From 1c24f9e297c3340c25182d7e956d2b92d6436ed6 Mon Sep 17 00:00:00 2001
From: Zack Buhman <zack@buhman.org>
Date: Sun, 26 Jan 2025 07:03:37 -0600
Subject: [PATCH] wiffle_screen_space: use sobel_fipr

---
 example/example.mk              |  1 +
 example/wiffle_screen_space.cpp | 32 +++++++++++++++++++++++---------
 sobel.cpp                       |  2 +-
 sobel_fipr.s                    |  9 +++++++++
 4 files changed, 34 insertions(+), 10 deletions(-)

diff --git a/example/example.mk b/example/example.mk
index 6147792..e1ea0c0 100644
--- a/example/example.mk
+++ b/example/example.mk
@@ -217,6 +217,7 @@ WIFFLE_SCREEN_SPACE_OBJ = \
 	holly/background.o \
 	holly/ta_fifo_polygon_converter.o \
 	sh7091/serial.o \
+	sobel_fipr.o \
 	sobel.o
 
 example/wiffle_screen_space.elf: LDSCRIPT = $(LIB)/main.lds
diff --git a/example/wiffle_screen_space.cpp b/example/wiffle_screen_space.cpp
index 029aed4..f632cae 100644
--- a/example/wiffle_screen_space.cpp
+++ b/example/wiffle_screen_space.cpp
@@ -253,6 +253,8 @@ static uint32_t inbuf[640 * 480] __attribute__((aligned(32)));
 static float temp[640 * 480] __attribute__((aligned(32)));
 static uint32_t outbuf[640 * 480] __attribute__((aligned(32)));
 
+extern "C" int sobel_fipr(float * a, uint32_t * i);
+
 void make_temp()
 {
   for (int i = 0; i < 640 * 480; i++) {
@@ -377,7 +379,7 @@ void main()
     uint32_t * in = (uint32_t *)&texture_memory64[texture_memory_alloc.texture.start / 4];
     //uint32_t * out = (uint32_t *)&texture_memory32[texture_memory_alloc.framebuffer[0].start / 4];
 
-    serial::string("ch1 dma start\n");
+    //serial::string("ch1 dma start\n");
     dma_transfer((uint32_t)in, (uint32_t)inbuf, 640 * 480 * 4 / 32);
 
     for (uint32_t i = 0; i < (sizeof (640 * 480 * 4)) / 32; i++) {
@@ -389,22 +391,34 @@ void main()
     }
 
     while ((sh7091.DMAC.CHCR1 & dmac::chcr::te::transfers_completed) == 0);
-    serial::string("ch1 dma end\n");
+    //serial::string("ch1 dma end\n");
 
-    serial::string("temp start\n");
+    //serial::string("temp start\n");
     make_temp();
-    serial::string("temp end\n");
+    //serial::string("temp end\n");
 
-    serial::string("convolve start\n");
-    convolve(temp, outbuf);
-    serial::string("convolve end\n");
+    //serial::string("convolve start\n");
+
+    //convolve(temp, outbuf);
+    int a = sobel_fipr(temp, outbuf);
+    //serial::integer<uint32_t>((uint32_t)temp);
+    //serial::integer<uint32_t>(a);
+    for (uint32_t i = 0; i < (sizeof (640 * 480 * 4)) / 32; i++) {
+      uint32_t address = (uint32_t)&outbuf[0];
+      asm volatile ("ocbwb @%0"
+                    :                          // output
+                    : "r" (address + (i * 32)) // input
+                    );
+    }
+
+    //serial::string("convolve end\n");
 
     uint32_t framebuffer = 0x11000000 + texture_memory_alloc.framebuffer[0].start; // TA FIFO - Direct Texture Path
     system.LMMODE0 = 1;
     system.LMMODE1 = 1; // 32-bit
-    serial::string("ch2 dma start\n");
+    //serial::string("ch2 dma start\n");
     ch2_dma_transfer((uint32_t)outbuf, framebuffer, (640 * 480 * 4) / 32);
-    serial::string("ch2 dma end\n");
+    //serial::string("ch2 dma end\n");
 
     while (!spg_status::vsync(holly.SPG_STATUS));
     holly.FB_R_SOF1 = texture_memory_alloc.framebuffer[0].start;
diff --git a/sobel.cpp b/sobel.cpp
index 98c83fa..5d08f2f 100644
--- a/sobel.cpp
+++ b/sobel.cpp
@@ -70,7 +70,7 @@ void convolve(float * in, uint32_t * out)
       float c = kernel2(in, x, y);
       int d = c > 100.f ? 0 : 0xffffffff;
 
-      out[y * 640 + x] = (uint8_t)d;
+      out[y * 640 + x] = d;
     }
   }
 }
diff --git a/sobel_fipr.s b/sobel_fipr.s
index 7d7540e..0909ce1 100644
--- a/sobel_fipr.s
+++ b/sobel_fipr.s
@@ -6,6 +6,10 @@ __setup:
         mov.l r9,@-r15
         mov.l r10,@-r15
         mov.l r11,@-r15
+        fmov.s  fr12,@-r15
+        fmov.s  fr13,@-r15
+        fmov.s  fr14,@-r15
+        fmov.s  fr15,@-r15
 
         fldi1 fr8    /* 1.0 */
         fldi1 fr9    /* 2.0 */
@@ -141,6 +145,11 @@ _loop_width:
 
         /* restore registers */
 _return:
