus = auto()
+    star = auto()
 
 @dataclass
 class Token:
+    filename: str
     start_ix: int
     line: int
     col: int
@@ -42,7 +45,10 @@ class LexerError(Exception):
     pass
 
 class Lexer:
-    def __init__(self, buf: memoryview, find_keyword,
+    def __init__(self,
+                 filename: str,
+                 buf: memoryview,
+                 find_keyword,
                  emit_newlines=False,
                  minus_is_token=False):
         self.start_ix = 0
@@ -53,6 +59,8 @@ class Lexer:
         self.find_keyword = find_keyword
         self.emit_newlines = emit_newlines
         self.minus_is_token = minus_is_token
+        self.filename = filename
+        self.nested_lexer = None
 
     def at_end_p(self):
         return self.current_ix >= len(self.buf)
@@ -73,7 +81,7 @@ class Lexer:
         return self.buf[self.current_ix]
 
     def pos(self):
-        return self.start_ix, self.line, self.col - (self.current_ix - self.start_ix)
+        return self.filename, self.start_ix, self.line, self.col - (self.current_ix - self.start_ix)
 
     def identifier(self):
         while not self.at_end_p() and self.peek() in identifier_characters:
@@ -84,8 +92,26 @@ class Lexer:
         else:
             return Token(*self.pos(), TT.identifier, self.lexeme(), None)
 
+    def include(self, filename):
+        dirname = path.dirname(self.filename)
+        new_filename = path.join(dirname, filename.decode('utf-8'))
+        with open(new_filename, 'rb') as f:
+            buf = f.read()
+        self.nested_lexer = Lexer(new_filename,
+                                  buf,
+                                  find_keyword=self.find_keyword,
+                                  emit_newlines=self.emit_newlines,
+                                  minus_is_token=self.minus_is_token)
+
     def lex_token(self):
         while True:
+            if self.nested_lexer is not None:
+                token = self.nested_lexer.lex_token()
+                if token.type is TT.eof:
+                    self.nested_lexer = None
+                else:
+                    return token
+
             self.start_ix = self.current_ix
 
             if self.at_end_p():
@@ -112,6 +138,8 @@ class Lexer:
                 return Token(*self.pos(), TT.semicolon, self.lexeme())
             elif c == ord(','):
                 return Token(*self.pos(), TT.comma, self.lexeme())
+            elif c == ord('*'):
+                return Token(*self.pos(), TT.star, self.lexeme())
             elif c == ord('-') and self.peek() == ord('-'):
                 self.advance()
                 while not self.at_end_p() and self.peek() != ord('\n'):
@@ -119,8 +147,24 @@ class Lexer:
             elif self.minus_is_token and c == ord('-'):
                 return Token(*self.pos(), TT.minus, self.lexeme())
             elif c == ord('#'):
-                while not self.at_end_p() and self.peek() != ord('\n'):
+                for c in b"include":
+                    o = self.advance()
+                    if o != c:
+                        token = Token(*self.pos(), None, self.lexeme())
+                        raise LexerError(f"unexpected character at line:{self.line} col:{self.col}, expected `#include`", token)
+                while self.peek() == ord(' '):
                     self.advance()
+                self.start_ix = self.current_ix
+                quote = self.advance()
+                if quote != ord('"'):
+                    token = Token(*self.pos(), None, self.lexeme())
+                    raise LexerError(f"unexpected character at line:{self.line} col:{self.col}, expected `\"`", token)
+                self.start_ix = self.current_ix
+                while self.peek() != ord('"'):
+                    self.advance()
+                filename = self.lexeme()
+                assert self.advance() == ord('"')
+                self.include(filename)
             elif c == ord(' ') or c == ord('\r') or c == ord('\t'):
                 pass
             elif c == ord('\n'):

regs/assembler/parser.py

@@ -22,8 +22,8 @@ class BaseParser:
         self.current_ix += 1
         return token
 
-    def match(self, token_type):
-        token = self.peek()
+    def match(self, token_type, offset=0):
+        token = self.peek(offset)
         return token.type == token_type
 
     def match_keyword(self, keyword):

regs/assembler/vs/__main__.py

@@ -22,8 +22,8 @@ out[0].xz    = VE_MAD    input[0].-y-_-0-_ temp[0].x_0_      temp[0].y_0_
 out[0].yw    = VE_MAD    input[0]._x_0     temp[0]._x_0      temp[0]._z_1
 """
 
-def frontend_inner(buf):
-    lexer = Lexer(buf, find_keyword)
+def frontend_inner(filename, buf):
+    lexer = Lexer(filename, buf, find_keyword)
     tokens = list(lexer.lex_tokens())
     parser = Parser(tokens)
     for ins in parser.instructions():
@@ -37,15 +37,15 @@ def frontend_inner(buf):
 
 def frontend(filename, buf):
     try:
-        yield from frontend_inner(buf)
+        yield from frontend_inner(filename, buf)
     except ParserError as e:
-        print_error(input_filename, buf, e)
+        print_error(e)
         raise
     except LexerError as e:
-        print_error(input_filename, buf, e)
+        print_error(e)
         raise
     except ValidatorError as e:
-        print_error(filename, buf, e)
+        print_error(e)
         raise
 
 if __name__ == "__main__":

regs/memory_controller_registers.inc

@@ -0,0 +1,628 @@
+{
+  .name = "MC_IND_INDEX",
+  .address = 0x70,
+},
+{
+  .name = "MC_IND_DATA",
+  .address = 0x74,
+},
+{
+  .name = "MC_STATUS",
+  .address = 0x0,
+},
+{
+  .name = "MC_ARB_MIN",
+  .address = 0x10,
+},
+{
+  .name = "MC_PT0_CNTL",
+  .address = 0x100,
+},
+{
+  .name = "MC_PT0_CONTEXT0_CNTL",
+  .address = 0x102,
+},
+{
+  .name = "MC_PT0_CONTEXT1_CNTL",
+  .address = 0x103,
+},
+{
+  .name = "MC_PT0_CONTEXT2_CNTL",
+  .address = 0x104,
+},
+{
+  .name = "MC_PT0_CONTEXT3_CNTL",
+  .address = 0x105,
+},
+{
+  .name = "MC_PT0_CONTEXT4_CNTL",
+  .address = 0x106,
+},
+{
+  .name = "MC_PT0_CONTEXT5_CNTL",
+  .address = 0x107,
+},
+{
+  .name = "MC_PT0_CONTEXT6_CNTL",
+  .address = 0x108,
+},
+{
+  .name = "MC_PT0_CONTEXT7_CNTL",
+  .address = 0x109,
+},
+{
+  .name = "MC_PT0_SYSTEM_APERTURE_LOW_ADDR",
+  .address = 0x112,
+},
+{
+  .name = "MC_PT0_SYSTEM_APERTURE_HIGH_ADDR",
+  .address = 0x114,
+},
+{
+  .name = "MC_PT0_SURFACE_PROBE",
+  .address = 0x116,
+},
+{
+  .name = "MC_PT0_SURFACE_PROBE_FAULT_STATUS",
+  .address = 0x118,
+},
+{
+  .name = "MC_PT0_PROTECTION_FAULT_STATUS",
+  .address = 0x11a,
+},
+{
+  .name = "MC_PT0_CONTEXT0_DEFAULT_READ_ADDR",
+  .address = 0x11c,
+},
+{
+  .name = "MC_PT0_CONTEXT1_DEFAULT_READ_ADDR",
+  .address = 0x11d,
+},
+{
+  .name = "MC_PT0_CONTEXT2_DEFAULT_READ_ADDR",
+  .address = 0x11e,
+},
+{
+  .name = "MC_PT0_CONTEXT3_DEFAULT_READ_ADDR",
+  .address = 0x11f,
+},
+{
+  .name = "MC_ARB_TIMERS",
+  .address = 0x12,
+},
+{
+  .name = "MC_PT0_CONTEXT4_DEFAULT_READ_ADDR",
+  .address = 0x120,
+},
+{
+  .name = "MC_PT0_CONTEXT5_DEFAULT_READ_ADDR",
+  .address = 0x121,
+},
+{
+  .name = "MC_PT0_CONTEXT6_DEFAULT_READ_ADDR",
+  .address = 0x122,
+},
+{
+  .name = "MC_PT0_CONTEXT7_DEFAULT_READ_ADDR",
+  .address = 0x123,
+},
+{
+  .name = "MC_PT0_CONTEXT0_FLAT_BASE_ADDR",
+  .address = 0x12c,
+},
+{
+  .name = "MC_PT0_CONTEXT1_FLAT_BASE_ADDR",
+  .address = 0x12d,
+},
+{
+  .name = "MC_PT0_CONTEXT2_FLAT_BASE_ADDR",
+  .address = 0x12e,
+},
+{
+  .name = "MC_PT0_CONTEXT3_FLAT_BASE_ADDR",
+  .address = 0x12f,
+},
+{
+  .name = "MC_ARB_DRAM_PENALTIES",
+  .address = 0x13,
+},
+{
+  .name = "MC_PT0_CONTEXT4_FLAT_BASE_ADDR",
+  .address = 0x130,
+},
+{
+  .name = "MC_PT0_CONTEXT5_FLAT_BASE_ADDR",
+  .address = 0x131,
+},
+{
+  .name = "MC_PT0_CONTEXT6_FLAT_BASE_ADDR",
+  .address = 0x132,
+},
+{
+  .name = "MC_PT0_CONTEXT7_FLAT_BASE_ADDR",
+  .address = 0x133,
+},
+{
+  .name = "MC_PT0_CONTEXT0_FLAT_START_ADDR",
+  .address = 0x13c,
+},
+{
+  .name = "MC_PT0_CONTEXT1_FLAT_START_ADDR",
+  .address = 0x13d,
+},
+{
+  .name = "MC_PT0_CONTEXT2_FLAT_START_ADDR",
+  .address = 0x13e,
+},
+{
+  .name = "MC_PT0_CONTEXT3_FLAT_START_ADDR",
+  .address = 0x13f,
+},
+{
+  .name = "MC_ARB_DRAM_PENALTIES2",
+  .address = 0x14,
+},
+{
+  .name = "MC_PT0_CONTEXT4_FLAT_START_ADDR",
+  .address = 0x140,
+},
+{
+  .name = "MC_PT0_CONTEXT5_FLAT_START_ADDR",
+  .address = 0x141,
+},
+{
+  .name = "MC_PT0_CONTEXT6_FLAT_START_ADDR",
+  .address = 0x142,
+},
+{
+  .name = "MC_PT0_CONTEXT7_FLAT_START_ADDR",
+  .address = 0x143,
+},
+{
+  .name = "MC_PT0_CONTEXT0_FLAT_END_ADDR",
+  .address = 0x14c,
+},
+{
+  .name = "MC_PT0_CONTEXT1_FLAT_END_ADDR",
+  .address = 0x14d,
+},
+{
+  .name = "MC_PT0_CONTEXT2_FLAT_END_ADDR",
+  .address = 0x14e,
+},
+{
+  .name = "MC_PT0_CONTEXT3_FLAT_END_ADDR",
+  .address = 0x14f,
+},
+{
+  .name = "MC_ARB_DRAM_PENALTIES3",
+  .address = 0x15,
+},
+{
+  .name = "MC_PT0_CONTEXT4_FLAT_END_ADDR",
+  .address = 0x150,
+},
+{
+  .name = "MC_PT0_CONTEXT5_FLAT_END_ADDR",
+  .address = 0x151,
+},
+{
+  .name = "MC_PT0_CONTEXT6_FLAT_END_ADDR",
+  .address = 0x152,
+},
+{
+  .name = "MC_PT0_CONTEXT7_FLAT_END_ADDR",
+  .address = 0x153,
+},
+{
+  .name = "MC_PT0_CONTEXT0_MULTI_LEVEL_BASE_ADDR",
+  .address = 0x15c,
+},
+{
+  .name = "MC_PT0_CONTEXT1_MULTI_LEVEL_BASE_ADDR",
+  .address = 0x15d,
+},
+{
+  .name = "MC_PT0_CONTEXT2_MULTI_LEVEL_BASE_ADDR",
+  .address = 0x15e,
+},
+{
+  .name = "MC_PT0_CONTEXT3_MULTI_LEVEL_BASE_ADDR",
+  .address = 0x15f,
+},
+{
+  .name = "MC_ARB_RATIO_CLK_SEQ",
+  .address = 0x16,
+},
+{
+  .name = "MC_PT0_CONTEXT4_MULTI_LEVEL_BASE_ADDR",
+  .address = 0x160,
+},
+{
+  .name = "MC_PT0_CONTEXT5_MULTI_LEVEL_BASE_ADDR",
+  .address = 0x161,
+},
+{
+  .name = "MC_PT0_CONTEXT6_MULTI_LEVEL_BASE_ADDR",
+  .address = 0x162,
+},
+{
+  .name = "MC_PT0_CONTEXT7_MULTI_LEVEL_BASE_ADDR",
+  .address = 0x163,
+},
+{
+  .name = "MC_PT0_CLIENT0_CNTL",
+  .address = 0x16c,
+},
+{
+  .name = "MC_PT0_CLIENT1_CNTL",
+  .address = 0x16d,
+},
+{
+  .name = "MC_PT0_CLIENT2_CNTL",
+  .address = 0x16e,
+},
+{
+  .name = "MC_PT0_CLIENT3_CNTL",
+  .address = 0x16f,
+},
+{
+  .name = "MC_ARB_RDWR_SWITCH",
+  .address = 0x17,
+},
+{
+  .name = "MC_PT0_CLIENT4_CNTL",
+  .address = 0x170,
+},
+{
+  .name = "MC_PT0_CLIENT5_CNTL",
+  .address = 0x171,
+},
+{
+  .name = "MC_PT0_CLIENT6_CNTL",
+  .address = 0x172,
+},
+{
+  .name = "MC_PT0_CLIENT7_CNTL",
+  .address = 0x173,
+},
+{
+  .name = "MC_PT0_CLIENT8_CNTL",
+  .address = 0x174,
+},
+{
+  .name = "MC_PT0_CLIENT9_CNTL",
+  .address = 0x175,
+},
+{
+  .name = "MC_PT0_CLIENT10_CNTL",
+  .address = 0x176,
+},
+{
+  .name = "MC_PT0_CLIENT11_CNTL",
+  .address = 0x177,
+},
+{
+  .name = "MC_PT0_CLIENT12_CNTL",
+  .address = 0x178,
+},
+{
+  .name = "MC_PT0_CLIENT13_CNTL",
+  .address = 0x179,
+},
+{
+  .name = "MC_PT0_CLIENT14_CNTL",
+  .address = 0x17a,
+},
+{
+  .name = "MC_PT0_CLIENT15_CNTL",
+  .address = 0x17b,
+},
+{
+  .name = "MC_PT0_CLIENT16_CNTL",
+  .address = 0x17c,
+},
+{
+  .name = "MC_SW_CNTL",
+  .address = 0x18,
+},
+{
+  .name = "MC_TIMING_CNTL_2",
+  .address = 0x3,
+},
+{
+  .name = "MC_WRITE_AGE1",
+  .address = 0x37,
+},
+{
+  .name = "MC_WRITE_AGE2",
+  .address = 0x38,
+},
+{
+  .name = "MC_FB_LOCATION",
+  .address = 0x4,
+},
+{
+  .name = "MC_AGP_LOCATION",
+  .address = 0x5,
+},
+{
+  .name = "AGP_BASE",
+  .address = 0x6,
+},
+{
+  .name = "MC_SEQ_DRAM",
+  .address = 0x60,
+},
+{
+  .name = "MC_SEQ_RAS_TIMING",
+  .address = 0x61,
+},
+{
+  .name = "MC_SEQ_CAS_TIMING",
+  .address = 0x62,
+},
+{
+  .name = "MC_SEQ_MISC_TIMING",
+  .address = 0x63,
+},
+{
+  .name = "MC_SEQ_RD_CTL_I0",
+  .address = 0x64,
+},
+{
+  .name = "MC_SEQ_RD_CTL_I1",
+  .address = 0x65,
+},
+{
+  .name = "MC_SEQ_WR_CTL_I0",
+  .address = 0x66,
+},
+{
+  .name = "MC_SEQ_WR_CTL_I1",
+  .address = 0x67,
+},
+{
+  .name = "MC_SEQ_IO_CTL_I0",
+  .address = 0x68,
+},
+{
+  .name = "MC_SEQ_IO_CTL_I1",
+  .address = 0x69,
+},
+{
+  .name = "MC_SEQ_NPL_CTL_I0",
+  .address = 0x6a,
+},
+{
+  .name = "MC_SEQ_NPL_CTL_I1",
+  .address = 0x6b,
+},
+{
+  .name = "MC_SEQ_CK_PAD_CNTL_I0",
+  .address = 0x6c,
+},
+{
+  .name = "MC_SEQ_CK_PAD_CNTL_I1",
+  .address = 0x6d,
+},
+{
+  .name = "MC_SEQ_CMD_PAD_CNTL_I0",
+  .address = 0x6e,
+},
+{
+  .name = "MC_SEQ_CMD_PAD_CNTL_I1",
+  .address = 0x6f,
+},
+{
+  .name = "AGP_BASE_2",
+  .address = 0x7,
+},
+{
+  .name = "MC_SEQ_DQ_PAD_CNTL_I0",
+  .address = 0x70,
+},
+{
+  .name = "MC_SEQ_DQ_PAD_CNTL_I1",
+  .address = 0x71,
+},
+{
+  .name = "MC_SEQ_QS_PAD_CNTL_I0",
+  .address = 0x72,
+},
+{
+  .name = "MC_SEQ_QS_PAD_CNTL_I1",
+  .address = 0x73,
+},
+{
+  .name = "MC_SEQ_A_PAD_CNTL_I0",
+  .address = 0x74,
+},
+{
+  .name = "MC_SEQ_A_PAD_CNTL_I1",
+  .address = 0x75,
+},
+{
+  .name = "MC_SEQ_CMD",
+  .address = 0x76,
+},
+{
+  .name = "MC_SEQ_STATUS",
+  .address = 0x77,
+},
+{
+  .name = "MC_CNTL0",
+  .address = 0x8,
+},
+{
+  .name = "MC_IO_PAD_CNTL_I0",
+  .address = 0x80,
+},
+{
+  .name = "MC_IO_PAD_CNTL_I1",
+  .address = 0x81,
+},
+{
+  .name = "MC_IO_PAD_CNTL",
+  .address = 0x82,
+},
+{
+  .name = "MC_IO_RD_DQ_CNTL_I0",
+  .address = 0x84,
+},
+{
+  .name = "MC_IO_RD_DQ_CNTL_I1",
+  .address = 0x85,
+},
+{
+  .name = "MC_IO_RD_QS_CNTL_I0",
+  .address = 0x86,
+},
+{
+  .name = "MC_IO_RD_QS_CNTL_I1",
+  .address = 0x87,
+},
+{
+  .name = "MC_IO_WR_CNTL_I0",
+  .address = 0x88,
+},
+{
+  .name = "MC_IO_WR_CNTL_I1",
+  .address = 0x89,
+},
+{
+  .name = "MC_IO_CK_PAD_CNTL_I0",
+  .address = 0x8a,
+},
+{
+  .name = "MC_IO_CK_PAD_CNTL_I1",
+  .address = 0x8b,
+},
+{
+  .name = "MC_IO_CMD_PAD_CNTL_I0",
+  .address = 0x8c,
+},
+{
+  .name = "MC_IO_CMD_PAD_CNTL_I1",
+  .address = 0x8d,
+},
+{
+  .name = "MC_IO_DQ_PAD_CNTL_I0",
+  .address = 0x8e,
+},
+{
+  .name = "MC_IO_DQ_PAD_CNTL_I1",
+  .address = 0x8f,
+},
+{
+  .name = "MC_CNTL1",
+  .address = 0x9,
+},
+{
+  .name = "MC_IO_QS_PAD_CNTL_I0",
+  .address = 0x90,
+},
+{
+  .name = "MC_IO_QS_PAD_CNTL_I1",
+  .address = 0x91,
+},
+{
+  .name = "MC_IO_A_PAD_CNTL_I0",
+  .address = 0x92,
+},
+{
+  .name = "MC_IO_A_PAD_CNTL_I1",
+  .address = 0x93,
+},
+{
+  .name = "MC_IO_WR_DQ_CNTL_I0",
+  .address = 0x94,
+},
+{
+  .name = "MC_IO_WR_DQ_CNTL_I1",
+  .address = 0x95,
+},
+{
+  .name = "MC_IO_WR_QS_CNTL_I0",
+  .address = 0x96,
+},
+{
+  .name = "MC_IO_WR_QS_CNTL_I1",
+  .address = 0x97,
+},
+{
+  .name = "MC_VENDOR_ID_I0",
+  .address = 0x98,
+},
+{
+  .name = "MC_VENDOR_ID_I1",
+  .address = 0x99,
+},
+{
+  .name = "MC_NPL_STATUS_I0",
+  .address = 0x9a,
+},
+{
+  .name = "MC_NPL_STATUS_I1",
+  .address = 0x9b,
+},
+{
+  .name = "MC_IO_RD_QS2_CNTL_I0",
+  .address = 0x9c,
+},
+{
+  .name = "MC_IO_RD_QS2_CNTL_I1",
+  .address = 0x9d,
+},
+{
+  .name = "MC_RFSH_CNTL",
+  .address = 0xa,
+},
+{
+  .name = "MC_IMP_CNTL",
+  .address = 0xa0,
+},
+{
+  .name = "MC_IMP_DEBUG",
+  .address = 0xa1,
+},
+{
+  .name = "MC_IMP_STATUS",
+  .address = 0xa2,
+},
+{
+  .name = "MC_RBS_MAP",
+  .address = 0xb0,
+},
+{
+  .name = "MC_RBS_CZT_HWM",
+  .address = 0xb1,
+},
+{
+  .name = "MC_RBS_SUN_HWM",
+  .address = 0xb2,
+},
+{
+  .name = "MC_RBS_MISC",
+  .address = 0xb3,
+},
+{
+  .name = "MC_PMG_CMD",
+  .address = 0xe0,
+},
+{
+  .name = "MC_PMG_CFG",
+  .address = 0xe1,
+},
+{
+  .name = "MC_MISC_0",
+  .address = 0xf0,
+},
+{
+  .name = "MC_MISC_1",
+  .address = 0xf1,
+},
+{
+  .name = "MC_DEBUG",
+  .address = 0xfe,
+},

regs/memory_controller_registers.txt

@@ -0,0 +1,471 @@
+MC_IND_INDEX
+MCDEC:0x70
+2-2
+MC_IND_DATA
+MCDEC:0x74
+2-2
+MC_STATUS
+MCIND:0x0
+2-2
+MC_ARB_MIN
+MCIND:0x10
+2-8
+MC_PT0_CNTL
+MCIND:0x100
+2-32
+MC_PT0_CONTEXT0_CNTL
+MCIND:0x102
+2-33
+MC_PT0_CONTEXT1_CNTL
+MCIND:0x103
+2-33
+MC_PT0_CONTEXT2_CNTL
+MCIND:0x104
+2-33
+MC_PT0_CONTEXT3_CNTL
+MCIND:0x105
+2-34
+MC_PT0_CONTEXT4_CNTL
+MCIND:0x106
+2-34
+MC_PT0_CONTEXT5_CNTL
+MCIND:0x107
+2-34
+MC_PT0_CONTEXT6_CNTL
+MCIND:0x108
+2-34
+MC_PT0_CONTEXT7_CNTL
+MCIND:0x109
+2-35
+MC_PT0_SYSTEM_APERTURE_LOW_ADDR
+MCIND:0x112
+2-35
+MC_PT0_SYSTEM_APERTURE_HIGH_ADDR
+MCIND:0x114
+2-35
+MC_PT0_SURFACE_PROBE
+MCIND:0x116
+2-35
+MC_PT0_SURFACE_PROBE_FAULT_STATUS
+MCIND:0x118
+2-36
+MC_PT0_PROTECTION_FAULT_STATUS
+MCIND:0x11A
+2-36
+MC_PT0_CONTEXT0_DEFAULT_READ_ADDR
+MCIND:0x11C
+2-36
+MC_PT0_CONTEXT1_DEFAULT_READ_ADDR
+MCIND:0x11D
+2-36
+MC_PT0_CONTEXT2_DEFAULT_READ_ADDR
+MCIND:0x11E
+2-37
+MC_PT0_CONTEXT3_DEFAULT_READ_ADDR
+MCIND:0x11F
+2-37
+MC_ARB_TIMERS
+MCIND:0x12
+2-8
+MC_PT0_CONTEXT4_DEFAULT_READ_ADDR
+MCIND:0x120
+2-37
+MC_PT0_CONTEXT5_DEFAULT_READ_ADDR
+MCIND:0x121
+2-37
+MC_PT0_CONTEXT6_DEFAULT_READ_ADDR
+MCIND:0x122
+2-37
+MC_PT0_CONTEXT7_DEFAULT_READ_ADDR
+MCIND:0x123
+2-37
+MC_PT0_CONTEXT0_FLAT_BASE_ADDR
+MCIND:0x12C
+2-37
+MC_PT0_CONTEXT1_FLAT_BASE_ADDR
+MCIND:0x12D
+2-38
+MC_PT0_CONTEXT2_FLAT_BASE_ADDR
+MCIND:0x12E
+2-38
+MC_PT0_CONTEXT3_FLAT_BASE_ADDR
+MCIND:0x12F
+2-38
+MC_ARB_DRAM_PENALTIES
+MCIND:0x13
+2-8
+MC_PT0_CONTEXT4_FLAT_BASE_ADDR
+MCIND:0x130
+2-38
+MC_PT0_CONTEXT5_FLAT_BASE_ADDR
+MCIND:0x131
+2-38
+MC_PT0_CONTEXT6_FLAT_BASE_ADDR
+MCIND:0x132
+2-38
+MC_PT0_CONTEXT7_FLAT_BASE_ADDR
+MCIND:0x133
+2-39
+MC_PT0_CONTEXT0_FLAT_START_ADDR
+MCIND:0x13C
+2-39
+MC_PT0_CONTEXT1_FLAT_START_ADDR
+MCIND:0x13D
+2-39
+MC_PT0_CONTEXT2_FLAT_START_ADDR
+MCIND:0x13E
+2-39
+MC_PT0_CONTEXT3_FLAT_START_ADDR
+MCIND:0x13F
+2-39
+MC_ARB_DRAM_PENALTIES2
+MCIND:0x14
+2-8
+MC_PT0_CONTEXT4_FLAT_START_ADDR
+MCIND:0x140
+2-39
+MC_PT0_CONTEXT5_FLAT_START_ADDR
+MCIND:0x141
+2-40
+MC_PT0_CONTEXT6_FLAT_START_ADDR
+MCIND:0x142
+2-40
+MC_PT0_CONTEXT7_FLAT_START_ADDR
+MCIND:0x143
+2-40
+MC_PT0_CONTEXT0_FLAT_END_ADDR
+MCIND:0x14C
+2-40
+MC_PT0_CONTEXT1_FLAT_END_ADDR
+MCIND:0x14D
+2-40
+MC_PT0_CONTEXT2_FLAT_END_ADDR
+MCIND:0x14E
+2-40
+MC_PT0_CONTEXT3_FLAT_END_ADDR
+MCIND:0x14F
+2-41
+MC_ARB_DRAM_PENALTIES3
+MCIND:0x15
+2-9
+MC_PT0_CONTEXT4_FLAT_END_ADDR
+MCIND:0x150
+2-41
+MC_PT0_CONTEXT5_FLAT_END_ADDR
+MCIND:0x151
+2-41
+MC_PT0_CONTEXT6_FLAT_END_ADDR
+MCIND:0x152
+2-41
+MC_PT0_CONTEXT7_FLAT_END_ADDR
+MCIND:0x153
+2-41
+MC_PT0_CONTEXT0_MULTI_LEVEL_BASE_ADDR
+MCIND:0x15C
+2-41
+MC_PT0_CONTEXT1_MULTI_LEVEL_BASE_ADDR
+MCIND:0x15D
+2-42
+MC_PT0_CONTEXT2_MULTI_LEVEL_BASE_ADDR
+MCIND:0x15E
+2-42
+MC_PT0_CONTEXT3_MULTI_LEVEL_BASE_ADDR
+MCIND:0x15F
+2-42
+MC_ARB_RATIO_CLK_SEQ
+MCIND:0x16
+2-9
+MC_PT0_CONTEXT4_MULTI_LEVEL_BASE_ADDR
+MCIND:0x160
+2-42
+MC_PT0_CONTEXT5_MULTI_LEVEL_BASE_ADDR
+MCIND:0x161
+2-42
+MC_PT0_CONTEXT6_MULTI_LEVEL_BASE_ADDR
+MCIND:0x162
+2-42
+MC_PT0_CONTEXT7_MULTI_LEVEL_BASE_ADDR
+MCIND:0x163
+2-42
+MC_PT0_CLIENT0_CNTL
+MCIND:0x16C
+2-43
+MC_PT0_CLIENT1_CNTL
+MCIND:0x16D
+2-43
+MC_PT0_CLIENT2_CNTL
+MCIND:0x16E
+2-44
+MC_PT0_CLIENT3_CNTL
+MCIND:0x16F
+2-45
+MC_ARB_RDWR_SWITCH
+MCIND:0x17
+2-9
+MC_PT0_CLIENT4_CNTL
+MCIND:0x170
+2-46
+MC_PT0_CLIENT5_CNTL
+MCIND:0x171
+2-47
+MC_PT0_CLIENT6_CNTL
+MCIND:0x172
+2-48
+MC_PT0_CLIENT7_CNTL
+MCIND:0x173
+2-49
+MC_PT0_CLIENT8_CNTL
+MCIND:0x174
+2-49
+MC_PT0_CLIENT9_CNTL
+MCIND:0x175
+2-50
+MC_PT0_CLIENT10_CNTL
+MCIND:0x176
+2-51
+MC_PT0_CLIENT11_CNTL
+MCIND:0x177
+2-52
+MC_PT0_CLIENT12_CNTL
+MCIND:0x178
+2-53
+MC_PT0_CLIENT13_CNTL
+MCIND:0x179
+2-54
+MC_PT0_CLIENT14_CNTL
+MCIND:0x17A
+2-55
+MC_PT0_CLIENT15_CNTL
+MCIND:0x17B
+2-55
+MC_PT0_CLIENT16_CNTL
+MCIND:0x17C
+2-56
+MC_SW_CNTL
+MCIND:0x18
+2-9
+MC_TIMING_CNTL_2
+MCIND:0x3
+2-3
+MC_WRITE_AGE1
+MCIND:0x37
+2-9
+MC_WRITE_AGE2
+MCIND:0x38
+2-10
+MC_FB_LOCATION
+MCIND:0x4
+2-3
+MC_AGP_LOCATION
+MCIND:0x5
+2-3
+AGP_BASE
+MCIND:0x6
+2-3
+MC_SEQ_DRAM
+MCIND:0x60
+2-10
+MC_SEQ_RAS_TIMING
+MCIND:0x61
+2-11
+MC_SEQ_CAS_TIMING
+MCIND:0x62
+2-11
+MC_SEQ_MISC_TIMING
+MCIND:0x63
+2-12
+MC_SEQ_RD_CTL_I0
+MCIND:0x64
+2-12
+MC_SEQ_RD_CTL_I1
+MCIND:0x65
+2-13
+MC_SEQ_WR_CTL_I0
+MCIND:0x66
+2-14
+MC_SEQ_WR_CTL_I1
+MCIND:0x67
+2-15
+MC_SEQ_IO_CTL_I0
+MCIND:0x68
+2-15
+MC_SEQ_IO_CTL_I1
+MCIND:0x69
+2-15
+MC_SEQ_NPL_CTL_I0
+MCIND:0x6A
+2-16
+MC_SEQ_NPL_CTL_I1
+MCIND:0x6B
+2-16
+MC_SEQ_CK_PAD_CNTL_I0
+MCIND:0x6C
+2-16
+MC_SEQ_CK_PAD_CNTL_I1
+MCIND:0x6D
+2-16
+MC_SEQ_CMD_PAD_CNTL_I0
+MCIND:0x6E
+2-17
+MC_SEQ_CMD_PAD_CNTL_I1
+MCIND:0x6F
+2-17
+AGP_BASE_2
+MCIND:0x7
+2-3
+MC_SEQ_DQ_PAD_CNTL_I0
+MCIND:0x70
+2-17
+MC_SEQ_DQ_PAD_CNTL_I1
+MCIND:0x71
+2-17
+MC_SEQ_QS_PAD_CNTL_I0
+MCIND:0x72
+2-18
+MC_SEQ_QS_PAD_CNTL_I1
+MCIND:0x73
+2-18
+MC_SEQ_A_PAD_CNTL_I0
+MCIND:0x74
+2-18
+MC_SEQ_A_PAD_CNTL_I1
+MCIND:0x75
+2-18
+MC_SEQ_CMD
+MCIND:0x76
+2-19
+MC_SEQ_STATUS
+MCIND:0x77
+2-19
+MC_CNTL0
+MCIND:0x8
+2-4
+MC_IO_PAD_CNTL_I0
+MCIND:0x80
+2-19
+MC_IO_PAD_CNTL_I1
+MCIND:0x81
+2-20
+MC_IO_PAD_CNTL
+MCIND:0x82
+2-21
+MC_IO_RD_DQ_CNTL_I0
+MCIND:0x84
+2-22
+MC_IO_RD_DQ_CNTL_I1
+MCIND:0x85
+2-22
+MC_IO_RD_QS_CNTL_I0
+MCIND:0x86
+2-22
+MC_IO_RD_QS_CNTL_I1
+MCIND:0x87
+2-22
+MC_IO_WR_CNTL_I0
+MCIND:0x88
+2-22
+MC_IO_WR_CNTL_I1
+MCIND:0x89
+2-23
+MC_IO_CK_PAD_CNTL_I0
+MCIND:0x8A
+2-23
+MC_IO_CK_PAD_CNTL_I1
+MCIND:0x8B
+2-23
+MC_IO_CMD_PAD_CNTL_I0
+MCIND:0x8C
+2-23
+MC_IO_CMD_PAD_CNTL_I1
+MCIND:0x8D
+2-24
+MC_IO_DQ_PAD_CNTL_I0
+MCIND:0x8E
+2-24
+MC_IO_DQ_PAD_CNTL_I1
+MCIND:0x8F
+2-24
+MC_CNTL1
+MCIND:0x9
+2-6
+MC_IO_QS_PAD_CNTL_I0
+MCIND:0x90
+2-25
+MC_IO_QS_PAD_CNTL_I1
+MCIND:0x91
+2-25
+MC_IO_A_PAD_CNTL_I0
+MCIND:0x92
+2-25
+MC_IO_A_PAD_CNTL_I1
+MCIND:0x93
+2-26
+MC_IO_WR_DQ_CNTL_I0
+MCIND:0x94
+2-26
+MC_IO_WR_DQ_CNTL_I1
+MCIND:0x95
+2-26
+MC_IO_WR_QS_CNTL_I0
+MCIND:0x96
+2-26
+MC_IO_WR_QS_CNTL_I1
+MCIND:0x97
+2-27
+MC_VENDOR_ID_I0
+MCIND:0x98
+2-27
+MC_VENDOR_ID_I1
+MCIND:0x99
+2-27
+MC_NPL_STATUS_I0
+MCIND:0x9A
+2-27
+MC_NPL_STATUS_I1
+MCIND:0x9B
+2-27
+MC_IO_RD_QS2_CNTL_I0
+MCIND:0x9C
+2-28
+MC_IO_RD_QS2_CNTL_I1
+MCIND:0x9D
+2-28
+MC_RFSH_CNTL
+MCIND:0xA
+2-8
+MC_IMP_CNTL
+MCIND:0xA0
+2-28
+MC_IMP_DEBUG
+MCIND:0xA1
+2-28
+MC_IMP_STATUS
+MCIND:0xA2
+2-28
+MC_RBS_MAP
+MCIND:0xB0
+2-29
+MC_RBS_CZT_HWM
+MCIND:0xB1
+2-30
+MC_RBS_SUN_HWM
+MCIND:0xB2
+2-30
+MC_RBS_MISC
+MCIND:0xB3
+2-30
+MC_PMG_CMD
+MCIND:0xE0
+2-31
+MC_PMG_CFG
+MCIND:0xE1
+2-31
+MC_MISC_0
+MCIND:0xF0
+2-31
+MC_MISC_1
+MCIND:0xF1
+2-31
+MC_DEBUG
+MCIND:0xFE
+2-32

regs/parse_packets.py

@@ -8,6 +8,7 @@ with open(sys.argv[1]) as f:
     values = [
         int(s.strip(), 16)
         for s in f.read().strip().split(",")
+        if s
     ]
 
 undocumented_registers = {
@@ -15,6 +16,17 @@ undocumented_registers = {
     0x2184: "VAP_VSM_VTX_ASSM",
 }
 
+def decode_print(register_name, value, paren=False, display_register_name=None):
+    if display_register_name is None:
+        display_register_name = register_name
+    decoded_value = decode_bits(register_name, value)
+    head = decoded_value[0][2:]
+    tail = indent('\n'.join([f"= {head}", *decoded_value[1:]]), '      ')
+    if paren:
+        print(f"  ({display_register_name})\n{tail}")
+    else:
+        print(f"  {display_register_name}\n{tail}")
+
 class Parser:
     def __init__(self, values):
         self.ix = 0
@@ -52,10 +64,7 @@ class Parser:
                 try:
                     if one_reg or value == 0:
                         assert False
-                    decoded_value = decode_bits(register_name, value)
-                    head = decoded_value[0][2:]
-                    tail = indent('\n'.join(decoded_value[1:]), '      ')
-                    print(f"  {register_name} = {head}\n{tail}")
+                    decode_print(register_name, value)
                 except AssertionError:
                     print(f"  {register_name} = 0x{value:08x}")
             else:
@@ -79,11 +88,83 @@ class Parser:
         it_opcode = (header >> 8) & 0xff
         count = (header >> 16) & 0x3fff
 
-        print(f"type 3: op:{it_opcode:02x} count:{count:04x}")
+        opcode_names = dict((v, k) for k, v in [
+            ("3D_DRAW_VBUF", 0x28),
+            ("3D_DRAW_IMMD", 0x29),
+            ("3D_DRAW_INDX", 0x2A),
+            ("LOAD_PALETTE", 0x2C),
+            ("3D_LOAD_VBPNTR", 0x2F),
+            ("INDX_BUFFER", 0x33),
+            ("3D_DRAW_VBUF_2", 0x34),
+            ("3D_DRAW_IMMD_2", 0x35),
+            ("3D_DRAW_INDX_2", 0x36),
+            ("3D_CLEAR_HIZ", 0x37),
+            ("3D_DRAW_128", 0x39),
+        ])
+        opcode_name = f"{it_opcode:02x}" if it_opcode not in opcode_names else opcode_names[it_opcode]
+
+        registers = {
+            "3D_DRAW_VBUF": ["VAP_VTX_FMT", "VAP_VF_CNTL"],
+            "3D_DRAW_IMMD": ["VAP_VTX_FMT", "VAP_VF_CNTL"],
+            "3D_DRAW_INDX": ["VAP_VTX_FMT", "VAP_VF_CNTL"],
+            "3D_LOAD_VBPNTR": ["VAP_VTX_NUM_ARRAYS",
+                               "VAP_VTX_AOS_ATTR01",
+                               "VAP_VTX_AOS_ADDR0",
+                               "VAP_VTX_AOS_ADDR1",
+                               "VAP_VTX_AOS_ATTR23",
+                               "VAP_VTX_AOS_ADDR2",
+                               "VAP_VTX_AOS_ADDR3",
+                               "VAP_VTX_AOS_ATTR45",
+                               "VAP_VTX_AOS_ADDR4",
+                               "VAP_VTX_AOS_ADDR5",
+                               "VAP_VTX_AOS_ATTR67",
+                               "VAP_VTX_AOS_ADDR6",
+                               "VAP_VTX_AOS_ADDR7",
+                               "VAP_VTX_AOS_ATTR89",
+                               "VAP_VTX_AOS_ADDR8",
+                               "VAP_VTX_AOS_ADDR9",
+                               "VAP_VTX_AOS_ATTR1011",
+                               "VAP_VTX_AOS_ADDR10",
+                               "VAP_VTX_AOS_ADDR11",
+                               "VAP_VTX_AOS_ATTR1213",
+                               "VAP_VTX_AOS_ADDR12",
+                               "VAP_VTX_AOS_ADDR13",
+                               "VAP_VTX_AOS_ATTR1415",
+                               "VAP_VTX_AOS_ADDR14",
+                               "VAP_VTX_AOS_ADDR15"],
+            "INDX_BUFFER": [(("ONE_REG_WR", (31, 31)), ("SKIP_COUNT", (18, 16)), ("DESTINATION", (12, 0))),
+                            (("BUFFER_BASE", (31, 0)),),
+                            (("BUFFER_SIZE", (31, 0)),)],
+            "3D_DRAW_VBUF_2": ["VAP_VF_CNTL"],
+            "3D_DRAW_IMMD_2": ["VAP_VF_CNTL"],
+            "3D_DRAW_INDX_2": ["VAP_VF_CNTL"],
+        }
+
+        print(f"type 3: op:{opcode_name} count:{count:04x}")
+        ix = 0
         while count >= 0:
             value = self.consume()
+            if opcode_name in registers and ix < len(registers[opcode_name]):
+                register_name = registers[opcode_name][ix]
+                if type(register_name) is str:
+                    if "_AOS_ATTR" in register_name:
+                        decode_print(register_name[:-2], value, paren=True, display_register_name=register_name)
+                    elif "_AOS_ADDR" in register_name:
+                        decode_print(register_name[:-1], value, paren=True, display_register_name=register_name)
+                    else:
+                        decode_print(register_name, value, paren=True)
+                else:
+                    print(f"  ({opcode_name}__{ix})")
+                    for i, desc in enumerate(register_name):
+                        eq_bar = '=' if i == 0 else '|'
+                        d_name, (high, low) = desc
+                        mask = (1 << ((high - low) + 1)) - 1
+                        v = (value >> low) & mask
+                        print(f'      {eq_bar} {opcode_name}__{ix}__{d_name}({v})')
+            else:
                 print(f"    {value:08x}")
             count -= 1
+            ix += 1
 
     def packet(self):
         value = self.peek()

regs/pcie_registers.inc

@@ -0,0 +1,648 @@
+{
+  .name = "PCIE_TX_CNTL",
+  .address = 0x1,
+},
+{
+  .name = "PCIE_TX_GART_CNTL",
+  .address = 0x10,
+},
+{
+  .name = "PCIE_TX_GART_DISCARD_RD_ADDR_LO",
+  .address = 0x11,
+},
+{
+  .name = "PCIE_TX_GART_DISCARD_RD_ADDR_HI",
+  .address = 0x12,
+},
+{
+  .name = "PCIE_TX_GART_BASE",
+  .address = 0x13,
+},
+{
+  .name = "PCIE_TX_GART_START_LO",
+  .address = 0x14,
+},
+{
+  .name = "PCIE_TX_GART_START_HI",
+  .address = 0x15,
+},
+{
+  .name = "PCIE_TX_GART_END_LO",
+  .address = 0x16,
+},
+{
+  .name = "PCIE_TX_GART_END_HI",
+  .address = 0x17,
+},
+{
+  .name = "PCIE_TX_GART_ERROR",
+  .address = 0x18,
+},
+{
+  .name = "PCIE_TX_SEQ",
+  .address = 0x2,
+},
+{
+  .name = "PCIE_TX_GART_LRU_MRU_PTR",
+  .address = 0x20,
+},
+{
+  .name = "PCIE_TX_GART_STATUS",
+  .address = 0x21,
+},
+{
+  .name = "PCIE_TX_GART_TLB_VALID",
+  .address = 0x22,
+},
+{
+  .name = "PCIE_TX_GART_TLB0_DATA",
+  .address = 0x23,
+},
+{
+  .name = "PCIE_TX_GART_TLB1_DATA",
+  .address = 0x24,
+},
+{
+  .name = "PCIE_TX_GART_TLB2_DATA",
+  .address = 0x25,
+},
+{
+  .name = "PCIE_TX_GART_TLB3_DATA",
+  .address = 0x26,
+},
+{
+  .name = "PCIE_TX_GART_TLB4_DATA",
+  .address = 0x27,
+},
+{
+  .name = "PCIE_TX_GART_TLB5_DATA",
+  .address = 0x28,
+},
+{
+  .name = "PCIE_TX_GART_TLB6_DATA",
+  .address = 0x29,
+},
+{
+  .name = "PCIE_TX_GART_TLB7_DATA",
+  .address = 0x2a,
+},
+{
+  .name = "PCIE_TX_GART_TLB8_DATA",
+  .address = 0x2b,
+},
+{
+  .name = "PCIE_TX_GART_TLB9_DATA",
+  .address = 0x2c,
+},
+{
+  .name = "PCIE_TX_GART_TLB10_DATA",
+  .address = 0x2d,
+},
+{
+  .name = "PCIE_TX_GART_TLB11_DATA",
+  .address = 0x2e,
+},
+{
+  .name = "PCIE_TX_GART_TLB12_DATA",
+  .address = 0x2f,
+},
+{
+  .name = "PCIE_TX_REPLAY",
+  .address = 0x3,
+},
+{
+  .name = "PCIE_TX_GART_TLB13_DATA",
+  .address = 0x30,
+},
+{
+  .name = "PCIE_TX_GART_TLB14_DATA",
+  .address = 0x31,
+},
+{
+  .name = "PCIE_TX_GART_TLB15_DATA",
+  .address = 0x32,
+},
+{
+  .name = "PCIE_TX_GART_TLB16_DATA",
+  .address = 0x33,
+},
+{
+  .name = "PCIE_TX_GART_TLB17_DATA",
+  .address = 0x34,
+},
+{
+  .name = "PCIE_TX_GART_TLB18_DATA",
+  .address = 0x35,
+},
+{
+  .name = "PCIE_TX_GART_TLB19_DATA",
+  .address = 0x36,
+},
+{
+  .name = "PCIE_TX_GART_TLB20_DATA",
+  .address = 0x37,
+},
+{
+  .name = "PCIE_TX_GART_TLB21_DATA",
+  .address = 0x38,
+},
+{
+  .name = "PCIE_TX_GART_TLB22_DATA",
+  .address = 0x39,
+},
+{
+  .name = "PCIE_TX_GART_TLB23_DATA",
+  .address = 0x3a,
+},
+{
+  .name = "PCIE_TX_GART_TLB24_DATA",
+  .address = 0x3b,
+},
+{
+  .name = "PCIE_TX_GART_TLB25_DATA",
+  .address = 0x3c,
+},
+{
+  .name = "PCIE_TX_GART_TLB26_DATA",
+  .address = 0x3d,
+},
+{
+  .name = "PCIE_TX_GART_TLB27_DATA",
+  .address = 0x3e,
+},
+{
+  .name = "PCIE_TX_GART_TLB28_DATA",
+  .address = 0x3f,
+},
+{
+  .name = "PCIE_TX_CREDITS_CONSUMED",
+  .address = 0x4,
+},
+{
+  .name = "PCIE_TX_GART_TLB29_DATA",
+  .address = 0x40,
+},
+{
+  .name = "PCIE_CLK_CNTL",
+  .address = 0x400,
+},
+{
+  .name = "PCIE_PRBS10",
+  .address = 0x401,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT0",
+  .address = 0x402,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT1",
+  .address = 0x403,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT2",
+  .address = 0x404,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT3",
+  .address = 0x405,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT4",
+  .address = 0x406,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT5",
+  .address = 0x407,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT6",
+  .address = 0x408,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT7",
+  .address = 0x409,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT8",
+  .address = 0x40a,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT9",
+  .address = 0x40b,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT10",
+  .address = 0x40c,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT11",
+  .address = 0x40d,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT12",
+  .address = 0x40e,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT13",
+  .address = 0x40f,
+},
+{
+  .name = "PCIE_TX_GART_TLB30_DATA",
+  .address = 0x41,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT14",
+  .address = 0x410,
+},
+{
+  .name = "PCIE_PRBS23_BITCNT15",
+  .address = 0x411,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT0",
+  .address = 0x412,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT1",
+  .address = 0x413,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT2",
+  .address = 0x414,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT3",
+  .address = 0x415,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT4",
+  .address = 0x416,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT5",
+  .address = 0x417,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT6",
+  .address = 0x418,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT7",
+  .address = 0x419,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT8",
+  .address = 0x41a,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT9",
+  .address = 0x41b,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT10",
+  .address = 0x41c,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT11",
+  .address = 0x41d,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT12",
+  .address = 0x41e,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT13",
+  .address = 0x41f,
+},
+{
+  .name = "PCIE_TX_GART_TLB31_DATA",
+  .address = 0x42,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT14",
+  .address = 0x420,
+},
+{
+  .name = "PCIE_PRBS23_ERRCNT15",
+  .address = 0x421,
+},
+{
+  .name = "PCIE_PRBS23_CTRL0",
+  .address = 0x422,
+},
+{
+  .name = "PCIE_PRBS23_CTRL1",
+  .address = 0x423,
+},
+{
+  .name = "PCIE_PRBS_EN",
+  .address = 0x424,
+},
+{
+  .name = "PCIE_XSTRAP1",
+  .address = 0x425,
+},
+{
+  .name = "PCIE_XSTRAP2",
+  .address = 0x426,
+},
+{
+  .name = "PCIE_XSTRAP5",
+  .address = 0x429,
+},
+{
+  .name = "PCIE_TX_CREDITS_CONSUMED_D",
+  .address = 0x5,
+},
+{
+  .name = "PCIE_TX_CREDITS_CONSUMED_CPLD",
+  .address = 0x6,
+},
+{
+  .name = "PCIE_FLOW_CNTL",
+  .address = 0x60,
+},
+{
+  .name = "PCIE_TXRX_DEBUG_SEQNUM",
+  .address = 0x61,
+},
+{
+  .name = "PCIE_TXRX_TEST_MODE",
+  .address = 0x62,
+},
+{
+  .name = "PCIE_TX_CREDITS_LIMIT",
+  .address = 0x7,
+},
+{
+  .name = "PCIE_RX_CNTL",
+  .address = 0x70,
+},
+{
+  .name = "PCIE_RX_NUM_NACK",
+  .address = 0x71,
+},
+{
+  .name = "PCIE_RX_NUM_NACK_GENERATED",
+  .address = 0x72,
+},
+{
+  .name = "PCIE_RX_ACK_NACK_LATENCY",
+  .address = 0x73,
+},
+{
+  .name = "PCIE_RX_ACK_NACK_LATENCY_THRESHOLD",
+  .address = 0x74,
+},
+{
+  .name = "PCIE_RX_TLP_HDR0",
+  .address = 0x75,
+},
+{
+  .name = "PCIE_RX_TLP_HDR1",
+  .address = 0x76,
+},
+{
+  .name = "PCIE_RX_TLP_HDR2",
+  .address = 0x77,
+},
+{
+  .name = "PCIE_RX_TLP_HDR3",
+  .address = 0x78,
+},
+{
+  .name = "PCIE_RX_TLP_HDR4",
+  .address = 0x79,
+},
+{
+  .name = "PCIE_RX_TLP_CRC",
+  .address = 0x7a,
+},
+{
+  .name = "PCIE_RX_DLP0",
+  .address = 0x7b,
+},
+{
+  .name = "PCIE_RX_DLP1",
+  .address = 0x7c,
+},
+{
+  .name = "PCIE_RX_DLP_CRC",
+  .address = 0x7d,
+},
+{
+  .name = "PCIE_RX_CREDITS_ALLOCATED",
+  .address = 0x7e,
+},
+{
+  .name = "PCIE_RX_CREDITS_ALLOCATED_D",
+  .address = 0x7f,
+},
+{
+  .name = "PCIE_TX_CREDITS_LIMIT_D",
+  .address = 0x8,
+},
+{
+  .name = "PCIE_RX_CREDITS_ALLOCATED_CPLD",
+  .address = 0x80,
+},
+{
+  .name = "PCIE_RX_CREDITS_RECEIVED",
+  .address = 0x81,
+},
+{
+  .name = "PCIE_RX_CREDITS_RECEIVED_D",
+  .address = 0x82,
+},
+{
+  .name = "PCIE_RX_CREDITS_RECEIVED_CPLD",
+  .address = 0x83,
+},
+{
+  .name = "PCIE_RX_MAL_TLP_COUNT",
+  .address = 0x84,
+},
+{
+  .name = "PCIE_RX_ERR_LOG",
+  .address = 0x85,
+},
+{
+  .name = "PCIE_RX_EXPECTED_SEQNUM",
+  .address = 0x86,
+},
+{
+  .name = "PCIE_TX_CREDITS_LIMIT_CPLD",
+  .address = 0x9,
+},
+{
+  .name = "PCIE_CI_CNTL",
+  .address = 0x90,
+},
+{
+  .name = "PCIE_CI_FLUSH_CNTL",
+  .address = 0x91,
+},
+{
+  .name = "PCIE_CI_PANIC",
+  .address = 0x92,
+},
+{
+  .name = "PCIE_CI_HANG",
+  .address = 0x93,
+},
+{
+  .name = "PCIE_LC_CNTL",
+  .address = 0xa0,
+},
+{
+  .name = "PCIE_LC_N_FTS_CNTL",
+  .address = 0xa1,
+},
+{
+  .name = "PCIE_LC_LINK_WIDTH_CNTL",
+  .address = 0xa2,
+},
+{
+  .name = "PCIE_LC_STATE0",
+  .address = 0xa5,
+},
+{
+  .name = "PCIE_LC_STATE1",
+  .address = 0xa6,
+},
+{
+  .name = "PCIE_LC_STATE2",
+  .address = 0xa7,
+},
+{
+  .name = "PCIE_LC_STATE3",
+  .address = 0xa8,
+},
+{
+  .name = "PCIE_LC_STATE4",
+  .address = 0xa9,
+},
+{
+  .name = "PCIE_LC_STATE5",
+  .address = 0xaa,
+},
+{
+  .name = "PCIE_LC_FORCE_SYNC_LOSS_CNTL",
+  .address = 0xab,
+},
+{
+  .name = "PCIE_P_CNTL",
+  .address = 0xb0,
+},
+{
+  .name = "PCIE_P_CNTL2",
+  .address = 0xb1,
+},
+{
+  .name = "PCIE_P_BUF_STATUS",
+  .address = 0xb2,
+},
+{
+  .name = "PCIE_P_DECODER_STATUS",
+  .address = 0xb3,
+},
+{
+  .name = "PCIE_P_MISC_DEBUG_STATUS",
+  .address = 0xb4,
+},
+{
+  .name = "PCIE_P_IMP_CNTL_STRENGTH",
+  .address = 0xc0,
+},
+{
+  .name = "PCIE_P_IMP_CNTL_UPDATE",
+  .address = 0xc1,
+},
+{
+  .name = "PCIE_P_STR_CNTL_UPDATE",
+  .address = 0xc2,
+},
+{
+  .name = "PCIE_P_PAD_MISC_CNTL",
+  .address = 0xc3,
+},
+{
+  .name = "PCIE_P_SYMSYNC_CTL",
+  .address = 0xc4,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNTL",
+  .address = 0xc5,
+},
+{
+  .name = "PCIE_ERR_CNTL",
+  .address = 0xe0,
+},
+{
+  .name = "PCIE_CLK_RST_CNTL",
+  .address = 0xe1,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_0",
+  .address = 0xf0,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_1",
+  .address = 0xf1,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_2",
+  .address = 0xf2,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_3",
+  .address = 0xf3,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_4",
+  .address = 0xf4,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_5",
+  .address = 0xf5,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_6",
+  .address = 0xf6,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_7",
+  .address = 0xf7,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_8",
+  .address = 0xf8,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_9",
+  .address = 0xf9,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_10",
+  .address = 0xfa,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_11",
+  .address = 0xfb,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_12",
+  .address = 0xfc,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_13",
+  .address = 0xfd,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_14",
+  .address = 0xfe,
+},
+{
+  .name = "PCIE_P_DECODE_ERR_CNT_15",
+  .address = 0xff,
+},

regs/pcie_registers.txt

@@ -0,0 +1,486 @@
+PCIE_TX_CNTL
+PCIEIND:0x1
+2-80
+PCIE_TX_GART_CNTL
+PCIEIND:0x10
+2-82
+PCIE_TX_GART_DISCARD_RD_ADDR_LO
+PCIEIND:0x11
+2-82
+PCIE_TX_GART_DISCARD_RD_ADDR_HI
+PCIEIND:0x12
+2-82
+PCIE_TX_GART_BASE
+PCIEIND:0x13
+2-83
+PCIE_TX_GART_START_LO
+PCIEIND:0x14
+2-83
+PCIE_TX_GART_START_HI
+PCIEIND:0x15
+2-83
+PCIE_TX_GART_END_LO
+PCIEIND:0x16
+2-83
+PCIE_TX_GART_END_HI
+PCIEIND:0x17
+2-83
+PCIE_TX_GART_ERROR
+PCIEIND:0x18
+2-83
+PCIE_TX_SEQ
+PCIEIND:0x2
+2-80
+PCIE_TX_GART_LRU_MRU_PTR
+PCIEIND:0x20
+2-84
+PCIE_TX_GART_STATUS
+PCIEIND:0x21
+2-84
+PCIE_TX_GART_TLB_VALID
+PCIEIND:0x22
+2-84
+PCIE_TX_GART_TLB0_DATA
+PCIEIND:0x23
+2-84
+PCIE_TX_GART_TLB1_DATA
+PCIEIND:0x24
+2-84
+PCIE_TX_GART_TLB2_DATA
+PCIEIND:0x25
+2-84
+PCIE_TX_GART_TLB3_DATA
+PCIEIND:0x26
+2-85
+PCIE_TX_GART_TLB4_DATA
+PCIEIND:0x27
+2-85
+PCIE_TX_GART_TLB5_DATA
+PCIEIND:0x28
+2-85
+PCIE_TX_GART_TLB6_DATA
+PCIEIND:0x29
+2-85
+PCIE_TX_GART_TLB7_DATA
+PCIEIND:0x2A
+2-85
+PCIE_TX_GART_TLB8_DATA
+PCIEIND:0x2B
+2-85
+PCIE_TX_GART_TLB9_DATA
+PCIEIND:0x2C
+2-85
+PCIE_TX_GART_TLB10_DATA
+PCIEIND:0x2D
+2-86
+PCIE_TX_GART_TLB11_DATA
+PCIEIND:0x2E
+2-86
+PCIE_TX_GART_TLB12_DATA
+PCIEIND:0x2F
+2-86
+PCIE_TX_REPLAY
+PCIEIND:0x3
+2-80
+PCIE_TX_GART_TLB13_DATA
+PCIEIND:0x30
+2-86
+PCIE_TX_GART_TLB14_DATA
+PCIEIND:0x31
+2-86
+PCIE_TX_GART_TLB15_DATA
+PCIEIND:0x32
+2-86
+PCIE_TX_GART_TLB16_DATA
+PCIEIND:0x33
+2-87
+PCIE_TX_GART_TLB17_DATA
+PCIEIND:0x34
+2-87
+PCIE_TX_GART_TLB18_DATA
+PCIEIND:0x35
+2-87
+PCIE_TX_GART_TLB19_DATA
+PCIEIND:0x36
+2-87
+PCIE_TX_GART_TLB20_DATA
+PCIEIND:0x37
+2-87
+PCIE_TX_GART_TLB21_DATA
+PCIEIND:0x38
+2-87
+PCIE_TX_GART_TLB22_DATA
+PCIEIND:0x39
+2-88
+PCIE_TX_GART_TLB23_DATA
+PCIEIND:0x3A
+2-88
+PCIE_TX_GART_TLB24_DATA
+PCIEIND:0x3B
+2-88
+PCIE_TX_GART_TLB25_DATA
+PCIEIND:0x3C
+2-88
+PCIE_TX_GART_TLB26_DATA
+PCIEIND:0x3D
+2-88
+PCIE_TX_GART_TLB27_DATA
+PCIEIND:0x3E
+2-88
+PCIE_TX_GART_TLB28_DATA
+PCIEIND:0x3F
+2-89
+PCIE_TX_CREDITS_CONSUMED
+PCIEIND:0x4
+2-81
+PCIE_TX_GART_TLB29_DATA
+PCIEIND:0x40
+2-89
+PCIE_CLK_CNTL
+PCIEIND:0x400
+2-107
+PCIE_PRBS10
+PCIEIND:0x401
+2-107
+PCIE_PRBS23_BITCNT0
+PCIEIND:0x402
+2-107
+PCIE_PRBS23_BITCNT1
+PCIEIND:0x403
+2-107
+PCIE_PRBS23_BITCNT2
+PCIEIND:0x404
+2-107
+PCIE_PRBS23_BITCNT3
+PCIEIND:0x405
+2-108
+PCIE_PRBS23_BITCNT4
+PCIEIND:0x406
+2-108
+PCIE_PRBS23_BITCNT5
+PCIEIND:0x407
+2-108
+PCIE_PRBS23_BITCNT6
+PCIEIND:0x408
+2-108
+PCIE_PRBS23_BITCNT7
+PCIEIND:0x409
+2-108
+PCIE_PRBS23_BITCNT8
+PCIEIND:0x40A
+2-108
+PCIE_PRBS23_BITCNT9
+PCIEIND:0x40B
+2-108
+PCIE_PRBS23_BITCNT10
+PCIEIND:0x40C
+2-109
+PCIE_PRBS23_BITCNT11
+PCIEIND:0x40D
+2-109
+PCIE_PRBS23_BITCNT12
+PCIEIND:0x40E
+2-109
+PCIE_PRBS23_BITCNT13
+PCIEIND:0x40F
+2-109
+PCIE_TX_GART_TLB30_DATA
+PCIEIND:0x41
+2-89
+PCIE_PRBS23_BITCNT14
+PCIEIND:0x410
+2-109
+PCIE_PRBS23_BITCNT15
+PCIEIND:0x411
+2-109
+PCIE_PRBS23_ERRCNT0
+PCIEIND:0x412
+2-110
+PCIE_PRBS23_ERRCNT1
+PCIEIND:0x413
+2-110
+PCIE_PRBS23_ERRCNT2
+PCIEIND:0x414
+2-110
+PCIE_PRBS23_ERRCNT3
+PCIEIND:0x415
+2-110
+PCIE_PRBS23_ERRCNT4
+PCIEIND:0x416
+2-110
+PCIE_PRBS23_ERRCNT5
+PCIEIND:0x417
+2-110
+PCIE_PRBS23_ERRCNT6
+PCIEIND:0x418
+2-110
+PCIE_PRBS23_ERRCNT7
+PCIEIND:0x419
+2-111
+PCIE_PRBS23_ERRCNT8
+PCIEIND:0x41A
+2-111
+PCIE_PRBS23_ERRCNT9
+PCIEIND:0x41B
+2-111
+PCIE_PRBS23_ERRCNT10
+PCIEIND:0x41C
+2-111
+PCIE_PRBS23_ERRCNT11
+PCIEIND:0x41D
+2-111
+PCIE_PRBS23_ERRCNT12
+PCIEIND:0x41E
+2-111
+PCIE_PRBS23_ERRCNT13
+PCIEIND:0x41F
+2-111
+PCIE_TX_GART_TLB31_DATA
+PCIEIND:0x42
+2-89
+PCIE_PRBS23_ERRCNT14
+PCIEIND:0x420
+2-112
+PCIE_PRBS23_ERRCNT15
+PCIEIND:0x421
+2-112
+PCIE_PRBS23_CTRL0
+PCIEIND:0x422
+2-112
+PCIE_PRBS23_CTRL1
+PCIEIND:0x423
+2-113
+PCIE_PRBS_EN
+PCIEIND:0x424
+2-113
+PCIE_XSTRAP1
+PCIEIND:0x425
+2-113
+PCIE_XSTRAP2
+PCIEIND:0x426
+2-114
+PCIE_XSTRAP5
+PCIEIND:0x429
+2-114
+PCIE_TX_CREDITS_CONSUMED_D
+PCIEIND:0x5
+2-81
+PCIE_TX_CREDITS_CONSUMED_CPLD
+PCIEIND:0x6
+2-81
+PCIE_FLOW_CNTL
+PCIEIND:0x60
+2-89
+PCIE_TXRX_DEBUG_SEQNUM
+PCIEIND:0x61
+2-90
+PCIE_TXRX_TEST_MODE
+PCIEIND:0x62
+2-90
+PCIE_TX_CREDITS_LIMIT
+PCIEIND:0x7
+2-81
+PCIE_RX_CNTL
+PCIEIND:0x70
+2-90
+PCIE_RX_NUM_NACK
+PCIEIND:0x71
+2-91
+PCIE_RX_NUM_NACK_GENERATED
+PCIEIND:0x72
+2-91
+PCIE_RX_ACK_NACK_LATENCY
+PCIEIND:0x73
+2-91
+PCIE_RX_ACK_NACK_LATENCY_THRESHOLD
+PCIEIND:0x74
+2-91
+PCIE_RX_TLP_HDR0
+PCIEIND:0x75
+2-91
+PCIE_RX_TLP_HDR1
+PCIEIND:0x76
+2-91
+PCIE_RX_TLP_HDR2
+PCIEIND:0x77
+2-91
+PCIE_RX_TLP_HDR3
+PCIEIND:0x78
+2-92
+PCIE_RX_TLP_HDR4
+PCIEIND:0x79
+2-92
+PCIE_RX_TLP_CRC
+PCIEIND:0x7A
+2-92
+PCIE_RX_DLP0
+PCIEIND:0x7B
+2-92
+PCIE_RX_DLP1
+PCIEIND:0x7C
+2-92
+PCIE_RX_DLP_CRC
+PCIEIND:0x7D
+2-92
+PCIE_RX_CREDITS_ALLOCATED
+PCIEIND:0x7E
+2-92
+PCIE_RX_CREDITS_ALLOCATED_D
+PCIEIND:0x7F
+2-93
+PCIE_TX_CREDITS_LIMIT_D
+PCIEIND:0x8
+2-81
+PCIE_RX_CREDITS_ALLOCATED_CPLD
+PCIEIND:0x80
+2-93
+PCIE_RX_CREDITS_RECEIVED
+PCIEIND:0x81
+2-93
+PCIE_RX_CREDITS_RECEIVED_D
+PCIEIND:0x82
+2-93
+PCIE_RX_CREDITS_RECEIVED_CPLD
+PCIEIND:0x83
+2-93
+PCIE_RX_MAL_TLP_COUNT
+PCIEIND:0x84
+2-93
+PCIE_RX_ERR_LOG
+PCIEIND:0x85
+2-94
+PCIE_RX_EXPECTED_SEQNUM
+PCIEIND:0x86
+2-94
+PCIE_TX_CREDITS_LIMIT_CPLD
+PCIEIND:0x9
+2-82
+PCIE_CI_CNTL
+PCIEIND:0x90
+2-94
+PCIE_CI_FLUSH_CNTL
+PCIEIND:0x91
+2-94
+PCIE_CI_PANIC
+PCIEIND:0x92
+2-94
+PCIE_CI_HANG
+PCIEIND:0x93
+2-95
+PCIE_LC_CNTL
+PCIEIND:0xA0
+2-95
+PCIE_LC_N_FTS_CNTL
+PCIEIND:0xA1
+2-95
+PCIE_LC_LINK_WIDTH_CNTL
+PCIEIND:0xA2
+2-97
+PCIE_LC_STATE0
+PCIEIND:0xA5
+2-95
+PCIE_LC_STATE1
+PCIEIND:0xA6
+2-96
+PCIE_LC_STATE2
+PCIEIND:0xA7
+2-96
+PCIE_LC_STATE3
+PCIEIND:0xA8
+2-96
+PCIE_LC_STATE4
+PCIEIND:0xA9
+2-96
+PCIE_LC_STATE5
+PCIEIND:0xAA
+2-96
+PCIE_LC_FORCE_SYNC_LOSS_CNTL
+PCIEIND:0xAB
+2-97
+PCIE_P_CNTL
+PCIEIND:0xB0
+2-97
+PCIE_P_CNTL2
+PCIEIND:0xB1
+2-98
+PCIE_P_BUF_STATUS
+PCIEIND:0xB2
+2-98
+PCIE_P_DECODER_STATUS
+PCIEIND:0xB3
+2-99
+PCIE_P_MISC_DEBUG_STATUS
+PCIEIND:0xB4
+2-100
+PCIE_P_IMP_CNTL_STRENGTH
+PCIEIND:0xC0
+2-101
+PCIE_P_IMP_CNTL_UPDATE
+PCIEIND:0xC1
+2-102
+PCIE_P_STR_CNTL_UPDATE
+PCIEIND:0xC2
+2-102
+PCIE_P_PAD_MISC_CNTL
+PCIEIND:0xC3
+2-102
+PCIE_P_SYMSYNC_CTL
+PCIEIND:0xC4
+2-102
+PCIE_P_DECODE_ERR_CNTL
+PCIEIND:0xC5
+2-103
+PCIE_ERR_CNTL
+PCIEIND:0xE0
+2-105
+PCIE_CLK_RST_CNTL
+PCIEIND:0xE1
+2-105
+PCIE_P_DECODE_ERR_CNT_0
+PCIEIND:0xF0
+2-103
+PCIE_P_DECODE_ERR_CNT_1
+PCIEIND:0xF1
+2-103
+PCIE_P_DECODE_ERR_CNT_2
+PCIEIND:0xF2
+2-103
+PCIE_P_DECODE_ERR_CNT_3
+PCIEIND:0xF3
+2-103
+PCIE_P_DECODE_ERR_CNT_4
+PCIEIND:0xF4
+2-103
+PCIE_P_DECODE_ERR_CNT_5
+PCIEIND:0xF5
+2-103
+PCIE_P_DECODE_ERR_CNT_6
+PCIEIND:0xF6
+2-104
+PCIE_P_DECODE_ERR_CNT_7
+PCIEIND:0xF7
+2-104
+PCIE_P_DECODE_ERR_CNT_8
+PCIEIND:0xF8
+2-104
+PCIE_P_DECODE_ERR_CNT_9
+PCIEIND:0xF9
+2-104
+PCIE_P_DECODE_ERR_CNT_10
+PCIEIND:0xFA
+2-104
+PCIE_P_DECODE_ERR_CNT_11
+PCIEIND:0xFB
+2-104
+PCIE_P_DECODE_ERR_CNT_12
+PCIEIND:0xFC
+2-105
+PCIE_P_DECODE_ERR_CNT_13
+PCIEIND:0xFD
+2-105
+PCIE_P_DECODE_ERR_CNT_14
+PCIEIND:0xFE
+2-105
+PCIE_P_DECODE_ERR_CNT_15
+PCIEIND:0xFF
+2-105

regs/rrg_registers.py

@@ -0,0 +1,37 @@
+import sys
+
+with open(sys.argv[1]) as f:
+    buf = f.read()
+
+prefixes = sys.argv[2:]
+assert len(prefixes) >= 1
+
+lines = [line.strip() for line in buf.strip().split()]
+
+assert len(lines) % 3 == 0
+
+def parse(lines):
+    for i in range(len(lines) // 3):
+        name = lines[i * 3 + 0]
+        address = lines[i * 3 + 1]
+        page = lines[i * 3 + 2]
+
+        assert '-' in page, page
+        orig_address = address
+        for prefix in prefixes:
+            if address.startswith(f"{prefix}:"):
+                address = address.removeprefix(f"{prefix}:")
+        assert address != orig_address
+        assert address.startswith("0x")
+        address = address.removeprefix("0x")
+        address = int(address, 16)
+        yield name, address, page
+
+
+for name, address, page in parse(lines):
+    print("{")
+    print(f"  .name = \"{name}\",")
+    print(f"  .address = {hex(address)},")
+    print("},")
+
+#print(f"#define {name} {hex(address)}")

shader_examples/mesa/gl_teximage2d.fs.txt

@@ -0,0 +1,12 @@
+0x00007807,
+0x02400000,
+0xe400f400,
+0x00000000,
+0x00000000,
+0x00000000,
+0x00078005,
+0x08020000,
+0x08020000,
+0x1c440220,
+0x1c60c003,
+0x00000005,

shader_examples/mesa/gl_teximage2d.vs.txt

@@ -0,0 +1,8 @@
+0x00f00203,
+0x00d10001,
+0x01248001,
+0x01248001,
+0x00f02203,
+0x00d10021,
+0x01248021,
+0x01248021,

src/Makefile

@@ -21,7 +21,7 @@ R500_COMMON = \
 	drm/drm.o \
 	file.o
 
-matrix_cubesphere: $(R500_COMMON) matrix_cubesphere.o | shaders
+%: $(R500_COMMON) %.o | shaders
 	$(CXX) $(LDFLAGS) $^ -o $@
 
 %.o: %.c
@@ -45,9 +45,9 @@ matrix_cubesphere: $(R500_COMMON) matrix_cubesphere.o | shaders
 shaders: $(SHADER_BIN)
 	@true
 
-#find . -type f ! -name "*.*" -delete
 clean:
-	find . -type f -name "*.o" -delete
+	find . -type f -name "*.o" -delete -print
+	find . -type f -executable ! -name '*.*' -delete -print
 
 .SUFFIXES:
 .INTERMEDIATE:

src/drm/drm.c

@@ -7,10 +7,10 @@
 #include "drm.h"
 #include "../r500/indirect_buffer.h" // for extern uint32_t ib[];
 
-void drm_radeon_cs(int fd,
+int drm_radeon_cs(int fd,
                   int colorbuffer_handle,
                   int zbuffer_handle,
-                   int flush_handle,
+                  int vertexbuffer_handle,
                   int * texturebuffer_handles,
                   int texturebuffer_handles_length,
                   int ib_dwords)
@@ -29,15 +29,23 @@ void drm_radeon_cs(int fd,
     .write_domain = 4, // RADEON_GEM_DOMAIN_VRAM
     .flags = 8,
   };
+  /*
   relocs[FLUSH_RELOC_INDEX] = (struct drm_radeon_cs_reloc){
     .handle = flush_handle,
     .read_domains = 2, // RADEON_GEM_DOMAIN_GTT
     .write_domain = 2, // RADEON_GEM_DOMAIN_GTT
     .flags = 0,
   };
+  */
+  relocs[VERTEXBUFFER_RELOC_INDEX] = (struct drm_radeon_cs_reloc){
+    .handle = vertexbuffer_handle,
+    .read_domains = 4, // RADEON_GEM_DOMAIN_VRAM
+    .write_domain = 4, // RADEON_GEM_DOMAIN_VRAM
+    .flags = 8,
+  };
 
   for (int i = 0; i < texturebuffer_handles_length; i++) {
-    relocs[3 + i] = (struct drm_radeon_cs_reloc){
+    relocs[TEXTUREBUFFER_RELOC_INDEX + i] = (struct drm_radeon_cs_reloc){
       .handle = texturebuffer_handles[i],
       .read_domains = 4, // RADEON_GEM_DOMAIN_VRAM
       .write_domain = 4, // RADEON_GEM_DOMAIN_VRAM
@@ -85,5 +93,70 @@ void drm_radeon_cs(int fd,
   int ret = drmCommandWriteRead(fd, DRM_RADEON_CS, &cs, (sizeof (struct drm_radeon_cs)));
   if (ret != 0) {
     perror("drmCommandWriteRead(DRM_RADEON_CS)");
+    return -1;
   }
+
+  return 0;
+}
+
+int drm_radeon_cs2(int fd,
+                   int * handles,
+                   int handles_length,
+                   int ib_dwords)
+{
+  struct drm_radeon_cs_reloc relocs[handles_length];
+
+  for (int i = 0; i < handles_length; i++) {
+    relocs[i] = (struct drm_radeon_cs_reloc){
+      .handle = handles[i],
+      .read_domains = 4, // RADEON_GEM_DOMAIN_VRAM
+      .write_domain = 4, // RADEON_GEM_DOMAIN_VRAM
+      .flags = 8,
+    };
+  }
+
+  const uint32_t flags[2] = {
+    5, // RADEON_CS_KEEP_TILING_FLAGS | RADEON_CS_END_OF_FRAME
+    0, // RADEON_CS_RING_GFX
+  };
+
+  struct drm_radeon_cs_chunk chunks[3] = {
+    {
+      .chunk_id = RADEON_CHUNK_ID_IB,
+      .length_dw = ib_dwords,
+      .chunk_data = (uint64_t)(uintptr_t)ib,
+    },
+    {
+      .chunk_id = RADEON_CHUNK_ID_RELOCS,
+      .length_dw = (sizeof (relocs)) / (sizeof (uint32_t)),
+      .chunk_data = (uint64_t)(uintptr_t)relocs,
+    },
+    {
+      .chunk_id = RADEON_CHUNK_ID_FLAGS,
+      .length_dw = (sizeof (flags)) / (sizeof (uint32_t)),
+      .chunk_data = (uint64_t)(uintptr_t)&flags,
+    },
+  };
+
+  uint64_t chunks_array[3] = {
+    (uint64_t)(uintptr_t)&chunks[0],
+    (uint64_t)(uintptr_t)&chunks[1],
+    (uint64_t)(uintptr_t)&chunks[2],
+  };
+
+  struct drm_radeon_cs cs = {
+    .num_chunks = 3,
+    .cs_id = 0,
+    .chunks = (uint64_t)(uintptr_t)chunks_array,
+    .gart_limit = 0,
+    .vram_limit = 0,
+  };
+
+  int ret = drmCommandWriteRead(fd, DRM_RADEON_CS, &cs, (sizeof (struct drm_radeon_cs)));
+  if (ret != 0) {
+    perror("drmCommandWriteRead(DRM_RADEON_CS)");
+    return -1;
+  }
+
+  return 0;
 }

src/drm/drm.h

@@ -2,21 +2,27 @@
 
 #define COLORBUFFER_RELOC_INDEX 0
 #define ZBUFFER_RELOC_INDEX 1
-#define FLUSH_RELOC_INDEX 2
-#define TEXTURE_RELOC_INDEX 3
+//#define FLUSH_RELOC_INDEX 2
+#define VERTEXBUFFER_RELOC_INDEX 2
+#define TEXTUREBUFFER_RELOC_INDEX 3
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-void drm_radeon_cs(int fd,
+int drm_radeon_cs(int fd,
                   int colorbuffer_handle,
                   int zbuffer_handle,
-                   int flush_handle,
+                  int vertexbuffer_handle,
                   int * texturebuffer_handles,
                   int texturebuffer_handles_length,
                   int ib_dwords);
 
+int drm_radeon_cs2(int fd,
+                   int * handles,
+                   int handles_length,
+                   int ib_dwords);
+
 #ifdef __cplusplus
 }
 #endif

src/matrix_cubesphere.cpp

@@ -67,7 +67,7 @@ void _3d_clear(struct shaders& shaders)
 
   //
 
-  ib_zbuffer(ZBUFFER_RELOC_INDEX, 7); // always
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 7); // always
 
   ib_texture__0();
 
@@ -315,7 +315,7 @@ vec3 _3d_light(struct shaders& shaders,
 
   //
 
-  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1); // less
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
 
   ib_texture__0();
 
@@ -403,9 +403,17 @@ void _3d_cube(struct shaders& shaders,
 
   //
 
-  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1); // less
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
 
-  ib_texture__1(TEXTURE_RELOC_INDEX);
+  int width = 1024;
+  int height = 1024;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX,
+                width, height,
+                macrotile, microtile,
+                clamp);
 
   ib_vap_stream_cntl__323();
 
@@ -473,16 +481,28 @@ void _3d_cube(struct shaders& shaders,
 int indirect_buffer(shaders& shaders,
                     float theta)
 {
+  int width = 1600;
+  int height = 1200;
+  int pitch = width;
+
   ib_ix = 0;
 
   ib_generic_initialization();
 
+  T0V(RB3D_BLENDCNTL, 0);
+  T0V(RB3D_ABLENDCNTL, 0);
+
+  ib_viewport(width, height);
+  ib_colorbuffer(COLORBUFFER_RELOC_INDEX, pitch, 0, 0);
+
+  T0V(GB_ENABLE, 0);
+
   T0V(US_OUT_FMT_0
       , US_OUT_FMT__OUT_FMT(0)  // C4_8
-      | US_OUT_FMT__C0_SEL(3)   // Blue
-      | US_OUT_FMT__C1_SEL(2)   // Green
-      | US_OUT_FMT__C2_SEL(1)   // Red
-      | US_OUT_FMT__C3_SEL(0)   // Alpha
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
       | US_OUT_FMT__OUT_SIGN(0)
       );
   T0V(US_OUT_FMT_1
@@ -495,8 +515,6 @@ int indirect_buffer(shaders& shaders,
       , US_OUT_FMT__OUT_FMT(15) // render target is not used
       );
 
-  ib_colorbuffer(COLORBUFFER_RELOC_INDEX);
-
   load_pvs_shaders(shaders.vertex, shaders.vertex_length);
   load_us_shaders(shaders.fragment, shaders.fragment_length);
 
@@ -527,7 +545,7 @@ int indirect_buffer(shaders& shaders,
 }
 
 const char * textures[] = {
-  "../texture/butterfly_1024x1024_argb8888.data",
+  "../texture/butterfly_1024x1024_rgba8888.data",
 };
 const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
 
@@ -571,13 +589,15 @@ int main()
   while (true) {
     int ib_dwords = indirect_buffer(shaders, theta);
 
-    drm_radeon_cs(fd,
+    int ret = drm_radeon_cs(fd,
                             colorbuffer_handle[colorbuffer_ix],
                             zbuffer_handle,
                             flush_handle,
                             texturebuffer_handle,
                             textures_length,
                             ib_dwords);
+    if (ret == -1)
+      break;
 
     primary_surface_address(rmmio, colorbuffer_ix);
 

src/matrix_cubesphere_indexed.cpp

@@ -0,0 +1,672 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+#include "../model/cubesphere.h"
+#include "../model/cubesphere_indexed.h"
+
+#define CLEAR_SHADER 0
+#define CUBESPHERE_SHADER 1
+#define LIGHT_SHADER 2
+
+#define INDEXBUFFER_RELOC_INDEX 4
+
+const char * vertex_shader_paths[] = {
+  "clear.vs.bin",
+  "matrix_cubesphere.vs.bin",
+  "light.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "clear.fs.bin",
+  "matrix_cubesphere.fs.bin",
+  "light.fs.bin"
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+void _3d_clear(struct shaders& shaders)
+{
+  ib_rs_instructions(0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , 0);
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 7); // always
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[CLEAR_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CLEAR_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT(600 * 12)
+      | GA_POINT_SIZE__WIDTH(800 * 12)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float center[] = {
+    800.0f, 600.0f,
+  };
+  const int vertex_count = 1;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(1) // point list
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < 2; i++) {
+    TF(center[i]);
+  }
+}
+
+mat4x4 perspective(float low1, float high1,
+                   float low2, float high2,
+                   float low3, float high3)
+{
+  float scale2 = (high2 - low2) / (high1 - low1);
+  float scale3 = (high3 - low3) / (high1 - low1);
+
+  mat4x4 m1 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, 1, -low1,
+                     0, 0, 0, 1
+                     );
+
+  mat4x4 m2 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, scale2, low2,
+                     0, 0, scale3, low3
+                     );
+
+  return m2 * m1;
+}
+
+void _3d_cube_inner(mat4x4 trans,
+                    mat4x4 world_trans,
+                    vec4 light_pos)
+{
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP_PVS
+  //////////////////////////////////////////////////////////////////////////////
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+
+    // 4
+    world_trans[0][0], world_trans[0][1], world_trans[0][2], world_trans[0][3],
+    world_trans[1][0], world_trans[1][1], world_trans[1][2], world_trans[1][3],
+    world_trans[2][0], world_trans[2][1], world_trans[2][2], world_trans[2][3],
+    world_trans[3][0], world_trans[3][1], world_trans[3][2], world_trans[3][3],
+
+    // 8
+    light_pos.x, light_pos.y, light_pos.z, light_pos.w,
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  //////////////////////////////////////////////////////////////////////////////
+  // AOS
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  T3(_3D_LOAD_VBPNTR, (6 - 1));
+  TU( // VAP_VTX_NUM_ARRAYS
+      VAP_VTX_NUM_ARRAYS__VTX_NUM_ARRAYS(3)
+    | VAP_VTX_NUM_ARRAYS__VC_FORCE_PREFETCH(1)
+    );
+  TU( // VAP_VTX_AOS_ATTR01
+      VAP_VTX_AOS_ATTR__VTX_AOS_COUNT0(3)
+    | VAP_VTX_AOS_ATTR__VTX_AOS_STRIDE0(8)
+    | VAP_VTX_AOS_ATTR__VTX_AOS_COUNT1(2)
+    | VAP_VTX_AOS_ATTR__VTX_AOS_STRIDE1(8)
+    );
+  TU( // VAP_VTX_AOS_ADDR0
+    (4 * 0)
+    );
+  TU( // VAP_VTX_AOS_ADDR1
+    (4 * 3)
+    );
+  TU( // VAP_VTX_AOS_ATTR23
+      VAP_VTX_AOS_ATTR__VTX_AOS_COUNT0(3)
+    | VAP_VTX_AOS_ATTR__VTX_AOS_STRIDE0(8)
+    );
+  TU( // VAP_VTX_AOS_ADDR2
+    (4 * 5)
+    );
+
+  // VAP_VTX_AOS_ADDR is an absolute address in VRAM. However, DRM_RADEON_CS
+  // modifies this to be an offset relative to the GEM buffer handles given via
+  // NOP:
+  T3(_NOP, 0);
+  TU(VERTEXBUFFER_RELOC_INDEX * 4); // index into relocs array for VAP_VTX_AOS_ADDR0
+  T3(_NOP, 0);
+  TU(VERTEXBUFFER_RELOC_INDEX * 4); // index into relocs array for VAP_VTX_AOS_ADDR1
+  T3(_NOP, 0);
+  TU(VERTEXBUFFER_RELOC_INDEX * 4); // index into relocs array for VAP_VTX_AOS_ADDR2
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int vertex_count = cubesphere_Cube_triangles_length;
+
+  T3(_3D_DRAW_INDX_2, (1 - 1));
+  TU( VAP_VF_CNTL__PRIM_TYPE(4)
+    | VAP_VF_CNTL__PRIM_WALK(1) // indexes in command stream , vertices from memory
+    | VAP_VF_CNTL__INDEX_SIZE(1) // 32 bits per index
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+
+  T3(_INDX_BUFFER, (3 - 1));
+  TU( // INDX_BUFFER__0
+      INDX_BUFFER__0__ONE_REG_WR(1)
+    | INDX_BUFFER__0__SKIP_COUNT(0)
+    | INDX_BUFFER__0__DESTINATION(VAP_PORT_IDX0 >> 2)
+    );
+  TU( // INDX_BUFFER__1
+      INDX_BUFFER__1__BUFFER_BASE(0) // replaced by reloc
+    );
+  TU( // INDX_BUFFER__2
+      INDX_BUFFER__2__BUFFER_SIZE(vertex_count) // in dwords
+    );
+  T3(_NOP, 0);
+  TU(INDEXBUFFER_RELOC_INDEX * 4); // index into relocs array for INDX_BUFFER__1__BUFFER_BASE
+}
+
+void _3d_light_inner(mat4x4 trans)
+{
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP_PVS
+  //////////////////////////////////////////////////////////////////////////////
+
+  const vec4 color = {1, 1, 0, 1};
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+
+    // 4
+    color[0], color[1], color[2], color[2],
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const model * model = &cubesphere_model;
+  const object * obj = model->object[0];
+  const int triangle_count = obj->triangle_count;
+  const int vertex_count = triangle_count * 3;
+
+  int dwords_per_vtx = 3;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(4)
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+
+  for (int i = 0; i < triangle_count; i++) {
+    for (int j = 0; j < 3; j++) {
+      vec3 p = model->position[obj->triangle[i][j].position];
+
+      TF(p.x);
+      TF(p.y);
+      TF(p.z);
+    }
+  }
+}
+
+vec3 _3d_light(struct shaders& shaders,
+               const mat4x4& view_to_clip,
+               float theta)
+{
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__3();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[LIGHT_SHADER]);
+  ib_vap_pvs(&shaders.vertex[LIGHT_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(0)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(0)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  // light
+  mat4x4 t1 = translate(vec3(1, 0, 0));
+  mat4x4 s = scale(0.1f);
+  mat4x4 rz = rotate_y(theta * 2.f);
+
+  mat4x4 world_trans = rz * t1 * s;
+
+  mat4x4 trans = view_to_clip * world_trans;
+
+  _3d_light_inner(trans);
+
+  vec3 light_pos = world_trans * light_pos;
+
+  return light_pos;
+}
+
+void _3d_cube(struct shaders& shaders,
+              const mat4x4& view_to_clip,
+              float theta,
+              const vec3& light_pos)
+{
+  ib_rs_instructions(4);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4)
+      | VAP_OUT_VTX_FMT_1__TEX_1_COMP_CNT(4)
+      | VAP_OUT_VTX_FMT_1__TEX_2_COMP_CNT(4)
+      | VAP_OUT_VTX_FMT_1__TEX_3_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 1024;
+  int height = 1024;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__323();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[CUBESPHERE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CUBESPHERE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(0)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(0)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(124)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  // cube
+  mat4x4 rx = rotate_x(1 * theta * 0.5f);
+  mat4x4 ry = rotate_y(0 * theta * 0.8f + 1.4f);
+  mat4x4 s = scale(0.9f);
+
+  mat4x4 world_trans = rx * ry * s;
+
+  mat4x4 trans = view_to_clip * world_trans;
+
+  _3d_cube_inner(trans, world_trans, light_pos);
+}
+
+int indirect_buffer(shaders& shaders,
+                    float theta)
+{
+  int width = 1600;
+  int height = 1200;
+  int pitch = width;
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  T0V(RB3D_BLENDCNTL, 0);
+  T0V(RB3D_ABLENDCNTL, 0);
+
+  ib_viewport(width, height);
+  ib_colorbuffer(COLORBUFFER_RELOC_INDEX, pitch, 0, 0);
+
+  T0V(GB_ENABLE, 0);
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 aspect = scale(vec3(3.0f/4.0f, 1, 1));
+  mat4x4 p = perspective(0.01f, 5.0f,
+                         0.001f, 0.999f,
+                         0.5f, 2.0f);
+  mat4x4 t = translate(vec3(0, 0, 3));
+  mat4x4 view_to_clip = aspect * p * t;
+
+  _3d_clear(shaders);
+  vec3 light_pos = _3d_light(shaders, view_to_clip, theta);
+  _3d_cube(shaders, view_to_clip, theta, light_pos);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // padding
+  //////////////////////////////////////////////////////////////////////////////
+
+  while ((ib_ix % 8) != 0) {
+    TU(0x80000000);
+  }
+
+  return ib_ix;
+}
+
+const char * textures[] = {
+  "../texture/butterfly_1024x1024_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+void transfer_vertex_index(void * vertexbuffer_ptr,
+                           void * indexbuffer_ptr)
+{
+  for (int i = 0; i < cubesphere_vertices_length; i++) {
+    ((float *)vertexbuffer_ptr)[i] = cubesphere_vertices[i];
+  }
+
+  for (int i = 0; i < cubesphere_vertices_length; i++) {
+    ((int *)indexbuffer_ptr)[i] = cubesphere_Cube_triangles[i];
+  }
+  asm volatile ("" ::: "memory");
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int vertexbuffer_handle;
+  int * texturebuffer_handle;
+  int indexbuffer_handle;
+
+  void * vertexbuffer_ptr;
+  void * indexbuffer_ptr;
+  void * colorbuffer_ptr[2];
+  void * zbuffer_ptr;
+
+  // cubesphere_Cube_triangles_length, cubesphere_vertices_length
+  const int vertexbuffer_size = cubesphere_vertices_length * 4;
+  const int indexbuffer_size = cubesphere_Cube_triangles_length * 4;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  vertexbuffer_handle = create_buffer(fd, vertexbuffer_size, &vertexbuffer_ptr);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+  indexbuffer_handle = create_buffer(fd, indexbuffer_size, &indexbuffer_ptr);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  transfer_vertex_index(vertexbuffer_ptr, indexbuffer_ptr);
+
+  int colorbuffer_ix = 0;
+  float theta = 0;
+
+  while (true) {
+    int ib_dwords = indirect_buffer(shaders, theta);
+
+    int handles[] = {
+      colorbuffer_handle[colorbuffer_ix], // 0
+      zbuffer_handle,                     // 1
+      vertexbuffer_handle,                // 2
+      texturebuffer_handle[0],            // 3
+      indexbuffer_handle,                 // 4
+    };
+    int handles_length = (sizeof (handles)) / (sizeof (handles[0]));
+
+    int ret = drm_radeon_cs2(fd,
+                             handles,
+                             handles_length,
+                             ib_dwords);
+    if (ret == -1)
+      break;
+
+    primary_surface_address(rmmio, colorbuffer_ix);
+
+    // next state
+    theta += 0.01f;
+    colorbuffer_ix = (colorbuffer_ix + 1) & 1;
+  }
+
+  close(fd);
+}

src/matrix_cubesphere_tiled.cpp

@@ -0,0 +1,821 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+#include "../model/cubesphere.h"
+
+#define CLEAR_SHADER 0
+#define CUBESPHERE_SHADER 1
+#define LIGHT_SHADER 2
+#define TEXTURE_TILE_SHADER 3
+
+const char * vertex_shader_paths[] = {
+  "clear.vs.bin",
+  "matrix_cubesphere.vs.bin",
+  "light.vs.bin",
+  "texture_tile.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "clear.fs.bin",
+  "matrix_cubesphere.fs.bin",
+  "light.fs.bin",
+  "texture_tile.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+void _3d_clear(const shaders& shaders)
+{
+  ib_rs_instructions(0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , 0);
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 7); // always
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[CLEAR_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CLEAR_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT(600 * 12)
+      | GA_POINT_SIZE__WIDTH(800 * 12)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float center[] = {
+    800.0f, 600.0f,
+  };
+  const int vertex_count = 1;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(1) // point list
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < 2; i++) {
+    TF(center[i]);
+  }
+}
+
+mat4x4 perspective(float low1, float high1,
+                   float low2, float high2,
+                   float low3, float high3)
+{
+  float scale2 = (high2 - low2) / (high1 - low1);
+  float scale3 = (high3 - low3) / (high1 - low1);
+
+  mat4x4 m1 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, 1, -low1,
+                     0, 0, 0, 1
+                     );
+
+  mat4x4 m2 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, scale2, low2,
+                     0, 0, scale3, low3
+                     );
+
+  return m2 * m1;
+}
+
+void _3d_cube_inner(mat4x4 trans,
+                    mat4x4 world_trans,
+                    vec4 light_pos)
+{
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP_PVS
+  //////////////////////////////////////////////////////////////////////////////
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+
+    // 4
+    world_trans[0][0], world_trans[0][1], world_trans[0][2], world_trans[0][3],
+    world_trans[1][0], world_trans[1][1], world_trans[1][2], world_trans[1][3],
+    world_trans[2][0], world_trans[2][1], world_trans[2][2], world_trans[2][3],
+    world_trans[3][0], world_trans[3][1], world_trans[3][2], world_trans[3][3],
+
+    // 8
+    light_pos.x, light_pos.y, light_pos.z, light_pos.w,
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const model * model = &cubesphere_model;
+  const object * obj = model->object[0];
+  const int triangle_count = obj->triangle_count;
+  const int vertex_count = triangle_count * 3;
+
+  const int dwords_per_vtx = 8;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(4)
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+
+  for (int i = 0; i < triangle_count; i++) {
+    for (int j = 0; j < 3; j++) {
+      vec3 p = model->position[obj->triangle[i][j].position];
+      vec2 t = model->texture[obj->triangle[i][j].texture];
+      vec3 n = model->normal[obj->triangle[i][j].normal];
+
+      TF(p.x);
+      TF(p.y);
+      TF(p.z);
+      TF(t.x);
+      TF(t.y);
+      TF(n.x);
+      TF(n.y);
+      TF(n.z);
+    }
+  }
+}
+
+void _3d_light_inner(mat4x4 trans)
+{
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP_PVS
+  //////////////////////////////////////////////////////////////////////////////
+
+  const vec4 color = {1, 1, 0, 1};
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+
+    // 4
+    color[0], color[1], color[2], color[2],
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const model * model = &cubesphere_model;
+  const object * obj = model->object[0];
+  const int triangle_count = obj->triangle_count;
+  const int vertex_count = triangle_count * 3;
+
+  int dwords_per_vtx = 3;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(4)
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+
+  for (int i = 0; i < triangle_count; i++) {
+    for (int j = 0; j < 3; j++) {
+      vec3 p = model->position[obj->triangle[i][j].position];
+
+      TF(p.x);
+      TF(p.y);
+      TF(p.z);
+    }
+  }
+}
+
+vec3 _3d_light(const shaders& shaders,
+               const mat4x4& view_to_clip,
+               float theta)
+{
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__3();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[LIGHT_SHADER]);
+  ib_vap_pvs(&shaders.vertex[LIGHT_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(0)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(0)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  // light
+  mat4x4 t1 = translate(vec3(1, 0, 0));
+  mat4x4 s = scale(0.1f);
+  mat4x4 rz = rotate_y(theta * 2.f);
+
+  mat4x4 world_trans = rz * t1 * s;
+
+  mat4x4 trans = view_to_clip * world_trans;
+
+  _3d_light_inner(trans);
+
+  vec3 light_pos = world_trans * light_pos;
+
+  return light_pos;
+}
+
+void _3d_cube(const shaders& shaders,
+              const mat4x4& view_to_clip,
+              float theta,
+              const vec3& light_pos)
+{
+  ib_rs_instructions(4);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4)
+      | VAP_OUT_VTX_FMT_1__TEX_1_COMP_CNT(4)
+      | VAP_OUT_VTX_FMT_1__TEX_2_COMP_CNT(4)
+      | VAP_OUT_VTX_FMT_1__TEX_3_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 1024;
+  int height = 1024;
+  int macrotile = 1;
+  int microtile = 1;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__323();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[CUBESPHERE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CUBESPHERE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(0)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(0)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  // cube
+  mat4x4 rx = rotate_x(1 * theta * 0.5f);
+  mat4x4 ry = rotate_y(0 * theta * 0.8f + 1.4f);
+  mat4x4 s = scale(0.9f);
+
+  mat4x4 world_trans = rx * ry * s;
+
+  mat4x4 trans = view_to_clip * world_trans;
+
+  _3d_cube_inner(trans, world_trans, light_pos);
+}
+
+int indirect_buffer(const shaders& shaders,
+                    float theta)
+{
+  int width = 1600;
+  int height = 1200;
+  int pitch = width;
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  T0V(RB3D_BLENDCNTL, 0);
+  T0V(RB3D_ABLENDCNTL, 0);
+
+  ib_viewport(width, height);
+  ib_colorbuffer(COLORBUFFER_RELOC_INDEX, pitch, 0, 0);
+
+  T0V(GB_ENABLE, 0);
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_3
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 aspect = scale(vec3(3.0f/4.0f, 1, 1));
+  mat4x4 p = perspective(0.01f, 5.0f,
+                         0.001f, 0.999f,
+                         0.5f, 2.0f);
+  mat4x4 t = translate(vec3(0, 0, 3));
+  mat4x4 view_to_clip = aspect * p * t;
+
+  _3d_clear(shaders);
+  vec3 light_pos = _3d_light(shaders, view_to_clip, theta);
+  _3d_cube(shaders, view_to_clip, theta, light_pos);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // padding
+  //////////////////////////////////////////////////////////////////////////////
+
+  while ((ib_ix % 8) != 0) {
+    TU(0x80000000);
+  }
+
+  return ib_ix;
+}
+
+int _tile_texture(const shaders& shaders,
+                  int input_reloc_index,
+                  int output_reloc_index)
+{
+  int width = 1024;
+  int height = 1024;
+  int pitch = width;
+  float x = (float)width * 0.5f;
+  float y = (float)height * 0.5f;
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  ib_viewport(width, height);
+  ib_colorbuffer(output_reloc_index, pitch, 1, 1); // macrotile, microtile
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__RED
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__BLUE
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_3
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  // shaders
+
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  // GA
+
+  T0V(GB_ENABLE
+    , GB_ENABLE__POINT_STUFF_ENABLE(1)
+    | GB_ENABLE__TEX0_SOURCE(2) // stuff with source texture coordinates s,t
+    );
+
+  T0Vf(GA_POINT_S0, 0.0f);
+  T0Vf(GA_POINT_T0, 1.0f);
+  T0Vf(GA_POINT_S1, 1.0f);
+  T0Vf(GA_POINT_T1, 0.0f);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // RS
+  //////////////////////////////////////////////////////////////////////////////
+
+  int rs_instructions = 1;
+
+  ib_rs_instructions(0);
+
+  T0V(RS_IP_0
+    , RS_IP__TEX_PTR_S(0)
+    | RS_IP__TEX_PTR_T(1)
+    | RS_IP__TEX_PTR_R(62) // constant 0.0
+    | RS_IP__TEX_PTR_Q(63) // constant 1.0
+    );
+
+  T0V(RS_COUNT
+      , RS_COUNT__IT_COUNT(2)
+      | RS_COUNT__IC_COUNT(0)
+      | RS_COUNT__W_ADDR(0)
+      | RS_COUNT__HIRES_EN(1)
+      );
+
+  T0V(RS_INST_COUNT
+      , RS_INST_COUNT__INST_COUNT(rs_instructions - 1));
+
+  T0V(RS_INST_0
+      , RS_INST__TEX_ID(0)
+      | RS_INST__TEX_CN(1)
+      | RS_INST__TEX_ADDR(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1)
+      );
+  T0V(VAP_OUT_VTX_FMT_1
+      , 0
+      );
+
+  //
+
+  T0V(ZB_CNTL, 0);
+  T0V(ZB_ZSTENCILCNTL, 0);
+
+  //
+
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 2; // clamp to [0.0, 1.0]
+  ib_texture__1(input_reloc_index,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__2();
+
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+
+  ib_ga_us(&shaders.fragment[TEXTURE_TILE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[TEXTURE_TILE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT((int)(y * 12.0f))
+      | GA_POINT_SIZE__WIDTH((int)(x * 12.0f))
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float center[] = {
+    x, y,
+  };
+  const int vertex_count = 1;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(1) // point list
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < 2; i++) {
+    TF(center[i]);
+  }
+
+  return ib_ix;
+}
+
+const char * textures[] = {
+  "../texture/butterfly_1024x1024_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  int flush_handle;
+  int test_handle;
+
+  void * colorbuffer_ptr[2];
+  void * zbuffer_ptr;
+  void * test_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  test_handle = create_buffer(fd, 1600 * 1200 * 4, &test_ptr);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+  fprintf(stderr, "test handle %d\n", test_handle);
+
+  int colorbuffer_ix = 0;
+  float theta = 0;
+
+  {
+    int ib_dwords = _tile_texture(shaders,
+                                  TEXTUREBUFFER_RELOC_INDEX, // input
+                                  COLORBUFFER_RELOC_INDEX);  // output
+
+    //int ib_dwords = indirect_buffer(shaders, theta);
+
+    printf("here2\n");
+
+    drm_radeon_cs(fd,
+                  test_handle, // colorbuffer
+                  zbuffer_handle, // unused
+                  flush_handle,
+                  texturebuffer_handle,
+                  textures_length,
+                  ib_dwords);
+  }
+
+  while (true) {
+    int ib_dwords = indirect_buffer(shaders, theta);
+
+    int ret = drm_radeon_cs(fd,
+                            colorbuffer_handle[colorbuffer_ix],
+                            zbuffer_handle,
+                            flush_handle,
+                            //texturebuffer_handle,
+                            //textures_length,
+                            &test_handle,
+                            1,
+                            ib_dwords);
+    if (ret == -1)
+      break;
+
+    primary_surface_address(rmmio, colorbuffer_ix);
+
+    // next state
+    theta += 0.01f;
+    colorbuffer_ix = (colorbuffer_ix + 1) & 1;
+  }
+
+
+  {
+    printf("test.data\n");
+    int out_fd = open("test.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, test_ptr, 1024 * 1024 * 4);
+    assert(write_length == 1024 * 1024 * 4);
+    close(out_fd);
+  }
+
+  close(fd);
+}

src/particle.cpp

@@ -0,0 +1,546 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+#include "../model/plane.h"
+
+#define CLEAR_SHADER 0
+#define PLANE_SHADER 1
+#define PARTICLE_SHADER 2
+#define TEXTURE_TILE_SHADER 3
+
+const char * vertex_shader_paths[] = {
+  "clear.vs.bin",
+  "particle_plane.vs.bin",
+  "particle_particle.vs.bin",
+  "texture_tile.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "clear.fs.bin",
+  "particle_plane.fs.bin",
+  "particle_particle.fs.bin",
+  "texture_tile.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+#define PLANE_TEXTURE 0
+#define PARTICLE_TEXTURE 1
+
+const char * textures[] = {
+  "../texture/plane_32x32_rgba8888.data",
+  "../texture/particle_32x32_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+void _3d_clear(struct shaders& shaders)
+{
+  ib_rs_instructions(0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , 0);
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 7); // always
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[CLEAR_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CLEAR_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT(600 * 12)
+      | GA_POINT_SIZE__WIDTH(800 * 12)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float center[] = {
+    800.0f, 600.0f,
+  };
+  const int vertex_count = 1;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(1) // point list
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < 2; i++) {
+    TF(center[i]);
+  }
+}
+
+mat4x4 perspective(float low1, float high1,
+                   float low2, float high2,
+                   float low3, float high3)
+{
+  float scale2 = (high2 - low2) / (high1 - low1);
+  float scale3 = (high3 - low3) / (high1 - low1);
+
+  mat4x4 m1 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, 1, -low1,
+                     0, 0, 0, 1
+                     );
+
+  mat4x4 m2 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, scale2, low2,
+                     0, 0, scale3, low3
+                     );
+
+  return m2 * m1;
+}
+
+void _3d_plane_inner(mat4x4 trans)
+{
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP_PVS
+  //////////////////////////////////////////////////////////////////////////////
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const model * model = &plane_model;
+  const object * obj = model->object[0];
+  const int triangle_count = obj->triangle_count;
+  const int vertex_count = triangle_count * 3;
+
+  const int dwords_per_vtx = 5;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(4)
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+
+  for (int i = 0; i < triangle_count; i++) {
+    for (int j = 0; j < 3; j++) {
+      vec3 p = model->position[obj->triangle[i][j].position];
+      vec2 t = model->texture[obj->triangle[i][j].texture];
+
+      TF(p.x);
+      TF(p.y);
+      TF(p.z);
+      TF(t.x);
+      TF(t.y);
+    }
+  }
+}
+
+void _3d_plane(struct shaders& shaders,
+               const mat4x4& world_to_clip,
+               float theta)
+{
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PLANE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__32();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PLANE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PLANE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 rx = rotate_x(-PI / 2.0f);
+  mat4x4 local_to_world = s * rx;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  _3d_plane_inner(trans);
+}
+
+void _3d_particle(struct shaders& shaders,
+                  const mat4x4& world_to_clip,
+                  float theta)
+{
+  // enable blending
+  T0V(RB3D_BLENDCNTL
+      , RB3D_BLENDCNTL__ALPHA_BLEND_ENABLE__ENABLE
+      | RB3D_BLENDCNTL__READ_ENABLE(1)
+      | RB3D_BLENDCNTL__SRCBLEND__GL_ONE
+      | RB3D_BLENDCNTL__DESTBLEND__GL_ONE
+      | RB3D_BLENDCNTL__SRC_ALPHA_0_NO_READ(0)
+      | RB3D_BLENDCNTL__SRC_ALPHA_1_NO_READ(0)
+      );
+
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PARTICLE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__32();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PLANE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PLANE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 local_to_world = s;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  _3d_plane_inner(trans);
+}
+
+int indirect_buffer(shaders& shaders,
+                    float theta)
+{
+  int width = 1600;
+  int height = 1200;
+  int pitch = width;
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  T0V(RB3D_BLENDCNTL, 0);
+  T0V(RB3D_ABLENDCNTL, 0);
+
+  ib_viewport(width, height);
+  ib_colorbuffer(COLORBUFFER_RELOC_INDEX, pitch, 0, 0);
+
+  T0V(GB_ENABLE, 0);
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 aspect = scale(vec3(3.0f/4.0f, 1, 1));
+  mat4x4 p = perspective(0.01f, 3.0f,
+                         0.001f, 0.999f,
+                         1.0f, 3.0f);
+  mat4x4 t = translate(vec3(0, 0, 1));
+  mat4x4 rx = rotate_x(-PI / 8.0f);
+  mat4x4 ry = rotate_y(theta * 0.8f);
+  mat4x4 world_to_view = t * rx * ry;
+
+  mat4x4 world_to_clip = aspect * p * world_to_view;
+
+  _3d_clear(shaders);
+  _3d_plane(shaders, world_to_clip, theta);
+  _3d_particle(shaders, world_to_clip, theta);
+  //_3d_zbuffer(shaders);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // padding
+  //////////////////////////////////////////////////////////////////////////////
+
+  while ((ib_ix % 8) != 0) {
+    TU(0x80000000);
+  }
+
+  return ib_ix;
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  int flush_handle;
+
+  void * colorbuffer_ptr[2];
+  void * zbuffer_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  int colorbuffer_ix = 0;
+  float theta = PI * 0.5;
+
+  while (true) {
+    int ib_dwords = indirect_buffer(shaders, theta);
+
+    drm_radeon_cs(fd,
+                  colorbuffer_handle[colorbuffer_ix],
+                  zbuffer_handle,
+                  flush_handle,
+                  texturebuffer_handle,
+                  textures_length,
+                  ib_dwords);
+
+    primary_surface_address(rmmio, colorbuffer_ix);
+
+    // next state
+    theta += 0.01f;
+    colorbuffer_ix = (colorbuffer_ix + 1) & 1;
+    break;
+  }
+
+  {
+    printf("colorbuffer0.data\n");
+    int out_fd = open("colorbuffer0.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, colorbuffer_ptr[0], colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+  {
+    printf("zbuffer.data\n");
+    int out_fd = open("zbuffer.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, zbuffer_ptr, colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+
+  close(fd);
+}

src/particle_oriented.cpp

@@ -0,0 +1,574 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+#include "../model/plane.h"
+
+#define CLEAR_SHADER 0
+#define PLANE_SHADER 1
+#define PARTICLE_SHADER 2
+#define TEXTURE_TILE_SHADER 3
+
+const char * vertex_shader_paths[] = {
+  "clear.vs.bin",
+  "particle_plane.vs.bin",
+  "particle_particle.vs.bin",
+  "texture_tile.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "clear.fs.bin",
+  "particle_plane.fs.bin",
+  "particle_particle.fs.bin",
+  "texture_tile.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+#define PLANE_TEXTURE 0
+#define PARTICLE_TEXTURE 1
+
+const char * textures[] = {
+  "../texture/plane_32x32_rgba8888.data",
+  "../texture/particle_32x32_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+void _3d_clear(struct shaders& shaders)
+{
+  ib_rs_instructions(0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , 0);
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 7); // always
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[CLEAR_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CLEAR_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT(600 * 12)
+      | GA_POINT_SIZE__WIDTH(800 * 12)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float center[] = {
+    800.0f, 600.0f,
+  };
+  const int vertex_count = 1;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(1) // point list
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < 2; i++) {
+    TF(center[i]);
+  }
+}
+
+mat4x4 perspective(float low1, float high1,
+                   float low2, float high2,
+                   float low3, float high3)
+{
+  float scale2 = (high2 - low2) / (high1 - low1);
+  float scale3 = (high3 - low3) / (high1 - low1);
+
+  mat4x4 m1 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, 1, -low1,
+                     0, 0, 0, 1
+                     );
+
+  mat4x4 m2 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, scale2, low2,
+                     0, 0, scale3, low3
+                     );
+
+  return m2 * m1;
+}
+
+void _3d_plane_inner(mat4x4 trans)
+{
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const model * model = &plane_model;
+  const object * obj = model->object[0];
+  const int triangle_count = obj->triangle_count;
+  const int vertex_count = triangle_count * 3;
+
+  const int dwords_per_vtx = 5;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(4)
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+
+  for (int i = 0; i < triangle_count; i++) {
+    for (int j = 0; j < 3; j++) {
+      vec3 p = model->position[obj->triangle[i][j].position];
+      vec2 t = model->texture[obj->triangle[i][j].texture];
+
+      TF(p.x);
+      TF(p.y);
+      TF(p.z);
+      TF(t.x);
+      TF(t.y);
+    }
+  }
+}
+
+void _3d_plane(struct shaders& shaders,
+               const mat4x4& world_to_clip,
+               float theta)
+{
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PLANE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__32();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PLANE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PLANE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 rx = rotate_x(-PI / 2.0f);
+  mat4x4 local_to_world = s * rx;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // consts
+  //////////////////////////////////////////////////////////////////////////////
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  // plane_inner
+
+  _3d_plane_inner(trans);
+}
+
+void _3d_particle(struct shaders& shaders,
+                  const mat4x4& world_to_clip,
+                  const mat4x4& world_to_view,
+                  float theta)
+{
+  // enable blending
+  T0V(RB3D_BLENDCNTL
+      , RB3D_BLENDCNTL__ALPHA_BLEND_ENABLE__ENABLE
+      | RB3D_BLENDCNTL__READ_ENABLE(1)
+      | RB3D_BLENDCNTL__SRCBLEND__GL_ONE
+      | RB3D_BLENDCNTL__DESTBLEND__GL_ONE
+      | RB3D_BLENDCNTL__SRC_ALPHA_0_NO_READ(0)
+      | RB3D_BLENDCNTL__SRC_ALPHA_1_NO_READ(0)
+      );
+
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PARTICLE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__32();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PARTICLE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PARTICLE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 local_to_world = s;
+
+  mat4x4 local_to_view = world_to_view * local_to_world;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // consts
+  //////////////////////////////////////////////////////////////////////////////
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+
+    // 4: particle_position
+    0, 0, 0, 0,
+
+    // 5: dx (right)
+    local_to_view[0][0], local_to_view[0][1], local_to_view[0][2], 0,
+
+    // 6: dy (up)
+    local_to_view[1][0], local_to_view[1][1], local_to_view[1][2], 0,
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  // plane_inner
+
+  _3d_plane_inner(trans);
+}
+
+int indirect_buffer(shaders& shaders,
+                    float theta)
+{
+  int width = 1600;
+  int height = 1200;
+  int pitch = width;
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  T0V(RB3D_BLENDCNTL, 0);
+  T0V(RB3D_ABLENDCNTL, 0);
+
+  ib_viewport(width, height);
+  ib_colorbuffer(COLORBUFFER_RELOC_INDEX, pitch, 0, 0);
+
+  T0V(GB_ENABLE, 0);
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 aspect = scale(vec3(3.0f/4.0f, 1, 1));
+  mat4x4 p = perspective(0.01f, 3.0f,
+                         0.001f, 0.999f,
+                         1.0f, 3.0f);
+  mat4x4 t = translate(vec3(0, 0, 1));
+  mat4x4 rx = rotate_x(-PI / 8.0f);
+  mat4x4 ry = rotate_y(theta * 0.8f);
+  mat4x4 world_to_view = t * rx * ry;
+
+  mat4x4 world_to_clip = aspect * p * world_to_view;
+
+  _3d_clear(shaders);
+  _3d_plane(shaders, world_to_clip, theta);
+  _3d_particle(shaders, world_to_clip, world_to_view, theta);
+  //_3d_zbuffer(shaders);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // padding
+  //////////////////////////////////////////////////////////////////////////////
+
+  while ((ib_ix % 8) != 0) {
+    TU(0x80000000);
+  }
+
+  return ib_ix;
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  int flush_handle;
+
+  void * colorbuffer_ptr[2];
+  void * zbuffer_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  int colorbuffer_ix = 0;
+  float theta = PI * 0.5;
+
+  while (true) {
+    int ib_dwords = indirect_buffer(shaders, theta);
+
+    drm_radeon_cs(fd,
+                  colorbuffer_handle[colorbuffer_ix],
+                  zbuffer_handle,
+                  flush_handle,
+                  texturebuffer_handle,
+                  textures_length,
+                  ib_dwords);
+
+    primary_surface_address(rmmio, colorbuffer_ix);
+
+    // next state
+    theta += 0.01f;
+    colorbuffer_ix = (colorbuffer_ix + 1) & 1;
+  }
+
+  {
+    printf("colorbuffer0.data\n");
+    int out_fd = open("colorbuffer0.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, colorbuffer_ptr[0], colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+  {
+    printf("zbuffer.data\n");
+    int out_fd = open("zbuffer.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, zbuffer_ptr, colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+
+  close(fd);
+}

src/particle_oriented_animated.cpp

@@ -0,0 +1,698 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+#include "../model/plane.h"
+
+#define CLEAR_SHADER 0
+#define PLANE_SHADER 1
+#define PARTICLE_SHADER 2
+#define TEXTURE_TILE_SHADER 3
+
+const char * vertex_shader_paths[] = {
+  "clear.vs.bin",
+  "particle_plane.vs.bin",
+  "particle_particle_animated.vs.bin",
+  "texture_tile.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "clear.fs.bin",
+  "particle_plane.fs.bin",
+  "particle_particle.fs.bin",
+  "texture_tile.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+#define PLANE_TEXTURE 0
+#define PARTICLE_TEXTURE 1
+
+const char * textures[] = {
+  "../texture/plane_32x32_rgba8888.data",
+  "../texture/particle_32x32_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+static inline uint32_t xorshift32(uint32_t state)
+{
+  uint32_t x = state;
+  x ^= x << 13;
+  x ^= x >> 17;
+  x ^= x << 5;
+  return x;
+}
+
+static inline float xorshift32f(uint32_t& state)
+{
+  state = xorshift32(state);
+  return (float)(state & 0xffffff) * (1.0f / 16777215.0f);
+}
+
+const float max_age = 3.0f;
+
+struct particle {
+  vec3 position;
+  float time;
+  float delta;
+  vec3 velocity;
+};
+
+void _3d_clear(const shaders& shaders)
+{
+  ib_rs_instructions(0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , 0);
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 7); // always
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[CLEAR_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CLEAR_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT(600 * 12)
+      | GA_POINT_SIZE__WIDTH(800 * 12)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float center[] = {
+    800.0f, 600.0f,
+  };
+  const int vertex_count = 1;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(1) // point list
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < 2; i++) {
+    TF(center[i]);
+  }
+}
+
+mat4x4 perspective(float low1, float high1,
+                   float low2, float high2,
+                   float low3, float high3)
+{
+  float scale2 = (high2 - low2) / (high1 - low1);
+  float scale3 = (high3 - low3) / (high1 - low1);
+
+  mat4x4 m1 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, 1, -low1,
+                     0, 0, 0, 1
+                     );
+
+  mat4x4 m2 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, scale2, low2,
+                     0, 0, scale3, low3
+                     );
+
+  return m2 * m1;
+}
+
+void _3d_plane_inner(mat4x4 trans)
+{
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const model * model = &plane_model;
+  const object * obj = model->object[0];
+  const int triangle_count = obj->triangle_count;
+  const int vertex_count = triangle_count * 3;
+
+  const int dwords_per_vtx = 5;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(4)
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+
+  for (int i = 0; i < triangle_count; i++) {
+    for (int j = 0; j < 3; j++) {
+      vec3 p = model->position[obj->triangle[i][j].position];
+      vec2 t = model->texture[obj->triangle[i][j].texture];
+
+      TF(p.x);
+      TF(p.y);
+      TF(p.z);
+      TF(t.x);
+      TF(t.y);
+    }
+  }
+}
+
+void _3d_plane(const shaders& shaders,
+               const mat4x4& world_to_clip,
+               float theta)
+{
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PLANE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__32();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PLANE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PLANE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 rx = rotate_x(-PI / 2.0f);
+  mat4x4 local_to_world = s * rx;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // consts
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  // plane_inner
+
+  _3d_plane_inner(trans);
+}
+
+void _3d_particle(const shaders& shaders,
+                  const mat4x4& world_to_clip,
+                  const mat4x4& world_to_view,
+                  const particle * particles,
+                  const int particles_length,
+                  const float theta)
+{
+  // enable blending
+  T0V(RB3D_BLENDCNTL
+      , RB3D_BLENDCNTL__ALPHA_BLEND_ENABLE__ENABLE
+      | RB3D_BLENDCNTL__READ_ENABLE(1)
+      | RB3D_BLENDCNTL__SRCBLEND__GL_ONE
+      | RB3D_BLENDCNTL__DESTBLEND__GL_ONE
+      | RB3D_BLENDCNTL__SRC_ALPHA_0_NO_READ(0)
+      | RB3D_BLENDCNTL__SRC_ALPHA_1_NO_READ(0)
+      );
+
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  //ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+  T0V(ZB_CNTL
+      , 0
+      );
+  T0V(ZB_ZSTENCILCNTL
+      , 0
+      );
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PARTICLE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__32();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PARTICLE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PARTICLE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 local_to_world = s;
+
+  mat4x4 local_to_view = world_to_view * local_to_world;
+
+  mat4x4 local_to_clip = world_to_clip * local_to_world;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // consts
+  //////////////////////////////////////////////////////////////////////////////
+
+  for (int i = 0; i < particles_length; i++) {
+    T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+    const vec3 position = particles[i].position;
+
+    // 6: position
+    const float scale = 0.005f;
+    //const float position_consts[] = { position.x, position.y, position.z, scale };
+    //ib_vap_pvs_const_offset(position_consts, (sizeof (position_consts)), 6);
+
+    const float consts[] = {
+      // 0: local space to clip space transformation matrix
+      local_to_clip[0][0], local_to_clip[0][1], local_to_clip[0][2], local_to_clip[0][3],
+      local_to_clip[1][0], local_to_clip[1][1], local_to_clip[1][2], local_to_clip[1][3],
+      local_to_clip[2][0], local_to_clip[2][1], local_to_clip[2][2], local_to_clip[2][3],
+      local_to_clip[3][0], local_to_clip[3][1], local_to_clip[3][2], local_to_clip[3][3],
+
+      // 4: dx ("right" change of basis vector)
+      local_to_view[0][0], local_to_view[0][1], local_to_view[0][2], 0,
+
+      // 5: dy ("up" change of basis vector)
+      local_to_view[1][0], local_to_view[1][1], local_to_view[1][2], 0,
+
+      // 6: particle position, scale
+      position.x, position.y, position.z, scale
+    };
+    ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+    // plane_inner
+
+    _3d_plane_inner(local_to_clip);
+  }
+}
+
+int indirect_buffer(const shaders& shaders,
+                    const particle * particles,
+                    const int particles_length,
+                    float theta)
+{
+  int width = 1600;
+  int height = 1200;
+  int pitch = width;
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  T0V(RB3D_BLENDCNTL, 0);
+  T0V(RB3D_ABLENDCNTL, 0);
+
+  ib_viewport(width, height);
+  ib_colorbuffer(COLORBUFFER_RELOC_INDEX, pitch, 0, 0);
+
+  T0V(GB_ENABLE, 0);
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 aspect = scale(vec3(3.0f/4.0f, 1, 1));
+  mat4x4 p = perspective(0.01f, 3.0f,
+                         0.001f, 0.999f,
+                         1.0f, 3.0f);
+  mat4x4 t = translate(vec3(0, 0, 1));
+  mat4x4 rx = rotate_x(-PI / 8.0f);
+  mat4x4 ry = rotate_y(theta * 0.8f);
+  mat4x4 world_to_view = t * rx * ry;
+
+  mat4x4 world_to_clip = aspect * p * world_to_view;
+
+  _3d_clear(shaders);
+  _3d_plane(shaders, world_to_clip, theta);
+  _3d_particle(shaders,
+               world_to_clip,
+               world_to_view,
+               particles,
+               particles_length,
+               theta);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // padding
+  //////////////////////////////////////////////////////////////////////////////
+
+  while ((ib_ix % 8) != 0) {
+    TU(0x80000000);
+  }
+
+  assert((unsigned int)ib_ix < (sizeof (ib)) / (sizeof (ib[0])));
+
+  return ib_ix;
+}
+
+void reset_particle(particle& p)
+{
+  //vec3 pos = normalize(p.position);
+
+  p.position = normalize(vec3(p.velocity.x,
+                              0,
+                              p.velocity.z)) * 20.0f;
+
+  //printf("position %f %f %f\n", p.position.x, p.position.y, p.position.z);
+
+  p.velocity = vec3(p.velocity.x,
+                    2.0f * p.delta,
+                    p.velocity.z);
+
+  //printf("velocity %f %f %f\n\n", p.velocity.x, p.velocity.y, p.velocity.z);
+}
+
+void init_particles(particle * particles, const int particles_length)
+{
+  uint32_t state = 0x12345678;
+
+  const float rl = 1.0f / (float)(particles_length);
+
+  for (int i = 0; i < particles_length; i++) {
+    float fi = ((float)i);
+
+    float sx = xorshift32f(state) * 2.0f - 1.0f;
+    float sy = xorshift32f(state) * 2.0f - 1.0f;
+    float sz = xorshift32f(state) * 2.0f - 1.0f;
+
+    float delta = xorshift32f(state) * 0.5f + 0.5f;
+
+    float vx = xorshift32f(state) * 2.0f - 1.0f;
+    float vz = xorshift32f(state) * 2.0f - 1.0f;
+
+    particles[i].time = max_age * sinf(fi * rl * 2) * 0.5f + 0.5f;
+    particles[i].delta = delta;
+    particles[i].position.x = sx;
+    particles[i].position.y = sy;
+    particles[i].position.z = sz;
+    particles[i].velocity = normalize(vec3(vx * 0.5f, 0.0f, vz * 0.5f));
+  }
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  int flush_handle;
+
+  void * colorbuffer_ptr[2];
+  void * zbuffer_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  int colorbuffer_ix = 0;
+  float theta = PI * 0.5;
+
+  particle particles[1000] = {};
+  const int particles_length = (sizeof (particles)) / (sizeof (particles[0]));
+  init_particles(particles, particles_length);
+
+  while (true) {
+    int ib_dwords = indirect_buffer(shaders,
+                                    particles,
+                                    particles_length,
+                                    theta);
+
+    int ret = drm_radeon_cs(fd,
+                            colorbuffer_handle[colorbuffer_ix],
+                            zbuffer_handle,
+                            flush_handle,
+                            texturebuffer_handle,
+                            textures_length,
+                            ib_dwords);
+    if (ret == -1)
+      break;
+
+    primary_surface_address(rmmio, colorbuffer_ix);
+
+    // next state
+    theta += 0.01f;
+    colorbuffer_ix = (colorbuffer_ix + 1) & 1;
+
+    //
+    // update particles
+    //
+    for (int i = 0; i < particles_length; i++) {
+      if (particles[i].time <= 0) {
+        particles[i].time += max_age;
+        reset_particle(particles[i]);
+      } else {
+        particles[i].time -= 0.01f;
+        particles[i].position += vec3(particles[i].velocity.x * 0.9f,
+                                      particles[i].velocity.y * 5.0f,
+                                      particles[i].velocity.z * 0.9f);
+        particles[i].velocity += vec3(0, -0.04, 0);
+        if (particles[i].position.y < 0) {
+          particles[i].position.y = fabsf(particles[i].position.y);
+          particles[i].velocity.y *= -0.6f;
+        }
+      }
+    }
+  }
+
+  {
+    printf("colorbuffer0.data\n");
+    int out_fd = open("colorbuffer0.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, colorbuffer_ptr[0], colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+  {
+    printf("zbuffer.data\n");
+    int out_fd = open("zbuffer.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, zbuffer_ptr, colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+
+  close(fd);
+}

src/particle_oriented_animated_fan.cpp

@@ -0,0 +1,774 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+
+#define CLEAR_SHADER 0
+#define PLANE_SHADER 1
+#define PARTICLE_SHADER 2
+#define TEXTURE_TILE_SHADER 3
+
+const char * vertex_shader_paths[] = {
+  "clear.vs.bin",
+  "particle_plane_fan.vs.bin",
+  "particle_particle_animated_fan.vs.bin",
+  "texture_tile.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "clear.fs.bin",
+  "particle_plane.fs.bin",
+  "particle_particle.fs.bin",
+  "texture_tile.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+#define PLANE_TEXTURE 0
+#define PARTICLE_TEXTURE 1
+
+const char * textures[] = {
+  "../texture/plane_32x32_rgba8888.data",
+  "../texture/particle_32x32_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+static inline uint32_t xorshift32(uint32_t state)
+{
+  uint32_t x = state;
+  x ^= x << 13;
+  x ^= x >> 17;
+  x ^= x << 5;
+  return x;
+}
+
+static inline float xorshift32f(uint32_t& state)
+{
+  state = xorshift32(state);
+  return (float)(state & 0xffffff) * (1.0f / 16777215.0f);
+}
+
+const float max_age = 3.0f;
+
+struct particle {
+  vec3 position;
+  float time;
+  float delta;
+  vec3 velocity;
+};
+
+void _3d_clear(const shaders& shaders)
+{
+  ib_rs_instructions(0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , 0);
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 7); // always
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[CLEAR_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CLEAR_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT(600 * 12)
+      | GA_POINT_SIZE__WIDTH(800 * 12)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float center[] = {
+    800.0f, 600.0f,
+  };
+  const int vertex_count = 1;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(1) // point list
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < 2; i++) {
+    TF(center[i]);
+  }
+}
+
+mat4x4 perspective(float low1, float high1,
+                   float low2, float high2,
+                   float low3, float high3)
+{
+  float scale2 = (high2 - low2) / (high1 - low1);
+  float scale3 = (high3 - low3) / (high1 - low1);
+
+  mat4x4 m1 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, 1, -low1,
+                     0, 0, 0, 1
+                     );
+
+  mat4x4 m2 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, scale2, low2,
+                     0, 0, scale3, low3
+                     );
+
+  return m2 * m1;
+}
+
+void _3d_plane_inner()
+{
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const vec2 vertices[] = {
+    {0.0, 0.0f},
+    {1.0, 0.0f},
+    {1.0, 1.0f},
+    {0.0, 1.0f},
+  };
+
+  const int vertex_count = 4;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(5) // triangle fan
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < vertex_count; i++) {
+    TF(vertices[i].x);
+    TF(vertices[i].y);
+  }
+}
+
+void _3d_particle_inner()
+{
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(3)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // AOS
+  //////////////////////////////////////////////////////////////////////////////
+
+  T3(_3D_LOAD_VBPNTR, (3 - 1));
+  TU( // VAP_VTX_NUM_ARRAYS
+      VAP_VTX_NUM_ARRAYS__VTX_NUM_ARRAYS(1)
+    | VAP_VTX_NUM_ARRAYS__VC_FORCE_PREFETCH(1)
+    );
+  TU( // VAP_VTX_AOS_ATTR01
+      VAP_VTX_AOS_ATTR__VTX_AOS_COUNT0(2)
+    | VAP_VTX_AOS_ATTR__VTX_AOS_STRIDE0(2)
+    );
+  TU( // VAP_VTX_AOS_ADDR0
+      (4 * 0);
+    );
+
+  T3(_NOP, 0);
+  TU(VERTEXBUFFER_RELOC_INDEX * 4); // index into relocs array for VAP_VTX_AOS_ADDR0
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int vertex_count = 4;
+  T3(_3D_DRAW_VBUF_2, (1 - 1));
+  TU( VAP_VF_CNTL__PRIM_TYPE(5) // triangle fan
+    | VAP_VF_CNTL__PRIM_WALK(2) // vertex list (data fetched from memory)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+}
+
+void _3d_plane(const shaders& shaders,
+               const mat4x4& world_to_clip,
+               float theta)
+{
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PLANE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PLANE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PLANE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 rx = rotate_x(-PI / 2.0f);
+  mat4x4 local_to_world = s * rx;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // consts
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+    // 4
+    -2.0f, 0, 0, 0,
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  // plane_inner
+
+  _3d_plane_inner();
+}
+
+void _3d_particle(const shaders& shaders,
+                  const mat4x4& world_to_clip,
+                  const mat4x4& world_to_view,
+                  const particle * particles,
+                  const int particles_length,
+                  const float theta)
+{
+  // enable blending
+  T0V(RB3D_BLENDCNTL
+      , RB3D_BLENDCNTL__ALPHA_BLEND_ENABLE__ENABLE
+      | RB3D_BLENDCNTL__READ_ENABLE(1)
+      | RB3D_BLENDCNTL__SRCBLEND__GL_ONE
+      | RB3D_BLENDCNTL__DESTBLEND__GL_ONE
+      | RB3D_BLENDCNTL__SRC_ALPHA_0_NO_READ(0)
+      | RB3D_BLENDCNTL__SRC_ALPHA_1_NO_READ(0)
+      );
+
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  //ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+  T0V(ZB_CNTL
+      , 0
+      );
+  T0V(ZB_ZSTENCILCNTL
+      , 0
+      );
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PARTICLE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PARTICLE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PARTICLE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 local_to_world = s;
+
+  mat4x4 local_to_view = world_to_view * local_to_world;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // consts
+  //////////////////////////////////////////////////////////////////////////////
+
+  for (int i = 0; i < particles_length; i++) {
+    T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+    const vec3 position = particles[i].position;
+
+    // 6: position
+    const float scale = 0.005f;
+    //const float position_consts[] = { position.x, position.y, position.z, scale };
+    //ib_vap_pvs_const_offset(position_consts, (sizeof (position_consts)), 6);
+
+    const float consts[] = {
+      // 0
+      trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+      trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+      trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+      trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+
+      // 4: dx (right)
+      local_to_view[0][0], local_to_view[0][1], local_to_view[0][2], 0,
+
+      // 5: dy (up)
+      local_to_view[1][0], local_to_view[1][1], local_to_view[1][2], 0,
+
+      // 6: xyz:position w:scale
+      position.x, position.y, position.z, scale,
+
+      // 7:
+      -2.0, 0, 0, 0,
+    };
+    ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+    // plane_inner
+    //_3d_plane_inner();
+    _3d_particle_inner();
+  }
+}
+
+int indirect_buffer(const shaders& shaders,
+                    const particle * particles,
+                    const int particles_length,
+                    float theta)
+{
+  int width = 1600;
+  int height = 1200;
+  int pitch = width;
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  T0V(RB3D_BLENDCNTL, 0);
+  T0V(RB3D_ABLENDCNTL, 0);
+
+  ib_viewport(width, height);
+  ib_colorbuffer(COLORBUFFER_RELOC_INDEX, pitch, 0, 0);
+
+  T0V(GB_ENABLE, 0);
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 aspect = scale(vec3(3.0f/4.0f, 1, 1));
+  mat4x4 p = perspective(0.01f, 3.0f,
+                         0.001f, 0.999f,
+                         1.0f, 3.0f);
+  mat4x4 t = translate(vec3(0, 0, 1));
+  mat4x4 rx = rotate_x(-PI / 8.0f);
+  mat4x4 ry = rotate_y(theta * 0.8f);
+  mat4x4 world_to_view = t * rx * ry;
+
+  mat4x4 world_to_clip = aspect * p * world_to_view;
+
+  _3d_clear(shaders);
+  _3d_plane(shaders, world_to_clip, theta);
+  _3d_particle(shaders,
+               world_to_clip,
+               world_to_view,
+               particles,
+               particles_length,
+               theta);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // padding
+  //////////////////////////////////////////////////////////////////////////////
+
+  while ((ib_ix % 8) != 0) {
+    TU(0x80000000);
+  }
+
+  assert((unsigned int)ib_ix < (sizeof (ib)) / (sizeof (ib[0])));
+
+  return ib_ix;
+}
+
+void reset_particle(particle& p)
+{
+  //vec3 pos = normalize(p.position);
+
+  p.position = normalize(vec3(p.velocity.x,
+                              0,
+                              p.velocity.z)) * 20.0f;
+
+  //printf("position %f %f %f\n", p.position.x, p.position.y, p.position.z);
+
+  p.velocity = vec3(p.velocity.x,
+                    2.0f * p.delta,
+                    p.velocity.z);
+
+  //printf("velocity %f %f %f\n\n", p.velocity.x, p.velocity.y, p.velocity.z);
+}
+
+void init_particles(particle * particles, const int particles_length)
+{
+  uint32_t state = 0x12345678;
+
+  const float rl = 1.0f / (float)(particles_length);
+
+  for (int i = 0; i < particles_length; i++) {
+    float fi = ((float)i);
+
+    float sx = xorshift32f(state) * 2.0f - 1.0f;
+    float sy = xorshift32f(state) * 2.0f - 1.0f;
+    float sz = xorshift32f(state) * 2.0f - 1.0f;
+
+    float delta = xorshift32f(state) * 0.5f + 0.5f;
+
+    float vx = xorshift32f(state) * 2.0f - 1.0f;
+    float vz = xorshift32f(state) * 2.0f - 1.0f;
+
+    particles[i].time = max_age * sinf(fi * rl * 2) * 0.5f + 0.5f;
+    particles[i].delta = delta;
+    particles[i].position.x = sx;
+    particles[i].position.y = sy;
+    particles[i].position.z = sz;
+    particles[i].velocity = normalize(vec3(vx * 0.5f, 0.0f, vz * 0.5f));
+  }
+}
+
+int init_particles_vertexbuffer(int fd, int particles_length)
+{
+  const vec2 vertices[] = {
+    {0.0, 0.0f},
+    {1.0, 0.0f},
+    {1.0, 1.0f},
+    {0.0, 1.0f},
+  };
+  const int vertex_count = 4;
+
+  const int size = particles_length * vertex_count * 2 * (sizeof (float));
+
+  void * ptr;
+  int handle = create_buffer(fd, size, &ptr);
+
+  float * ptrf = (float*)ptr;
+
+  int ix = 0;
+  for (int j = 0; j < particles_length; j++) {
+    for (int i = 0; i < vertex_count; i++) {
+      ptrf[ix++] = vertices[i].x;
+      ptrf[ix++] = vertices[i].y;
+    }
+  }
+  munmap(ptr, size);
+  printf("init vertexbuffer %d %d\n", ix, size);
+
+  return handle;
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  //int flush_handle;
+  int vertexbuffer_handle;
+
+  void * colorbuffer_ptr[2];
+  void * zbuffer_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  //flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  int colorbuffer_ix = 0;
+  float theta = PI * 0.5;
+
+  particle particles[10] = {};
+  const int particles_length = (sizeof (particles)) / (sizeof (particles[0]));
+  vertexbuffer_handle = init_particles_vertexbuffer(fd, particles_length);
+  init_particles(particles, particles_length);
+  fprintf(stderr, "vertexbuffer handle %d\n", vertexbuffer_handle);
+
+  while (true) {
+    int ib_dwords = indirect_buffer(shaders,
+                                    particles,
+                                    particles_length,
+                                    theta);
+
+    int ret = drm_radeon_cs(fd,
+                            colorbuffer_handle[colorbuffer_ix],
+                            zbuffer_handle,
+                            vertexbuffer_handle,
+                            texturebuffer_handle,
+                            textures_length,
+                            ib_dwords);
+    if (ret == -1)
+      break;
+
+    primary_surface_address(rmmio, colorbuffer_ix);
+
+    // next state
+    theta += 0.01f;
+    colorbuffer_ix = (colorbuffer_ix + 1) & 1;
+
+    //
+    // update particles
+    //
+    for (int i = 0; i < particles_length; i++) {
+      if (particles[i].time <= 0) {
+        particles[i].time += max_age;
+        reset_particle(particles[i]);
+      } else {
+        particles[i].time -= 0.01f;
+        particles[i].position += vec3(particles[i].velocity.x * 0.9f,
+                                      particles[i].velocity.y * 5.0f,
+                                      particles[i].velocity.z * 0.9f);
+        particles[i].velocity += vec3(0, -0.04, 0);
+        if (particles[i].position.y < 0) {
+          particles[i].position.y = fabsf(particles[i].position.y);
+          particles[i].velocity.y *= -0.6f;
+        }
+      }
+    }
+  }
+
+  {
+    printf("colorbuffer0.data\n");
+    int out_fd = open("colorbuffer0.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, colorbuffer_ptr[0], colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+  {
+    printf("zbuffer.data\n");
+    int out_fd = open("zbuffer.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, zbuffer_ptr, colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+
+  close(fd);
+}

src/particle_oriented_animated_quad_vbuf.cpp

@@ -0,0 +1,806 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+
+#define CLEAR_SHADER 0
+#define PLANE_SHADER 1
+#define PARTICLE_SHADER 2
+#define TEXTURE_TILE_SHADER 3
+
+const char * vertex_shader_paths[] = {
+  "clear.vs.bin",
+  "particle_plane_fan.vs.bin",
+  "particle_particle_animated_quad_vbuf.vs.bin",
+  "texture_tile.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "clear.fs.bin",
+  "particle_plane.fs.bin",
+  "particle_particle.fs.bin",
+  "texture_tile.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+#define PLANE_TEXTURE 0
+#define PARTICLE_TEXTURE 1
+
+const char * textures[] = {
+  "../texture/plane_32x32_rgba8888.data",
+  "../texture/particle_32x32_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+static inline uint32_t xorshift32(uint32_t state)
+{
+  uint32_t x = state;
+  x ^= x << 13;
+  x ^= x >> 17;
+  x ^= x << 5;
+  return x;
+}
+
+static inline float xorshift32f(uint32_t& state)
+{
+  state = xorshift32(state);
+  return (float)(state & 0xffffff) * (1.0f / 16777215.0f);
+}
+
+const float max_age = 3.0f;
+
+struct particle {
+  vec3 position;
+  float time;
+  float delta;
+  vec3 velocity;
+};
+
+void _3d_clear(const shaders& shaders)
+{
+  ib_rs_instructions(0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , 0);
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 7); // always
+
+  ib_texture__0();
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+  ib_ga_us(&shaders.fragment[CLEAR_SHADER]);
+  ib_vap_pvs(&shaders.vertex[CLEAR_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP INDEX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_INDEX_OFFSET, 0);
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(0)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT(600 * 12)
+      | GA_POINT_SIZE__WIDTH(800 * 12)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float center[] = {
+    800.0f, 600.0f,
+  };
+  const int vertex_count = 1;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(1) // point list
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < 2; i++) {
+    TF(center[i]);
+  }
+}
+
+mat4x4 perspective(float low1, float high1,
+                   float low2, float high2,
+                   float low3, float high3)
+{
+  float scale2 = (high2 - low2) / (high1 - low1);
+  float scale3 = (high3 - low3) / (high1 - low1);
+
+  mat4x4 m1 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, 1, -low1,
+                     0, 0, 0, 1
+                     );
+
+  mat4x4 m2 = mat4x4(1, 0, 0, 0,
+                     0, 1, 0, 0,
+                     0, 0, scale2, low2,
+                     0, 0, scale3, low3
+                     );
+
+  return m2 * m1;
+}
+
+void _3d_plane_inner()
+{
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const vec2 vertices[] = {
+    {0.0, 0.0f},
+    {1.0, 0.0f},
+    {1.0, 1.0f},
+    {0.0, 1.0f},
+  };
+
+  const int vertex_count = 4;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(5) // triangle fan
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < vertex_count; i++) {
+    TF(vertices[i].x);
+    TF(vertices[i].y);
+  }
+}
+
+void _3d_particle_inner(int particles_length, int position_offset)
+{
+  const int vertex_count = 4 * particles_length;
+
+  assert(vertex_count <= 0xffffff);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VF
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_VF_MAX_VTX_INDX
+      , VAP_VF_MAX_VTX_INDX__MAX_INDX(vertex_count - 1)
+      );
+  T0V(VAP_VF_MIN_VTX_INDX
+      , VAP_VF_MIN_VTX_INDX__MIN_INDX(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // AOS
+  //////////////////////////////////////////////////////////////////////////////
+
+  T3(_3D_LOAD_VBPNTR, (4 - 1));
+  TU( // VAP_VTX_NUM_ARRAYS
+      VAP_VTX_NUM_ARRAYS__VTX_NUM_ARRAYS(2)
+    | VAP_VTX_NUM_ARRAYS__VC_FORCE_PREFETCH(1)
+    );
+  TU( // VAP_VTX_AOS_ATTR01
+      VAP_VTX_AOS_ATTR__VTX_AOS_COUNT0(3)
+    | VAP_VTX_AOS_ATTR__VTX_AOS_STRIDE0(3)
+    | VAP_VTX_AOS_ATTR__VTX_AOS_COUNT1(2)
+    | VAP_VTX_AOS_ATTR__VTX_AOS_STRIDE1(2)
+    );
+  TU( // VAP_VTX_AOS_ADDR0
+      (4 * position_offset);
+    );
+  TU( // VAP_VTX_AOS_ADDR1
+      (4 * 0);
+    );
+
+  T3(_NOP, 0);
+  TU(VERTEXBUFFER_RELOC_INDEX * 4); // index into relocs array for VAP_VTX_AOS_ADDR0
+  T3(_NOP, 0);
+  TU(VERTEXBUFFER_RELOC_INDEX * 4); // index into relocs array for VAP_VTX_AOS_ADDR0
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_ALT_NUM_VERTICES,
+      vertex_count);
+
+  T3(_3D_DRAW_VBUF_2, (1 - 1));
+  TU( VAP_VF_CNTL__PRIM_TYPE(13) // quad list
+    | VAP_VF_CNTL__PRIM_WALK(2) // vertex list (data fetched from memory)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(1)
+    | VAP_VF_CNTL__NUM_VERTICES(0)
+    );
+}
+
+void _3d_plane(const shaders& shaders,
+               const mat4x4& world_to_clip,
+               float theta)
+{
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PLANE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__2();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PLANE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PLANE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 rx = rotate_x(-PI / 2.0f);
+  mat4x4 local_to_world = s * rx;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // consts
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+    // 4
+    -2.0f, 0, 0, 0,
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  // plane_inner
+
+  _3d_plane_inner();
+}
+
+void _3d_particle(const shaders& shaders,
+                  const mat4x4& world_to_clip,
+                  const mat4x4& world_to_view,
+                  const particle * particles,
+                  const int particles_length,
+                  const float theta,
+                  float * vertexbuffer_ptr)
+{
+  // enable blending
+  T0V(RB3D_BLENDCNTL
+      , RB3D_BLENDCNTL__ALPHA_BLEND_ENABLE__ENABLE
+      | RB3D_BLENDCNTL__READ_ENABLE(1)
+      | RB3D_BLENDCNTL__SRCBLEND__GL_ONE
+      | RB3D_BLENDCNTL__DESTBLEND__GL_ONE
+      | RB3D_BLENDCNTL__SRC_ALPHA_0_NO_READ(0)
+      | RB3D_BLENDCNTL__SRC_ALPHA_1_NO_READ(0)
+      );
+
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1));
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4));
+
+  //
+
+  //ib_zbuffer(ZBUFFER_RELOC_INDEX, 1600, 1); // less
+  T0V(ZB_CNTL
+      , 0
+      );
+  T0V(ZB_ZSTENCILCNTL
+      , 0
+      );
+
+  int width = 32;
+  int height = 32;
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 0; // wrap/repeat
+  ib_texture__1(TEXTUREBUFFER_RELOC_INDEX + PARTICLE_TEXTURE,
+                width, height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__32();
+
+  // shaders
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(4)
+      );
+  ib_ga_us(&shaders.fragment[PARTICLE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[PARTICLE_SHADER]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__PS_UCP_MODE(3)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_X_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Z_OFFSET_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(0) // enable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(0)  // enable W division
+      | VAP_VTE_CNTL__VTX_W0_FMT(1)
+      | VAP_VTE_CNTL__SERIAL_PROC_ENA(0)
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // matrix
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 s = scale(1.0f);
+  mat4x4 local_to_world = s;
+
+  mat4x4 local_to_view = world_to_view * local_to_world;
+
+  mat4x4 trans = world_to_clip * local_to_world;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // consts
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_PVS_STATE_FLUSH_REG, 0x00000000);
+
+  const float scale = 0.005f;
+
+  const float consts[] = {
+    // 0
+    trans[0][0], trans[0][1], trans[0][2], trans[0][3],
+    trans[1][0], trans[1][1], trans[1][2], trans[1][3],
+    trans[2][0], trans[2][1], trans[2][2], trans[2][3],
+    trans[3][0], trans[3][1], trans[3][2], trans[3][3],
+
+    // 4: dx (right)
+    local_to_view[0][0], local_to_view[0][1], local_to_view[0][2], 0,
+
+    // 5: dy (up)
+    local_to_view[1][0], local_to_view[1][1], local_to_view[1][2], 0,
+
+    // 6: xyz:position w:scale
+    0, 0, 0, scale,
+
+    // 7:
+    -2.0, 0, 0, 0,
+  };
+  ib_vap_pvs_const_cntl(consts, (sizeof (consts)));
+
+  int offset = particles_length * 4 * 2;
+  int ix = 0;
+  for (int i = 0; i < particles_length; i++) {
+    const vec3& position = particles[i].position;
+    for (int j = 0; j < 4; j++) {
+      vertexbuffer_ptr[offset + ix] = position.x;
+      ix++;
+      vertexbuffer_ptr[offset + ix] = position.y;
+      ix++;
+      vertexbuffer_ptr[offset + ix] = position.z;
+      ix++;
+    };
+  }
+  asm volatile ("" ::: "memory");
+
+  _3d_particle_inner(particles_length, offset);
+}
+
+int indirect_buffer(const shaders& shaders,
+                    const particle * particles,
+                    const int particles_length,
+                    float theta,
+                    float * vertexbuffer_ptr)
+{
+  int width = 1600;
+  int height = 1200;
+  int pitch = width;
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  T0V(RB3D_BLENDCNTL, 0);
+  T0V(RB3D_ABLENDCNTL, 0);
+
+  ib_viewport(width, height);
+  ib_colorbuffer(COLORBUFFER_RELOC_INDEX, pitch, 0, 0);
+
+  T0V(GB_ENABLE, 0);
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  mat4x4 aspect = scale(vec3(3.0f/4.0f, 1, 1));
+  mat4x4 p = perspective(0.01f, 3.0f,
+                         0.001f, 0.999f,
+                         1.0f, 3.0f);
+  mat4x4 t = translate(vec3(0, 0, 1));
+  mat4x4 rx = rotate_x(-PI / 8.0f);
+  mat4x4 ry = rotate_y(theta * 0.8f);
+  mat4x4 world_to_view = t * rx * ry;
+
+  mat4x4 world_to_clip = aspect * p * world_to_view;
+
+  _3d_clear(shaders);
+  _3d_plane(shaders, world_to_clip, theta);
+  _3d_particle(shaders,
+               world_to_clip,
+               world_to_view,
+               particles,
+               particles_length,
+               theta,
+               vertexbuffer_ptr);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // padding
+  //////////////////////////////////////////////////////////////////////////////
+
+  while ((ib_ix % 8) != 0) {
+    TU(0x80000000);
+  }
+
+  assert((unsigned int)ib_ix < (sizeof (ib)) / (sizeof (ib[0])));
+
+  return ib_ix;
+}
+
+void reset_particle(particle& p)
+{
+  //vec3 pos = normalize(p.position);
+
+  p.position = normalize(vec3(p.velocity.x,
+                              0,
+                              p.velocity.z)) * 20.0f;
+
+  //printf("position %f %f %f\n", p.position.x, p.position.y, p.position.z);
+
+  p.velocity = vec3(p.velocity.x,
+                    2.0f * p.delta,
+                    p.velocity.z);
+
+  //printf("velocity %f %f %f\n\n", p.velocity.x, p.velocity.y, p.velocity.z);
+}
+
+void init_particles(particle * particles, const int particles_length)
+{
+  uint32_t state = 0x12345678;
+
+  const float rl = 1.0f / (float)(particles_length);
+
+  for (int i = 0; i < particles_length; i++) {
+    float fi = ((float)i);
+
+    float sx = xorshift32f(state) * 2.0f - 1.0f;
+    float sy = xorshift32f(state) * 2.0f - 1.0f;
+    float sz = xorshift32f(state) * 2.0f - 1.0f;
+
+    float delta = xorshift32f(state) * 0.5f + 0.5f;
+
+    float vx = xorshift32f(state) * 2.0f - 1.0f;
+    float vz = xorshift32f(state) * 2.0f - 1.0f;
+
+    particles[i].time = max_age * sinf(fi * rl * 2) * 0.5f + 0.5f;
+    particles[i].delta = delta;
+    particles[i].position.x = sx;
+    particles[i].position.y = sy;
+    particles[i].position.z = sz;
+    particles[i].velocity = normalize(vec3(vx * 0.5f, 0.0f, vz * 0.5f));
+  }
+}
+
+int init_particles_vertexbuffer(int fd, int particles_length, float ** ptr_out)
+{
+  const vec2 vertices[] = {
+    {0.0, 0.0f},
+    {1.0, 0.0f},
+    {1.0, 1.0f},
+    {0.0, 1.0f},
+  };
+  const int vertex_count = 4;
+
+  const int size = particles_length * vertex_count * 2 * (sizeof (float))
+                 + particles_length * vertex_count * 3 * (sizeof (float));
+  printf("%d size %d\n", particles_length, size);
+
+  void * ptr;
+  int handle = create_buffer(fd, size, &ptr);
+
+  float * ptrf = (float*)ptr;
+
+  int ix = 0;
+  for (int j = 0; j < particles_length; j++) {
+    for (int i = 0; i < vertex_count; i++) {
+      ptrf[ix++] = vertices[i].x;
+      ptrf[ix++] = vertices[i].y;
+    }
+  }
+  printf("init vertexbuffer %d %d\n", ix, size);
+
+  assert(ptr_out != NULL);
+  *ptr_out = ptrf;
+
+  return handle;
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  //int flush_handle;
+  int vertexbuffer_handle;
+
+  void * colorbuffer_ptr[2];
+  void * zbuffer_ptr;
+  float * vertexbuffer_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  //flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  int colorbuffer_ix = 0;
+  float theta = PI * 0.5;
+
+  particle particles[512 * 256] = {};
+  const int particles_length = (sizeof (particles)) / (sizeof (particles[0]));
+  vertexbuffer_handle = init_particles_vertexbuffer(fd, particles_length, &vertexbuffer_ptr);
+  init_particles(particles, particles_length);
+  fprintf(stderr, "vertexbuffer handle %d\n", vertexbuffer_handle);
+
+  while (true) {
+    int ib_dwords = indirect_buffer(shaders,
+                                    particles,
+                                    particles_length,
+                                    theta,
+                                    vertexbuffer_ptr);
+
+    int ret = drm_radeon_cs(fd,
+                            colorbuffer_handle[colorbuffer_ix],
+                            zbuffer_handle,
+                            vertexbuffer_handle,
+                            texturebuffer_handle,
+                            textures_length,
+                            ib_dwords);
+    if (ret == -1)
+      break;
+
+    primary_surface_address(rmmio, colorbuffer_ix);
+
+    // next state
+    theta += 0.01f;
+    colorbuffer_ix = (colorbuffer_ix + 1) & 1;
+
+    //
+    // update particles
+    //
+    for (int i = 0; i < particles_length; i++) {
+      if (particles[i].time <= 0) {
+        particles[i].time += max_age;
+        reset_particle(particles[i]);
+      } else {
+        particles[i].time -= 0.01f;
+        particles[i].position += vec3(particles[i].velocity.x * 0.9f,
+                                      particles[i].velocity.y * 5.0f,
+                                      particles[i].velocity.z * 0.9f);
+        particles[i].velocity += vec3(0, -0.04, 0);
+        if (particles[i].position.y < 0) {
+          particles[i].position.y = fabsf(particles[i].position.y);
+          particles[i].velocity.y *= -0.6f;
+        }
+      }
+    }
+  }
+
+  {
+    printf("colorbuffer0.data\n");
+    int out_fd = open("colorbuffer0.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, colorbuffer_ptr[0], colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+  {
+    printf("zbuffer.data\n");
+    int out_fd = open("zbuffer.data", O_RDWR|O_CREAT, 0644);
+    assert(out_fd >= 0);
+    ssize_t write_length = write(out_fd, zbuffer_ptr, colorbuffer_size);
+    assert(write_length == colorbuffer_size);
+    close(out_fd);
+  }
+
+  close(fd);
+}

src/particle_particle.fs.asm

@@ -0,0 +1,10 @@
+-- temp[0]: texture coordinate
+
+TEX TEX_SEM_WAIT TEX_SEM_ACQUIRE
+  temp[0].rgba = LD tex[0].rgba temp[0].rgaa ;
+
+OUT TEX_SEM_WAIT
+src0.a = temp[0] ,
+src0.rgb = temp[0] :
+  out[0].a    = MAX src0.a src0.a ,
+  out[0].rgb  = MAX src0.rgb src0.rgb ;

src/particle_particle.vs.asm

@@ -0,0 +1,30 @@
+-- const[0-3]: transform matrix
+-- const[4]: particle_position
+-- const[5]: dx
+-- const[6]: dy
+-- input[0]: position coordinate
+-- input[1]: texture coordinate
+
+--
+-- dot(m[0], v), dot(m[1], v), dot(m[2], v), dot(m[3], v)
+--
+
+-- ppos = particle_position
+temp[0].xyz  = VE_ADD  const[4].xyz_  const[4].000_ ;
+
+-- ppos = position.xxx * dx + ppos
+temp[0].xyz  = VE_MAD  input[0].xxx_  const[5].xyz_ temp[0].xyz_ ;
+-- ppos = position.yyy * dy + ppos
+temp[0].xyz  = VE_MAD  input[0].yyy_  const[6].xyz_ temp[0].xyz_ ;
+
+-- scale
+temp[0].xyzw = VE_MUL  temp[0].xyz1   temp[0].1111 ;
+
+-- ppos = transform_matrix * ppos
+temp[1].x    = VE_DOT  const[0].xyzw  temp[0].xyzw ;
+temp[1].y    = VE_DOT  const[1].xyzw  temp[0].xyzw ;
+temp[1].z    = VE_DOT  const[2].xyzw  temp[0].xyzw ;
+temp[1].w    = VE_DOT  const[3].xyzw  temp[0].xyzw ;
+
+out[0].xyzw  = VE_MAD   temp[1].xyzw   temp[1].1111 temp[1].0000 ;
+out[1].xyzw  = VE_MAX  input[1].xyzw  input[1].xyzw ;

src/particle_particle_animated.vs.asm

@@ -0,0 +1,32 @@
+-- const[0-3]  : transform matrix
+-- const[4]    : dx
+-- const[5]    : dy
+-- const[6].xyz: particle_position
+-- const[6].w  : scale
+
+-- input[0]: position coordinate
+-- input[1]: texture coordinate
+
+--
+-- dot(m[0], v), dot(m[1], v), dot(m[2], v), dot(m[3], v)
+--
+
+-- ppos = particle_position
+temp[0].xyz  = VE_ADD  const[6].xyz_  const[6].000_ ;
+
+-- ppos = position.xxx * dx + ppos
+temp[0].xyz  = VE_MAD  input[0].xxx_  const[4].xyz_ temp[0].xyz_ ;
+-- ppos = position.yyy * dy + ppos
+temp[0].xyz  = VE_MAD  input[0].yyy_  const[5].xyz_ temp[0].xyz_ ;
+
+-- ppos *= scale
+temp[0].xyzw = VE_MUL  temp[0].xyz1   const[6].www1 ;
+
+-- ppos = transform_matrix * ppos
+temp[1].x    = VE_DOT  const[0].xyzw  temp[0].xyzw ;
+temp[1].y    = VE_DOT  const[1].xyzw  temp[0].xyzw ;
+temp[1].z    = VE_DOT  const[2].xyzw  temp[0].xyzw ;
+temp[1].w    = VE_DOT  const[3].xyzw  temp[0].xyzw ;
+
+out[0].xyzw  = VE_MAD   temp[1].xyzw   temp[1].1111 temp[1].0000 ;
+out[1].xyzw  = VE_MAX  input[1].xyzw  input[1].xyzw ;

src/particle_particle_animated_fan.vs.asm

@@ -0,0 +1,37 @@
+-- const[0-3]  : transform matrix
+-- const[4]    : dx
+-- const[5]    : dy
+-- const[6].xyz: particle_position
+-- const[6].w  : scale
+-- const[7].x  : -2.0
+
+-- input[0]: texture coordinate
+
+--
+-- dot(m[0], v), dot(m[1], v), dot(m[2], v), dot(m[3], v)
+--
+
+-- calculate position from texture coordinate
+-- x = y * -2 + 1
+-- y = x * -2 + 1
+temp[2].xy   = VE_MAD  input[0].yx__  const[7].xx__  input[0].11__ ;
+
+-- ppos = particle_position
+temp[0].xyz  = VE_ADD  const[6].xyz_  const[6].000_ ;
+
+-- ppos = position.xxx * dx + ppos
+temp[0].xyz  = VE_MAD   temp[2].xxx_  const[4].xyz_  temp[0].xyz_ ;
+-- ppos = position.yyy * dy + ppos
+temp[0].xyz  = VE_MAD   temp[2].yyy_  const[5].xyz_  temp[0].xyz_ ;
+
+-- ppos *= scale
+temp[0].xyzw = VE_MUL  temp[0].xyz1   const[6].www1 ;
+
+-- ppos = transform_matrix * ppos
+temp[1].x    = VE_DOT  const[0].xyzw  temp[0].xyzw ;
+temp[1].y    = VE_DOT  const[1].xyzw  temp[0].xyzw ;
+temp[1].z    = VE_DOT  const[2].xyzw  temp[0].xyzw ;
+temp[1].w    = VE_DOT  const[3].xyzw  temp[0].xyzw ;
+
+out[0].xyzw  = VE_MAD   temp[1].xyzw   temp[1].1111  temp[1].0000 ;
+out[1].xyzw  = VE_MAX  input[0].xy00  input[0].xy00 ;

src/particle_particle_animated_quad_vbuf.vs.asm

@@ -0,0 +1,38 @@
+-- const[0-3]  : transform matrix
+-- const[4]    : dx
+-- const[5]    : dy
+---- const[6].xyz: particle_position
+-- const[6].w  : scale
+-- const[7].x  : -2.0
+
+-- input[0]: particle position
+-- input[1]: texture coordinate
+
+--
+-- dot(m[0], v), dot(m[1], v), dot(m[2], v), dot(m[3], v)
+--
+
+-- calculate position from texture coordinate
+-- x = y * -2 + 1
+-- y = x * -2 + 1
+temp[2].xy   = VE_MAD  input[1].yx__  const[7].xx__  input[1].11__ ;
+
+-- ppos = particle_position
+temp[0].xyz  = VE_ADD  input[0].xyz_  input[0].000_ ;
+
+-- ppos = position.xxx * dx + ppos
+temp[0].xyz  = VE_MAD   temp[2].xxx_  const[4].xyz_   temp[0].xyz_ ;
+-- ppos = position.yyy * dy + ppos
+temp[0].xyz  = VE_MAD   temp[2].yyy_  const[5].xyz_   temp[0].xyz_ ;
+
+-- ppos = vec4(ppos.xyz * scale.xyz, age)
+temp[0].xyzw = VE_MAD  temp[0].xyz0   const[6].www0  input[0].000w ;
+
+-- ppos = transform_matrix * ppos
+temp[1].x    = VE_DOT  const[0].xyzw  temp[0].xyz1 ;
+temp[1].y    = VE_DOT  const[1].xyzw  temp[0].xyz1 ;
+temp[1].z    = VE_DOT  const[2].xyzw  temp[0].xyz1 ;
+temp[1].w    = VE_DOT  const[3].xyzw  temp[0].xyz1 ;
+
+out[0].xyzw  = VE_ADD   temp[1].xyzw   temp[1].0000 ;
+out[1].xyzw  = VE_ADD  input[1].xy00   temp[0].00zw ; -- age

src/particle_particle_colored.fs.asm

@@ -0,0 +1,43 @@
+-- temp[0].xy: texture coordinate
+-- temp[0].a : age
+
+TEX TEX_SEM_WAIT TEX_SEM_ACQUIRE
+  temp[2].rgb = LD tex[0].rgba temp[0].rgaa ;
+
+-- i = vec3(0.25, 0.40625, 0.5625);
+src0.a = float(40)   , -- 0.25
+src1.a = float(45)   , -- 0.40625
+src2.rgb = float(49) : -- 0.5625
+  temp[1].rgb = MAD src0.a10 src1.1a0 src2.00r ;
+
+-- t = i + vec3(td)
+src0.a   = temp[0] ,
+src0.rgb = temp[1] ,
+src1.rgb = temp[0] :
+  temp[1].rgb = MAD src1.bbb src0.aaa src0.rgb ;
+
+-- j = fract(t)
+src0.rgb = temp[1] :
+  temp[1].rgb = FRC src0.rgb ;
+
+-- k = cos(j * 2π)
+src0.rgb = temp[1] :
+              COS src0.r ,
+  temp[1].r = SOP ;
+src0.rgb = temp[1] :
+              COS src0.g ,
+  temp[1].g = SOP ;
+src0.rgb = temp[1] :
+              COS src0.b ,
+  temp[1].b = SOP ;
+
+-- l = k * vec3(0.5, 0.5, 0.5) + vec3(0.5, 0.5, 0.5)
+src0.rgb = temp[1] ,
+src1.rgb = float(48) : -- 0.5
+  temp[1].rgb = MAD src0.rgb src1.rrr src1.rrr ;
+
+OUT TEX_SEM_WAIT
+src0.rgb = temp[2] ,
+src1.rgb = temp[1] :
+  out[0].a    = MAX src0.1 src0.1 ,
+  out[0].rgb  = MAD src0.rgb src1.rgb src0.000 ;

src/particle_physics.fs.asm

@@ -0,0 +1,121 @@
+-- temp[0].rgb : position
+-- temp[0].a   : age
+-- temp[1].rgb : velocity
+-- temp[1].a   : delta
+
+-- temp[2].rgb : reset__position
+-- temp[2].a   : reset__age
+-- temp[3].rgb : reset__velocity
+-- temp[3].a   : reset__delta
+
+-- temp[4].rgb : update__position
+-- temp[4].a   : update__age
+-- temp[5].rgb : update__velocity
+-- temp[5].a   : update__delta
+
+-- temp[6].rgb : temp
+
+-- velocity_scale.rgb = vec3(0.003 , 0.01, 0.003)
+-- delta_age = 0.01
+-- const[0] = { velocity_scale.rgb, delta_age }
+-- gravity = -0.05
+-- velocity_attenuation = -0.7
+-- const[1] = { velocity_attenuation, gravity, max_age, 0 }
+
+-- out[0].rgb  : position
+-- out[0].a    : age
+
+TEX
+  temp[0].rgba = LD tex[0].rgba temp[0].rgaa ;
+
+TEX TEX_SEM_WAIT TEX_SEM_ACQUIRE
+  temp[1].rgba = LD tex[1].rgba temp[0].rgaa ;
+
+-- update_particle (position)
+TEX_SEM_WAIT
+src0.a   = temp[0]  , -- age
+src0.rgb = temp[0]  , -- position
+src1.a   = const[0] , -- delta_age
+src1.rgb = const[0] , -- scale
+src2.rgb = temp[1]  : -- velocity
+  temp[4].a   = MAD src0.a   src0.1   src1.a   , -- update__age      = (age * 1) - delta_age
+  temp[4].rgb = MAD src2.rgb src1.rgb src0.rgb ; -- update__position = (velocity * scale) + position
+
+-- update_particle (velocity gravity)
+-- p.velocity.y += -0.05;
+src0.rgb = temp[1]  , -- velocity
+src1.rgb = const[1] : -- gravity (g)
+  temp[5].rgb = MAD src0.rgb src1.111 src1.0g0 ;
+
+-- update_particle (velocity bounce)
+-- p.velocity.y *= -0.7;
+src0.rgb = temp[5]  , -- velocity <gravity>
+src1.rgb = const[1] : -- velocity_attenuation (r)
+  temp[6].rgb = MAD src0.rgb src1.1r1 src1.000 ;
+
+-- update_particle (velocity bounce)
+-- p.velocity = (p.position.y >= 0) ? temp[5] : temp[6]
+src0.rgb = temp[5] , -- velocity <gravity>
+src1.rgb = temp[6] , -- velocity <bounce>
+src2.rgb = temp[4] : -- position
+  temp[5].rgb = CMP src0.rgb src1.rgb src2.ggg ;
+
+-- position.y = abs(position.y)
+src0.rgb = temp[4] :
+  temp[4].g = MAX |src0.0g0| |src0.0g0| ;
+
+--
+-- reset
+--
+
+-- normalize(vec3(velocity.x, 0, velocity.z))
+src0.rgb = temp[1] : -- velocity
+              DP3 src0.r0b src0.r0b ,
+  temp[2].a = DP ;
+src0.a   = temp[2] :
+  temp[2].a   = RSQ |src0.a| ;
+src0.a   = temp[2] ,
+src0.rgb = temp[1] , -- velocity
+src1.a   = temp[1] : -- delta
+  temp[2].rgb = MAD src0.r0b src0.a0a src1.0a0 ;
+
+-- age = age + max_age
+-- reset__position = reset__position * 20
+src0.a   = temp[0]  , -- age
+src1.rgb = const[1] , -- max_age
+src2.a   = const[1] , -- reset_radius
+src0.rgb = temp[2]  : -- reset__position
+  temp[2].a   = MAD src0.a src0.1 src1.b ,
+  temp[2].rgb = MAD src0.rgb src2.aaa src2.000 ;
+
+-- reset__velocity
+-- (p.velocity.x * 0.5 + 0.5)
+-- velocity.xz = velocity.xz
+src0.rgb = temp[1]   , -- velocity.x
+src1.a   = float(48) : -- 0.5
+  temp[3].a   = MAD src0.r src1.a src1.a ,
+  temp[3].rgb = MAX src0.r0b src0.r0b ;
+
+-- reset__velocity
+-- velocity.y = (temp[3].a * delta + 2.0)
+src0.a   = temp[1]   , -- delta
+src1.a   = float(56) , -- 1.0
+src2.a   = temp[3]   : --
+  temp[3].g = MAD src2.0a0 src0.0a0 src1.0a0 ;
+
+OUT
+src0.a   = temp[4]  , -- update__age
+src1.a   = temp[2]  , -- reset__age
+src2.a   = temp[0]  , -- age
+src0.rgb = temp[4]  , -- update__position
+src1.rgb = temp[2]  : -- reset__position
+  out[0].a    = CMP src0.a src1.a src2.a ,
+  out[0].rgb  = CMP src0.rgb src1.rgb src2.aaa ;
+
+OUT TEX_SEM_WAIT
+src0.a   = temp[1]  , -- delta
+src2.a   = temp[0]  , -- age
+src0.rgb = temp[5]  , -- update__velocity
+src1.rgb = temp[3]  : -- reset__velocity
+  out[1].a    = MAX src0.a src0.a , -- constant
+  out[1].rgb  = CMP src0.rgb src1.rgb src2.aaa ;

src/particle_physics.vs.asm

@@ -0,0 +1,2 @@
+out[0].xyzw = VE_ADD  input[0].xyzw  input[0].0000 ;
+out[1].xyzw = VE_ADD  input[0].xy00  const[0].xy00 ;

src/particle_plane.fs.asm

@@ -0,0 +1,10 @@
+-- temp[0]: texture coordinate
+
+TEX TEX_SEM_WAIT TEX_SEM_ACQUIRE
+  temp[0].rgba = LD tex[0].rgba temp[0].rgaa ;
+
+OUT TEX_SEM_WAIT
+src0.a = temp[0] ,
+src0.rgb = temp[0] :
+  out[0].a    = MAX src0.a src0.a ,
+  out[0].rgb  = MAX src0.rgb src0.rgb ;

src/particle_plane.vs.asm

@@ -0,0 +1,15 @@
+-- const[0-3]: transform matrix
+-- input[0]: position coordinate
+-- input[1]: texture coordinate
+
+--
+-- dot(m[0], v), dot(m[1], v), dot(m[2], v), dot(m[3], v)
+--
+
+temp[1].x   = VE_DOT  const[0].xyzw  input[0].xyzw ;
+temp[1].y   = VE_DOT  const[1].xyzw  input[0].xyzw ;
+temp[1].z   = VE_DOT  const[2].xyzw  input[0].xyzw ;
+temp[1].w   = VE_DOT  const[3].xyzw  input[0].xyzw ;
+
+out[0].xyzw = VE_MAD   temp[1].xyzw   temp[1].1111 temp[1].0000 ;
+out[1].xyzw = VE_MAX  input[1].xyzw  input[1].xyzw ;

src/particle_plane_fan.vs.asm

@@ -0,0 +1,20 @@
+-- const[0-3]: transform matrix
+-- const[4].x: -2.0
+-- input[0]: texture coordinate
+
+--
+-- dot(m[0], v), dot(m[1], v), dot(m[2], v), dot(m[3], v)
+--
+
+-- calculate position from texture coordinate
+-- x = y * -2 + 1
+-- y = x * -2 + 1
+temp[1].xyzw = VE_MAD  input[0].yx00  const[4].xx00  input[0].1101 ;
+
+temp[2].x    = VE_DOT  const[0].xyzw   temp[1].xyzw ;
+temp[2].y    = VE_DOT  const[1].xyzw   temp[1].xyzw ;
+temp[2].z    = VE_DOT  const[2].xyzw   temp[1].xyzw ;
+temp[2].w    = VE_DOT  const[3].xyzw   temp[1].xyzw ;
+
+out[0].xyzw  = VE_MAD   temp[2].xyzw   temp[2].1111 temp[2].0000 ;
+out[1].xyzw  = VE_MAX  input[0].xy00  input[0].xy00 ;

src/r500/command_processor.h

@@ -2,8 +2,10 @@
 
 #define _NOP 0x10
 #define _3D_LOAD_VBPNTR 0x2f
+#define _INDX_BUFFER 0x33
 #define _3D_DRAW_VBUF_2 0x34
 #define _3D_DRAW_IMMD_2 0x35
+#define _3D_DRAW_INDX_2 0x36
 
 #define TYPE_0_COUNT(c) (((c) & 0x3fff) << 16)
 #define TYPE_0_ONE_REG (1 << 15)
@@ -11,3 +13,11 @@
 
 #define TYPE_3_COUNT(c) (((c) & 0x3fff) << 16)
 #define TYPE_3_OPCODE(o) (((o) & 0xff) << 8)
+
+#define INDX_BUFFER__0__ONE_REG_WR(n) (((n) & 1) << 31)
+#define INDX_BUFFER__0__SKIP_COUNT(n) (((n) & 0x7) << 16)
+#define INDX_BUFFER__0__DESTINATION(n) (((n) & 0x1fff) << 0)
+
+#define INDX_BUFFER__1__BUFFER_BASE(n) ((n) << 0)
+
+#define INDX_BUFFER__2__BUFFER_SIZE(n) ((n) << 0)

src/r500/indirect_buffer.c

@@ -9,7 +9,7 @@
 #include "3d_registers_undocumented.h"
 #include "3d_registers_bits.h"
 
-union u32_f32 ib[16384];
+union u32_f32 ib[16384 * 100];
 volatile int ib_ix;
 
 void ib_generic_initialization()
@@ -54,8 +54,6 @@ void ib_generic_initialization()
       , RB3D_CCTL__INDEPENDENT_COLORFORMAT_ENABLE(1)
       );
   T0V(RB3D_ROPCNTL, 0x00000000);
-  T0V(RB3D_BLENDCNTL, 0x00000000);
-  T0V(RB3D_ABLENDCNTL, 0x00000000);
   T0V(RB3D_COLOR_CHANNEL_MASK
       , RB3D_COLOR_CHANNEL_MASK__BLUE_MASK(1)
       | RB3D_COLOR_CHANNEL_MASK__GREEN_MASK(1)
@@ -136,11 +134,6 @@ void ib_generic_initialization()
       | GA_ROUND_MODE__GEOMETRY_MASK(0)
       );
 
-  T0Vf(GA_POINT_S0, 0.0f);
-  T0Vf(GA_POINT_T0, 1.0f);
-  T0Vf(GA_POINT_S1, 1.0f);
-  T0Vf(GA_POINT_T1, 0.0f);
-
   T0V(GA_COLOR_CONTROL
       , GA_COLOR_CONTROL__RGB0_SHADING(2)
       | GA_COLOR_CONTROL__ALPHA0_SHADING(2)
@@ -160,7 +153,6 @@ void ib_generic_initialization()
       );
 
   T0V(GB_SELECT, 0x00000000);
-  T0V(GB_ENABLE, 0x00000000);
 
   T0V(GB_MSPOS0
       , GB_MSPOS0__MS_X0(6)
@@ -278,33 +270,10 @@ void ib_generic_initialization()
   T0V(US_W_FMT
       , US_W_FMT__W_FMT(0) // W is always zero
       );
-
-  //////////////////////////////////////////////////////////////////////////////
-  // SC
-  //////////////////////////////////////////////////////////////////////////////
-
-  T0V(SC_SCISSOR0
-      , SC_SCISSOR0__XS0(0)
-      | SC_SCISSOR0__YS0(0)
-      );
-  T0V(SC_SCISSOR1
-      , SC_SCISSOR1__XS1(1600 - 1)
-      | SC_SCISSOR1__YS1(1200 - 1)
-      );
-
-  //////////////////////////////////////////////////////////////////////////////
-  // VAP
-  //////////////////////////////////////////////////////////////////////////////
-
-  T0Vf(VAP_VPORT_XSCALE,   800.0f);
-  T0Vf(VAP_VPORT_XOFFSET,  800.0f);
-  T0Vf(VAP_VPORT_YSCALE,  -600.0f);
-  T0Vf(VAP_VPORT_YOFFSET,  600.0f);
-  T0Vf(VAP_VPORT_ZSCALE,     0.5f);
-  T0Vf(VAP_VPORT_ZOFFSET,    0.5f);
 }
 
-void ib_colorbuffer(int reloc_index)
+void ib_colorbuffer(int reloc_index, int pitch,
+                    int macrotile, int microtile)
 {
 
   //////////////////////////////////////////////////////////////////////////////
@@ -318,8 +287,10 @@ void ib_colorbuffer(int reloc_index)
   TU(reloc_index * 4); // index into relocs array
 
   T0V(RB3D_COLORPITCH0
-      , RB3D_COLORPITCH__COLORPITCH(1600 >> 1)
-      | RB3D_COLORPITCH__COLORFORMAT(6) // ARGB8888
+      , RB3D_COLORPITCH__COLORPITCH(pitch >> 1)
+      | RB3D_COLORPITCH__COLORTILE(macrotile)
+      | RB3D_COLORPITCH__COLORMICROTILE(microtile)
+      | RB3D_COLORPITCH__COLORFORMAT__ARGB8888
       );
   // The COLORPITCH NOP is ignored/not applied due to
   // RADEON_CS_KEEP_TILING_FLAGS, but is still required.
@@ -327,7 +298,102 @@ void ib_colorbuffer(int reloc_index)
   TU(reloc_index * 4); // index into relocs array
 }
 
-void ib_zbuffer(int reloc_index, int zfunc)
+void ib_colorbuffer2(int buffer_index,
+                     int reloc_index,
+                     int pitch,
+                     int macrotile, int microtile,
+                     int colorformat)
+{
+  assert(buffer_index >= 0 && buffer_index <= 3);
+
+  int reg_offset = buffer_index * 4;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // CB
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(RB3D_COLOROFFSET0 + reg_offset
+      , 0x00000000 // value replaced by kernel from relocs
+      );
+  T3(_NOP, 0);
+  TU(reloc_index * 4); // index into relocs array
+
+  T0V(RB3D_COLORPITCH0 + reg_offset
+      , RB3D_COLORPITCH__COLORPITCH(pitch >> 1)
+      | RB3D_COLORPITCH__COLORTILE(macrotile)
+      | RB3D_COLORPITCH__COLORMICROTILE(microtile)
+      | RB3D_COLORPITCH__COLORFORMAT(colorformat)
+      );
+  // The COLORPITCH NOP is ignored/not applied due to
+  // RADEON_CS_KEEP_TILING_FLAGS, but is still required.
+  T3(_NOP, 0);
+  TU(reloc_index * 4); // index into relocs array
+}
+
+void ib_colorbuffer3(int buffer_index,
+                     int reloc_index,
+                     int offset,
+                     int pitch,
+                     int macrotile, int microtile,
+                     int colorformat)
+{
+  assert(buffer_index >= 0 && buffer_index <= 3);
+
+  int reg_offset = buffer_index * 4;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // CB
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(RB3D_COLOROFFSET0 + reg_offset
+      , offset // value replaced by kernel from relocs
+      );
+  T3(_NOP, 0);
+  TU(reloc_index * 4); // index into relocs array
+
+  T0V(RB3D_COLORPITCH0 + reg_offset
+      , RB3D_COLORPITCH__COLORPITCH(pitch >> 1)
+      | RB3D_COLORPITCH__COLORTILE(macrotile)
+      | RB3D_COLORPITCH__COLORMICROTILE(microtile)
+      | RB3D_COLORPITCH__COLORFORMAT(colorformat)
+      );
+  // The COLORPITCH NOP is ignored/not applied due to
+  // RADEON_CS_KEEP_TILING_FLAGS, but is still required.
+  T3(_NOP, 0);
+  TU(reloc_index * 4); // index into relocs array
+}
+
+void ib_viewport(int width, int height)
+{
+  //////////////////////////////////////////////////////////////////////////////
+  // SC
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(SC_SCISSOR0
+      , SC_SCISSOR0__XS0(0)
+      | SC_SCISSOR0__YS0(0)
+      );
+  T0V(SC_SCISSOR1
+      , SC_SCISSOR1__XS1(width - 1)
+      | SC_SCISSOR1__YS1(height - 1)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  float x = ((float)width) * 0.5f;
+  float y = ((float)height) * 0.5f;
+
+  T0Vf(VAP_VPORT_XSCALE,   x);
+  T0Vf(VAP_VPORT_XOFFSET,  x);
+  T0Vf(VAP_VPORT_YSCALE,  -y);
+  T0Vf(VAP_VPORT_YOFFSET,  y);
+  T0Vf(VAP_VPORT_ZSCALE,   0.5f);
+  T0Vf(VAP_VPORT_ZOFFSET,  0.5f);
+}
+
+void ib_zbuffer(int reloc_index, int pitch, int zfunc)
 {
   //////////////////////////////////////////////////////////////////////////////
   // ZB
@@ -349,7 +415,7 @@ void ib_zbuffer(int reloc_index, int zfunc)
   TU(reloc_index * 4); // index into relocs array
 
   T0V(ZB_DEPTHPITCH
-      , ZB_DEPTHPITCH__DEPTHPITCH(1600 >> 2)
+      , ZB_DEPTHPITCH__DEPTHPITCH(pitch >> 2)
       | ZB_DEPTHPITCH__DEPTHMACROTILE(1)
       | ZB_DEPTHPITCH__DEPTHMICROTILE(1)
       );
@@ -518,7 +584,10 @@ void ib_texture__0()
   T0V(TX_ENABLE, 0x00000000);
 }
 
-void ib_texture__1(int reloc_index)
+void ib_texture__1(int reloc_index,
+                   int width, int height,
+                   int macrotile, int microtile,
+                   int clamp)
 {
   //////////////////////////////////////////////////////////////////////////////
   // TX
@@ -528,17 +597,21 @@ void ib_texture__1(int reloc_index)
 
   T0V(TX_ENABLE
       , TX_ENABLE__TEX_0_ENABLE__ENABLE);
+
   T0V(TX_FILTER0_0
-      , TX_FILTER0__MAG_FILTER__LINEAR
+      , TX_FILTER0__CLAMP_S(clamp)
+      | TX_FILTER0__CLAMP_T(clamp)
+      | TX_FILTER0__MAG_FILTER__LINEAR
       | TX_FILTER0__MIN_FILTER__LINEAR
       );
   T0V(TX_FILTER1_0
       , TX_FILTER1__LOD_BIAS(1)
+      | TX_FILTER1__BORDER_FIX(1)
       );
   T0V(TX_BORDER_COLOR_0, 0);
   T0V(TX_FORMAT0_0
-      , TX_FORMAT0__TXWIDTH(1024 - 1)
-      | TX_FORMAT0__TXHEIGHT(1024 - 1)
+      , TX_FORMAT0__TXWIDTH(width - 1)
+      | TX_FORMAT0__TXHEIGHT(height - 1)
       );
 
   T0V(TX_FORMAT1_0
@@ -552,10 +625,110 @@ void ib_texture__1(int reloc_index)
   T0V(TX_FORMAT2_0, 0);
 
   T0V(TX_OFFSET_0
-      //, TX_OFFSET__MACRO_TILE(1)
-      //| TX_OFFSET__MICRO_TILE(1)
+      , TX_OFFSET__MACRO_TILE(macrotile)
+      | TX_OFFSET__MICRO_TILE(microtile)
+      );
+
+  T3(_NOP, 0);
+  TU(reloc_index * 4); // index into relocs array
+}
+
+void ib_texture__1_float32(int reloc_index,
+                           int width, int height,
+                           int macrotile, int microtile,
+                           int clamp)
+{
+  //////////////////////////////////////////////////////////////////////////////
+  // TX
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(TX_INVALTAGS, 0x00000000);
+
+  T0V(TX_ENABLE
+      , TX_ENABLE__TEX_0_ENABLE__ENABLE);
+
+  T0V(TX_FILTER0_0
+      , TX_FILTER0__CLAMP_S(clamp)
+      | TX_FILTER0__CLAMP_T(clamp)
+      | TX_FILTER0__MAG_FILTER__POINT
+      | TX_FILTER0__MIN_FILTER__POINT
+      | TX_FILTER0__ID(0)
+      );
+  T0V(TX_FILTER1_0
+      , TX_FILTER1__LOD_BIAS(1)
+      | TX_FILTER1__BORDER_FIX(1)
+      );
+  T0V(TX_BORDER_COLOR_0, 0);
+  T0V(TX_FORMAT0_0
+      , TX_FORMAT0__TXWIDTH(width - 1)
+      | TX_FORMAT0__TXHEIGHT(height - 1)
+      );
+
+  T0V(TX_FORMAT1_0
+      , TX_FORMAT1__TXFORMAT__TX_FMT_32F_32F_32F_32F
+      | TX_FORMAT1__SEL_ALPHA(3)
+      | TX_FORMAT1__SEL_RED(0)
+      | TX_FORMAT1__SEL_GREEN(1)
+      | TX_FORMAT1__SEL_BLUE(2)
+      | TX_FORMAT1__TEX_COORD_TYPE__2D
+      );
+  T0V(TX_FORMAT2_0, 0);
+
+  T0V(TX_OFFSET_0
+      , TX_OFFSET__MACRO_TILE(macrotile)
+      | TX_OFFSET__MICRO_TILE(microtile)
+      );
+
+  T3(_NOP, 0);
+  TU(reloc_index * 4); // index into relocs array
+}
+
+void ib_texture2(int texture_index,
+                 int reloc_index,
+                 int width, int height,
+                 int macrotile, int microtile,
+                 int clamp,
+                 int txformat)
+{
+  assert(texture_index >= 0 && texture_index <= 15);
+
+  int texture_offset = texture_index * 4;
+
+  T0V(TX_FILTER0_0 + texture_offset
+      , TX_FILTER0__CLAMP_S(clamp)
+      | TX_FILTER0__CLAMP_T(clamp)
+      | TX_FILTER0__MAG_FILTER__POINT
+      | TX_FILTER0__MIN_FILTER__POINT
+      | TX_FILTER0__ID(texture_index)
+      );
+  T0V(TX_FILTER1_0 + texture_offset
+      , TX_FILTER1__LOD_BIAS(1)
+      | TX_FILTER1__BORDER_FIX(0)
+      );
+  T0V(TX_BORDER_COLOR_0 + texture_offset
       , 0
       );
+  T0V(TX_FORMAT0_0 + texture_offset
+      , TX_FORMAT0__TXWIDTH(width - 1)
+      | TX_FORMAT0__TXHEIGHT(height - 1)
+      );
+
+  T0V(TX_FORMAT1_0 + texture_offset
+      , TX_FORMAT1__TXFORMAT(txformat)
+      | TX_FORMAT1__SEL_ALPHA(3)
+      | TX_FORMAT1__SEL_RED(0)
+      | TX_FORMAT1__SEL_GREEN(1)
+      | TX_FORMAT1__SEL_BLUE(2)
+      | TX_FORMAT1__TEX_COORD_TYPE__2D
+      );
+  T0V(TX_FORMAT2_0 + texture_offset
+      , 0
+      );
+
+  T0V(TX_OFFSET_0 + texture_offset
+      , TX_OFFSET__MACRO_TILE(macrotile)
+      | TX_OFFSET__MICRO_TILE(microtile)
+      );
 
   T3(_NOP, 0);
   TU(reloc_index * 4); // index into relocs array
@@ -617,7 +790,7 @@ void ib_vap_pvs_const_cntl(const float * consts, int size)
 
   T0V(VAP_PVS_CONST_CNTL
       , VAP_PVS_CONST_CNTL__PVS_CONST_BASE_OFFSET(0)
-      | VAP_PVS_CONST_CNTL__PVS_MAX_CONST_ADDR((size / 4) - 1)
+      | VAP_PVS_CONST_CNTL__PVS_MAX_CONST_ADDR(consts_length - 1)
       );
 
   T0V(VAP_PVS_VECTOR_INDX_REG
@@ -629,6 +802,34 @@ void ib_vap_pvs_const_cntl(const float * consts, int size)
     TF(consts[i]);
 }
 
+void ib_vap_pvs_const_offset(const float * consts, int size, int offset)
+{
+  assert(size % 16 == 0);
+
+  const int consts_length = size / 4;
+
+  T0V(VAP_PVS_VECTOR_INDX_REG
+      , VAP_PVS_VECTOR_INDX_REG__OCTWORD_OFFSET(1024 + offset)
+      );
+
+  T0_ONE_REG(VAP_PVS_VECTOR_DATA_REG_128, (consts_length - 1));
+  for (int i = 0; i < consts_length; i++)
+    TF(consts[i]);
+}
+
+void ib_ga_consts(const float * consts, int consts_length, int index)
+{
+  assert(consts_length % 4 == 0);
+
+  T0V(GA_US_VECTOR_INDEX
+      , GA_US_VECTOR_INDEX__INDEX(index)
+      | GA_US_VECTOR_INDEX__TYPE(1)
+      );
+  T0_ONE_REG(GA_US_VECTOR_DATA, (consts_length - 1));
+  for (int i = 0; i < consts_length; i++)
+    TF(consts[i]);
+}
+
 void ib_vap_stream_cntl__2()
 {
   //////////////////////////////////////////////////////////////////////////////
@@ -671,6 +872,66 @@ void ib_vap_stream_cntl__3()
       );
 }
 
+void ib_vap_stream_cntl__32()
+{
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP_PROG_STREAM_CNTL
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_PROG_STREAM_CNTL_0
+      , VAP_PROG_STREAM_CNTL__DATA_TYPE_0__FLOAT_3
+      | VAP_PROG_STREAM_CNTL__SKIP_DWORDS_0(0)
+      | VAP_PROG_STREAM_CNTL__DST_VEC_LOC_0(0)
+      | VAP_PROG_STREAM_CNTL__LAST_VEC_0(0)
+      | VAP_PROG_STREAM_CNTL__DATA_TYPE_1__FLOAT_2
+      | VAP_PROG_STREAM_CNTL__SKIP_DWORDS_1(0)
+      | VAP_PROG_STREAM_CNTL__DST_VEC_LOC_1(1)
+      | VAP_PROG_STREAM_CNTL__LAST_VEC_1(1)
+      );
+  T0V(VAP_PROG_STREAM_CNTL_EXT_0
+      , VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_X_0__SELECT_X
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_Y_0__SELECT_Y
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_Z_0__SELECT_Z
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_W_0__SELECT_FP_ONE
+      | VAP_PROG_STREAM_CNTL_EXT__WRITE_ENA_0(0b1111) // XYZW
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_X_1__SELECT_X
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_Y_1__SELECT_Y
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_Z_1__SELECT_FP_ZERO
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_W_1__SELECT_FP_ONE
+      | VAP_PROG_STREAM_CNTL_EXT__WRITE_ENA_1(0b1111) // XYZW
+      );
+}
+
+void ib_vap_stream_cntl__42()
+{
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP_PROG_STREAM_CNTL
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_PROG_STREAM_CNTL_0
+      , VAP_PROG_STREAM_CNTL__DATA_TYPE_0__FLOAT_4
+      | VAP_PROG_STREAM_CNTL__SKIP_DWORDS_0(0)
+      | VAP_PROG_STREAM_CNTL__DST_VEC_LOC_0(0)
+      | VAP_PROG_STREAM_CNTL__LAST_VEC_0(0)
+      | VAP_PROG_STREAM_CNTL__DATA_TYPE_1__FLOAT_2
+      | VAP_PROG_STREAM_CNTL__SKIP_DWORDS_1(0)
+      | VAP_PROG_STREAM_CNTL__DST_VEC_LOC_1(1)
+      | VAP_PROG_STREAM_CNTL__LAST_VEC_1(1)
+      );
+  T0V(VAP_PROG_STREAM_CNTL_EXT_0
+      , VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_X_0__SELECT_X
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_Y_0__SELECT_Y
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_Z_0__SELECT_Z
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_W_0__SELECT_W
+      | VAP_PROG_STREAM_CNTL_EXT__WRITE_ENA_0(0b1111) // XYZW
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_X_1__SELECT_X
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_Y_1__SELECT_Y
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_Z_1__SELECT_FP_ZERO
+      | VAP_PROG_STREAM_CNTL_EXT__SWIZZLE_SELECT_W_1__SELECT_FP_ONE
+      | VAP_PROG_STREAM_CNTL_EXT__WRITE_ENA_1(0b1111) // XYZW
+      );
+}
+
 void ib_vap_stream_cntl__323()
 {
   //////////////////////////////////////////////////////////////////////////////

src/r500/indirect_buffer.h

@@ -7,23 +7,23 @@
 
 #define T0(address, count)                                              \
   do {                                                                  \
-    ib[ib_ix++].u32 = TYPE_0_COUNT(count) | TYPE_0_BASE_INDEX(address >> 2);   \
+    ib[ib_ix++].u32 = TYPE_0_COUNT(count) | TYPE_0_BASE_INDEX((address) >> 2); \
   } while (0);
 
 #define T0_ONE_REG(address, count)                                      \
   do {                                                                  \
-    ib[ib_ix++].u32 = TYPE_0_COUNT(count) | TYPE_0_ONE_REG | TYPE_0_BASE_INDEX(address >> 2); \
+    ib[ib_ix++].u32 = TYPE_0_COUNT(count) | TYPE_0_ONE_REG | TYPE_0_BASE_INDEX((address) >> 2); \
   } while (0);
 
 #define T0V(address, value)                                             \
   do {                                                                  \
-    ib[ib_ix++].u32 = TYPE_0_COUNT(0) | TYPE_0_BASE_INDEX(address >> 2);   \
+    ib[ib_ix++].u32 = TYPE_0_COUNT(0) | TYPE_0_BASE_INDEX((address) >> 2); \
     ib[ib_ix++].u32 = value;                                               \
   } while (0);
 
 #define T0Vf(address, value)                                            \
   do {                                                                  \
-    ib[ib_ix++].u32 = TYPE_0_COUNT(0) | TYPE_0_BASE_INDEX(address >> 2);   \
+    ib[ib_ix++].u32 = TYPE_0_COUNT(0) | TYPE_0_BASE_INDEX((address) >> 2); \
     ib[ib_ix++].f32 = value;                                               \
   } while (0);
 
@@ -51,20 +51,49 @@ union u32_f32 {
   float f32;
 };
 
-extern union u32_f32 ib[16384];
+extern union u32_f32 ib[16384 * 100];
 extern volatile int ib_ix;
 
 void ib_generic_initialization();
-void ib_colorbuffer(int reloc_index);
-void ib_zbuffer(int reloc_index, int zfunc);
+void ib_viewport(int width, int height);
+void ib_colorbuffer(int reloc_index, int pitch, int macrotile, int microtile);
+void ib_colorbuffer2(int buffer_index,
+                     int reloc_index,
+                     int pitch,
+                     int macrotile, int microtile,
+                     int colorformat);
+void ib_colorbuffer3(int buffer_index,
+                     int reloc_index,
+                     int offset,
+                     int pitch,
+                     int macrotile, int microtile,
+                     int colorformat);
+void ib_zbuffer(int reloc_index, int pitch, int zfunc);
 void ib_rs_instructions(int count);
 void ib_texture__0();
-void ib_texture__1(int reloc_index);
+void ib_texture__1(int reloc_index,
+                   int width, int height,
+                   int macrotile, int microtile,
+                   int clamp);
+void ib_texture__1_float32(int reloc_index,
+                           int width, int height,
+                           int macrotile, int microtile,
+                           int clamp);
+void ib_texture2(int texture_index,
+                 int reloc_index,
+                 int width, int height,
+                 int macrotile, int microtile,
+                 int clamp,
+                 int txformat);
 void ib_vap_pvs(struct shader_offset * offset);
 void ib_ga_us(struct shader_offset * offset);
 void ib_vap_pvs_const_cntl(const float * consts, int size);
+void ib_vap_pvs_const_offset(const float * consts, int size, int offset);
+void ib_ga_consts(const float * consts, int consts_length, int index);
 void ib_vap_stream_cntl__2();
 void ib_vap_stream_cntl__3();
+void ib_vap_stream_cntl__32();
+void ib_vap_stream_cntl__42();
 void ib_vap_stream_cntl__323();
 
 #ifdef __cplusplus

src/texture_tile.fs.asm

@@ -0,0 +1,8 @@
+TEX TEX_SEM_WAIT TEX_SEM_ACQUIRE
+  temp[0].rgba = LD tex[0].rgba temp[0].rgaa ;
+
+OUT TEX_SEM_WAIT
+src0.a = temp[0],
+src0.rgb = temp[0] :
+  out[0].a    = MAX src0.a src0.a ,
+  out[0].rgb  = MAX src0.rgb src0.rgb ;

src/texture_tile.vs.asm

@@ -0,0 +1 @@
+out[0].xyzw = VE_ADD  input[0].xyzw  input[0].0000 ;

src/tx_rt.cpp

@@ -0,0 +1,293 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+#include "../model/plane.h"
+
+#define TEXTURE_TILE_SHADER 0
+
+const char * vertex_shader_paths[] = {
+  "tx_rt.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "tx_rt.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+#define PARTICLE_TEXTURE 0
+
+const char * textures[] = {
+  "../texture/butterfly_1024x1024_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+int _tile_texture(const shaders& shaders,
+                  int input_reloc_index,
+                  int output_reloc_index)
+{
+  int viewport_width = 1600;
+  int viewport_height = 1200;
+  int texture_width = 1024;
+  int texture_height = 1024;
+  float vx = ((float)viewport_width) * 0.5f;
+  float vy = ((float)viewport_height) * 0.5f;
+  float tx = 0.5f / ((float)texture_width);
+  float ty = 0.5f / ((float)texture_height);
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  T0V(SC_SCISSOR0
+      , SC_SCISSOR0__XS0(0)
+      | SC_SCISSOR0__YS0(0)
+      );
+  T0V(SC_SCISSOR1
+      , SC_SCISSOR1__XS1(viewport_width - 1)
+      | SC_SCISSOR1__YS1(viewport_height - 1)
+      );
+  T0Vf(VAP_VPORT_XSCALE, (float)viewport_width);
+  T0Vf(VAP_VPORT_YSCALE, (float)viewport_height);
+
+  ib_colorbuffer(output_reloc_index, viewport_width, 0, 0); // macrotile, microtile
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(0)  // C4_8
+      | US_OUT_FMT__C0_SEL__BLUE
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__RED
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_3
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  // shaders
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  // GA
+
+  T0V(GB_ENABLE
+      , 0
+    );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // RS
+  //////////////////////////////////////////////////////////////////////////////
+
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1)
+      );
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4)
+      );
+
+  //
+
+  T0V(ZB_CNTL, 0);
+  T0V(ZB_ZSTENCILCNTL, 0);
+
+  //
+
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 2; // clamp to [0.0, 1.0]
+  ib_texture__1(input_reloc_index,
+                texture_width, texture_height,
+                macrotile, microtile,
+                clamp);
+
+  ib_vap_stream_cntl__2();
+
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+
+  ib_ga_us(&shaders.fragment[TEXTURE_TILE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[TEXTURE_TILE_SHADER]);
+
+  // fragment constants
+  const float fragment_consts[] = {
+    tx, ty, 0, 0,
+  };
+  int fragment_consts_length = (sizeof (fragment_consts)) / (sizeof (fragment_consts[0]));
+
+  T0V(GA_US_VECTOR_INDEX
+      , GA_US_VECTOR_INDEX__INDEX(0)
+      | GA_US_VECTOR_INDEX__TYPE(1)
+      );
+  T0_ONE_REG(GA_US_VECTOR_DATA, (fragment_consts_length - 1));
+  for (int i = 0; i < fragment_consts_length; i++)
+    TF(fragment_consts[i]);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // GA POINT SIZE
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(GA_POINT_SIZE
+      , GA_POINT_SIZE__HEIGHT((int)(vy * 12.0f))
+      | GA_POINT_SIZE__WIDTH((int)(vx * 12.0f))
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float vertices[] = {
+    0.0f, 0.0f,
+    1.0f, 0.0f,
+    1.0f, 1.0f,
+    0.0f, 1.0f,
+  };
+  const int vertex_count = 4;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(5) // triangle fan
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < vertex_count * 2; i++) {
+    TF(vertices[i]);
+  }
+
+  return ib_ix;
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  int flush_handle;
+
+  void * colorbuffer_ptr[2];
+  void * zbuffer_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  int colorbuffer_ix = 0;
+  float theta = PI * 0.5;
+
+  while (true) {
+    //int ib_dwords = indirect_buffer(shaders, theta);
+    int ib_dwords = _tile_texture(shaders,
+                                  TEXTUREBUFFER_RELOC_INDEX, // input
+                                  COLORBUFFER_RELOC_INDEX);  // output
+
+    int ret = drm_radeon_cs(fd,
+                            colorbuffer_handle[colorbuffer_ix],
+                            zbuffer_handle,
+                            flush_handle,
+                            texturebuffer_handle,
+                            textures_length,
+                            ib_dwords);
+    if (ret == -1)
+      break;
+
+    primary_surface_address(rmmio, colorbuffer_ix);
+
+    // next state
+    theta += 0.01f;
+    colorbuffer_ix = (colorbuffer_ix + 1) & 1;
+
+    break;
+  }
+
+  close(fd);
+}

src/tx_rt.fs.asm

@@ -0,0 +1,8 @@
+TEX TEX_SEM_WAIT TEX_SEM_ACQUIRE
+  temp[0].rgba = LD tex[0].rgba temp[0].rgaa ;
+
+OUT TEX_SEM_WAIT
+src0.a = temp[0],
+src0.rgb = temp[0] :
+  out[0].a    = MAX src0.a src0.a ,
+  out[0].rgb  = MAX src0.rgb src0.rgb ;

src/tx_rt.vs.asm

@@ -0,0 +1,2 @@
+out[0].xyzw = VE_ADD  input[0].xyzw  input[0].0000 ;
+out[1].xyzw = VE_ADD  input[0].xyzw  input[0].0000 ;

src/tx_rt_float.cpp

@@ -0,0 +1,335 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+#include "../model/plane.h"
+
+#define TEXTURE_TILE_SHADER 0
+
+const char * vertex_shader_paths[] = {
+  "tx_rt_float.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "tx_rt_float.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+#define PARTICLE_TEXTURE 0
+
+const char * textures[] = {
+  "../texture/butterfly_1024x1024_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+const int floatbuffer_width = 64;
+const int floatbuffer_height = 64;
+
+int _floatbuffer(const shaders& shaders,
+                 int input_reloc_index,
+                 int output_reloc_index)
+{
+  int viewport_width = floatbuffer_width;
+  int viewport_height = floatbuffer_height;
+  int texture_width = floatbuffer_width;
+  int texture_height = floatbuffer_height;
+  float vx = ((float)viewport_width) * 0.5f;
+  float vy = ((float)viewport_height) * 0.5f;
+  float tx = 0.5f / ((float)texture_width);
+  float ty = 0.5f / ((float)texture_height);
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  printf("vp %d %d \n", viewport_width, viewport_height);
+
+  T0V(SC_SCISSOR0
+      , SC_SCISSOR0__XS0(0)
+      | SC_SCISSOR0__YS0(0)
+      );
+  T0V(SC_SCISSOR1
+      , SC_SCISSOR1__XS1(viewport_width - 1)
+      | SC_SCISSOR1__YS1(viewport_height - 1)
+      );
+  T0Vf(VAP_VPORT_XSCALE, (float)viewport_width);
+  T0Vf(VAP_VPORT_YSCALE, (float)viewport_height);
+
+  ib_colorbuffer(output_reloc_index, viewport_width, 0, 0); // macrotile, microtile
+
+  T0V(RB3D_COLORPITCH0
+      , RB3D_COLORPITCH__COLORPITCH(viewport_width >> 1)
+      | RB3D_COLORPITCH__COLORTILE(0)
+      | RB3D_COLORPITCH__COLORMICROTILE(0)
+      | RB3D_COLORPITCH__COLORFORMAT__ARGB32323232
+      );
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(21)  // C4_32_FP
+      | US_OUT_FMT__C0_SEL__RED
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__BLUE
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_3
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  // shaders
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  // GA
+
+  T0V(GB_ENABLE
+      , 0
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // RS
+  //////////////////////////////////////////////////////////////////////////////
+
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1)
+      );
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4)
+      );
+
+  //
+
+  T0V(ZB_CNTL, 0);
+  T0V(ZB_ZSTENCILCNTL, 0);
+
+  //
+
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 2; // clamp to [0.0, 1.0]
+  ib_texture__1_float32(input_reloc_index,
+                        texture_width, texture_height,
+                        macrotile, microtile,
+                        clamp);
+
+  ib_vap_stream_cntl__2();
+
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+
+  ib_ga_us(&shaders.fragment[TEXTURE_TILE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[TEXTURE_TILE_SHADER]);
+
+  const float vertex_consts[] = {
+    tx, ty, 0, 0,
+  };
+  const int vertex_consts_size = (sizeof (vertex_consts));
+  ib_vap_pvs_const_cntl(vertex_consts, vertex_consts_size);
+
+  // fragment constants
+
+  const float fragment_consts[] = {
+    1234.0f, 0, 0, 0,
+  };
+  int fragment_consts_length = (sizeof (fragment_consts)) / (sizeof (fragment_consts[0]));
+  ib_ga_consts(fragment_consts, fragment_consts_length, 0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float vertices[] = {
+    0.0f, 0.0f,
+    1.0f, 0.0f,
+    1.0f, 1.0f,
+    0.0f, 1.0f,
+  };
+  const int vertex_count = 4;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(5) // triangle fan
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < vertex_count * 2; i++) {
+    TF(vertices[i]);
+  }
+
+  return ib_ix;
+}
+
+void floatbuffer_data(void * ptr, int size)
+{
+  float * f32 = (float*)ptr;
+  int vector_length = size / (4 * 4);
+
+  int offset = 16384;
+  for (int i = 0; i < vector_length; i++) {
+    f32[i * 4 + 0] = offset--;
+    f32[i * 4 + 1] = offset--;
+    f32[i * 4 + 2] = offset--;
+    f32[i * 4 + 3] = offset--;
+  }
+}
+
+void floatbuffer_compare(void * a, void * b, int size)
+{
+  float * a_f32 = (float*)a;
+  float * b_f32 = (float*)b;
+  int vector_length = size / (4 * 4);
+
+  int offset = 1;
+  for (int i = 0; i < vector_length; i++) {
+    if (   (a_f32[i * 4 + 0] != b_f32[i * 4 + 0])
+        || (a_f32[i * 4 + 1] != b_f32[i * 4 + 1])
+        || (a_f32[i * 4 + 2] != b_f32[i * 4 + 2])
+        || (a_f32[i * 4 + 3] != b_f32[i * 4 + 3])) {
+      printf("a[%d] = [% 2.02f  % 2.02f  % 2.02f  % 2.02f] ; ", i,
+             a_f32[i * 4 + 0], a_f32[i * 4 + 1], a_f32[i * 4 + 2], a_f32[i * 4 + 3]);
+      printf("b[%d] = [% 2.02f  % 2.02f  % 2.02f  % 2.02f] \n", i,
+             b_f32[i * 4 + 0], b_f32[i * 4 + 1], b_f32[i * 4 + 2], b_f32[i * 4 + 3]);
+    }
+  }
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  const int floatbuffer_size = floatbuffer_width * floatbuffer_height * 4 * 4;
+  const int floatbuffer_count = 2;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  int flush_handle;
+  int floatbuffer_handle[floatbuffer_count];
+
+  void * colorbuffer_ptr[2];
+  void * floatbuffer_ptr[floatbuffer_count];
+  void * zbuffer_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  floatbuffer_handle[0] = create_buffer(fd, floatbuffer_size, &floatbuffer_ptr[0]);
+  floatbuffer_handle[1] = create_buffer(fd, floatbuffer_size, &floatbuffer_ptr[1]);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "floatbuffer handle[0] %d\n", floatbuffer_handle[0]);
+  fprintf(stderr, "floatbuffer handle[1] %d\n", floatbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  int colorbuffer_ix = 0;
+
+  floatbuffer_data(floatbuffer_ptr[0], floatbuffer_size);
+
+  while (true) {
+    int ib_dwords = _floatbuffer(shaders,
+                                 TEXTUREBUFFER_RELOC_INDEX + 0, // input
+                                 COLORBUFFER_RELOC_INDEX);      // output
+
+    int ret = drm_radeon_cs(fd,
+                            floatbuffer_handle[1],
+                            zbuffer_handle,
+                            flush_handle,
+                            floatbuffer_handle,
+                            floatbuffer_count,
+                            ib_dwords);
+    if (ret == -1)
+      break;
+
+    break;
+  }
+
+  floatbuffer_compare(floatbuffer_ptr[0], floatbuffer_ptr[1], floatbuffer_size);
+
+  close(fd);
+}

src/tx_rt_float.fs.asm

@@ -0,0 +1,8 @@
+TEX TEX_SEM_WAIT TEX_SEM_ACQUIRE
+  temp[0].rgba = LD tex[0].rgba temp[0].rgaa ;
+
+OUT TEX_SEM_WAIT
+src0.a = temp[0],
+src0.rgb = temp[0] :
+  out[0].a    = MAX src0.a src0.a ,
+  out[0].rgb  = MAX src0.rgb src0.rgb ;

src/tx_rt_float.vs.asm

@@ -0,0 +1,2 @@
+out[0].xyzw = VE_ADD  input[0].xyzw  input[0].0000 ;
+out[1].xyzw = VE_ADD  input[0].xy00  const[0].0000 ;

src/tx_rt_float_4x.cpp

@@ -0,0 +1,343 @@
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <fcntl.h>
+#include <sys/mman.h>
+
+#include "r500/3d_registers.h"
+#include "r500/3d_registers_undocumented.h"
+#include "r500/3d_registers_bits.h"
+#include "r500/indirect_buffer.h"
+#include "r500/shader.h"
+#include "r500/display_controller.h"
+
+#include "drm/buffer.h"
+#include "drm/drm.h"
+
+#include "math/float_types.hpp"
+#include "math/transform.hpp"
+#include "math/constants.hpp"
+
+#include "../model/model2.h"
+#include "../model/plane.h"
+
+#define TEXTURE_TILE_SHADER 0
+
+const char * vertex_shader_paths[] = {
+  "tx_rt_float.vs.bin",
+};
+const int vertex_shader_paths_length = (sizeof (vertex_shader_paths)) / (sizeof (vertex_shader_paths[0]));
+const char * fragment_shader_paths[] = {
+  "tx_rt_float.fs.bin",
+};
+const int fragment_shader_paths_length = (sizeof (fragment_shader_paths)) / (sizeof (fragment_shader_paths[0]));
+
+#define PARTICLE_TEXTURE 0
+
+const char * textures[] = {
+  "../texture/butterfly_1024x1024_rgba8888.data",
+};
+const int textures_length = (sizeof (textures)) / (sizeof (textures[0]));
+
+struct shaders {
+  struct shader_offset * vertex;
+  struct shader_offset * fragment;
+  int vertex_length;
+  int fragment_length;
+};
+
+const int floatbuffer_width = 64;
+const int floatbuffer_height = 64;
+
+int _floatbuffer(const shaders& shaders,
+                 int input_reloc_index,
+                 int output_reloc_index)
+{
+  int viewport_width = floatbuffer_width * 4;
+  int viewport_height = floatbuffer_height;
+  int texture_width = floatbuffer_width;
+  int texture_height = floatbuffer_height;
+  float vx = ((float)viewport_width) * 0.5f;
+  float vy = ((float)viewport_height) * 0.5f;
+  float tx = 0.5f / ((float)texture_width);
+  float ty = 0.5f / ((float)texture_height);
+
+  ib_ix = 0;
+
+  ib_generic_initialization();
+
+  printf("vp %d %d \n", viewport_width, viewport_height);
+
+  T0V(SC_SCISSOR0
+      , SC_SCISSOR0__XS0(0)
+      | SC_SCISSOR0__YS0(0)
+      );
+  T0V(SC_SCISSOR1
+      , SC_SCISSOR1__XS1(viewport_width - 1)
+      | SC_SCISSOR1__YS1(viewport_height - 1)
+      );
+  T0Vf(VAP_VPORT_XSCALE, (float)viewport_width);
+  T0Vf(VAP_VPORT_YSCALE, (float)viewport_height);
+
+  ib_colorbuffer(output_reloc_index, viewport_width, 0, 0); // macrotile, microtile
+
+  T0V(RB3D_COLORPITCH0
+      , RB3D_COLORPITCH__COLORPITCH(viewport_width >> 1)
+      | RB3D_COLORPITCH__COLORTILE(0)
+      | RB3D_COLORPITCH__COLORMICROTILE(0)
+      | RB3D_COLORPITCH__COLORFORMAT__ARGB32323232
+      );
+
+  T0V(US_OUT_FMT_0
+      , US_OUT_FMT__OUT_FMT(21)  // C4_32_FP
+      | US_OUT_FMT__C0_SEL__RED
+      | US_OUT_FMT__C1_SEL__GREEN
+      | US_OUT_FMT__C2_SEL__BLUE
+      | US_OUT_FMT__C3_SEL__ALPHA
+      | US_OUT_FMT__OUT_SIGN(0)
+      );
+  T0V(US_OUT_FMT_1
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_2
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+  T0V(US_OUT_FMT_3
+      , US_OUT_FMT__OUT_FMT(15) // render target is not used
+      );
+
+  // shaders
+  load_pvs_shaders(shaders.vertex, shaders.vertex_length);
+  load_us_shaders(shaders.fragment, shaders.fragment_length);
+
+  // GA
+
+  T0V(GB_ENABLE
+      , 0
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // RS
+  //////////////////////////////////////////////////////////////////////////////
+
+  ib_rs_instructions(1);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP OUT
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_OUT_VTX_FMT_0
+      , VAP_OUT_VTX_FMT_0__VTX_POS_PRESENT(1)
+      );
+  T0V(VAP_OUT_VTX_FMT_1
+      , VAP_OUT_VTX_FMT_1__TEX_0_COMP_CNT(4)
+      );
+
+  //
+
+  T0V(ZB_CNTL, 0);
+  T0V(ZB_ZSTENCILCNTL, 0);
+
+  //
+
+  int macrotile = 0;
+  int microtile = 0;
+  int clamp = 2; // clamp to [0.0, 1.0]
+  ib_texture__1_float32(input_reloc_index,
+                        texture_width, texture_height,
+                        macrotile, microtile,
+                        clamp);
+
+  ib_vap_stream_cntl__2();
+
+  T0V(US_PIXSIZE
+      , US_PIXSIZE__PIX_SIZE(1)
+      );
+
+  ib_ga_us(&shaders.fragment[TEXTURE_TILE_SHADER]);
+  ib_vap_pvs(&shaders.vertex[TEXTURE_TILE_SHADER]);
+
+  const float vertex_consts[] = {
+    tx, ty, 0, 0,
+  };
+  const int vertex_consts_size = (sizeof (vertex_consts));
+  ib_vap_pvs_const_cntl(vertex_consts, vertex_consts_size);
+
+  // fragment constants
+
+  const float fragment_consts[] = {
+    1234.0f, 0, 0, 0,
+  };
+  int fragment_consts_length = (sizeof (fragment_consts)) / (sizeof (fragment_consts[0]));
+  ib_ga_consts(fragment_consts, fragment_consts_length, 0);
+
+  //////////////////////////////////////////////////////////////////////////////
+  // VAP
+  //////////////////////////////////////////////////////////////////////////////
+
+  T0V(VAP_CLIP_CNTL
+      , VAP_CLIP_CNTL__CLIP_DISABLE(1)
+      );
+
+  T0V(VAP_VTE_CNTL
+      , VAP_VTE_CNTL__VPORT_X_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VPORT_Y_SCALE_ENA(1)
+      | VAP_VTE_CNTL__VTX_XY_FMT(1) // disable W division
+      | VAP_VTE_CNTL__VTX_Z_FMT(1)  // disable W division
+      );
+
+  T0V(VAP_CNTL_STATUS
+      , VAP_CNTL_STATUS__PVS_BYPASS(0)
+      );
+
+  //////////////////////////////////////////////////////////////////////////////
+  // 3D_DRAW
+  //////////////////////////////////////////////////////////////////////////////
+
+  const int dwords_per_vtx = 2;
+
+  T0V(VAP_VTX_SIZE
+      , VAP_VTX_SIZE__DWORDS_PER_VTX(dwords_per_vtx)
+      );
+
+  const float vertices[] = {
+    0.0f, 0.0f,
+    1.0f, 0.0f,
+    1.0f, 1.0f,
+    0.0f, 1.0f,
+  };
+  const int vertex_count = 4;
+  T3(_3D_DRAW_IMMD_2, (1 + vertex_count * dwords_per_vtx) - 1);
+  TU( VAP_VF_CNTL__PRIM_TYPE(5) // triangle fan
+    | VAP_VF_CNTL__PRIM_WALK(3)
+    | VAP_VF_CNTL__INDEX_SIZE(0)
+    | VAP_VF_CNTL__VTX_REUSE_DIS(0)
+    | VAP_VF_CNTL__DUAL_INDEX_MODE(0)
+    | VAP_VF_CNTL__USE_ALT_NUM_VERTS(0)
+    | VAP_VF_CNTL__NUM_VERTICES(vertex_count)
+    );
+  for (int i = 0; i < vertex_count * 2; i++) {
+    TF(vertices[i]);
+  }
+
+  return ib_ix;
+}
+
+void floatbuffer_data(void * ptr, int size)
+{
+  float * f32 = (float*)ptr;
+  int vector_length = size / (4 * 4);
+
+  int offset = 16384;
+  for (int i = 0; i < vector_length; i++) {
+    f32[i * 4 + 0] = offset--;
+    f32[i * 4 + 1] = offset--;
+    f32[i * 4 + 2] = offset--;
+    f32[i * 4 + 3] = offset--;
+  }
+}
+
+void floatbuffer_compare(void * a, void * b, int size)
+{
+  float * a_f32 = (float*)a;
+  float * b_f32 = (float*)b;
+  int vector_length = size / (4 * 4);
+
+  int matches = 0;
+
+  int ix = 0;
+  for (int i = 0; i < vector_length; i++) {
+    for (int j = 0; j < 4; j++) {
+      if (   (a_f32[i * 4 + 0] != b_f32[ix * 4 + 0])
+             || (a_f32[i * 4 + 1] != b_f32[ix * 4 + 1])
+             || (a_f32[i * 4 + 2] != b_f32[ix * 4 + 2])
+             || (a_f32[i * 4 + 3] != b_f32[ix * 4 + 3])) {
+        printf("a[%d] = [% 2.02f  % 2.02f  % 2.02f  % 2.02f] ; ", i,
+               a_f32[i * 4 + 0], a_f32[i * 4 + 1], a_f32[i * 4 + 2], a_f32[i * 4 + 3]);
+        printf("b[%d] = [% 2.02f  % 2.02f  % 2.02f  % 2.02f] \n", i,
+               b_f32[ix * 4 + 0], b_f32[ix * 4 + 1], b_f32[ix * 4 + 2], b_f32[ix * 4 + 3]);
+      } else {
+        matches += 1;
+      }
+      ix++;
+    }
+  }
+  printf("vector_length %d matches %d\n", vector_length, matches);
+}
+
+int main()
+{
+  struct shaders shaders = {
+    .vertex = load_shaders(vertex_shader_paths, vertex_shader_paths_length),
+    .fragment = load_shaders(fragment_shader_paths, fragment_shader_paths_length),
+    .vertex_length = vertex_shader_paths_length,
+    .fragment_length = fragment_shader_paths_length,
+  };
+
+  void * rmmio = map_pci_resource2();
+
+  int fd = open("/dev/dri/card0", O_RDWR | O_CLOEXEC);
+  assert(fd != -1);
+
+  const int colorbuffer_size = 1600 * 1200 * 4;
+  const int floatbuffer_size = floatbuffer_width * floatbuffer_height * 4 * 4;
+  const int floatbuffer_count = 2;
+  int colorbuffer_handle[2];
+  int zbuffer_handle;
+  int * texturebuffer_handle;
+  int flush_handle;
+  int floatbuffer_handle[floatbuffer_count];
+
+  void * colorbuffer_ptr[2];
+  void * floatbuffer_ptr[floatbuffer_count];
+  void * zbuffer_ptr;
+
+  // colorbuffer
+  colorbuffer_handle[0] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[0]);
+  colorbuffer_handle[1] = create_buffer(fd, colorbuffer_size, &colorbuffer_ptr[1]);
+  zbuffer_handle = create_buffer(fd, colorbuffer_size, &zbuffer_ptr);
+  flush_handle = create_flush_buffer(fd);
+  texturebuffer_handle = load_textures(fd, textures, textures_length);
+
+  floatbuffer_handle[0] = create_buffer(fd, floatbuffer_size, &floatbuffer_ptr[0]);
+  floatbuffer_handle[1] = create_buffer(fd, floatbuffer_size * 4, &floatbuffer_ptr[1]);
+
+  fprintf(stderr, "colorbuffer handle[0] %d\n", colorbuffer_handle[0]);
+  fprintf(stderr, "colorbuffer handle[1] %d\n", colorbuffer_handle[1]);
+  fprintf(stderr, "floatbuffer handle[0] %d\n", floatbuffer_handle[0]);
+  fprintf(stderr, "floatbuffer handle[1] %d\n", floatbuffer_handle[1]);
+  fprintf(stderr, "zbuffer handle %d\n", zbuffer_handle);
+
+  int colorbuffer_ix = 0;
+
+  floatbuffer_data(floatbuffer_ptr[0], floatbuffer_size);
+
+  while (true) {
+    int ib_dwords = _floatbuffer(shaders,
+                                 TEXTUREBUFFER_RELOC_INDEX + 0, // input
+                                 COLORBUFFER_RELOC_INDEX);      // output
+
+    int ret = drm_radeon_cs(fd,
+                            floatbuffer_handle[1],
+                            zbuffer_handle,
+                            flush_handle,
+                            floatbuffer_handle,
+                            floatbuffer_count,
+                            ib_dwords);
+    if (ret == -1)
+      break;
+
+    break;
+  }
+
+  floatbuffer_compare(floatbuffer_ptr[0], floatbuffer_ptr[1], floatbuffer_size);
+
+  close(fd);
+}

src/vertex_buffer_copy.fs.asm

@@ -0,0 +1,10 @@
+-- temp[0]: texture coordinate
+
+TEX TEX_SEM_WAIT TEX_SEM_ACQUIRE
+  temp[0].rgba = LD tex[0].rgba temp[0].rgaa ;
+
+OUT TEX_SEM_WAIT
+src0.a = temp[0] ,
+src0.rgb = temp[0] :
+  out[0].a    = MAX src0.a src0.a ,
+  out[0].rgb  = MAX src0.rgb src0.rgb ;

src/vertex_buffer_copy.vs.asm

@@ -0,0 +1,2 @@
+out[0].xyzw = VE_ADD  input[0].xyzw  input[0].0000 ;
+out[1].xyzw = VE_ADD  input[0].xy00  input[0].0000 ;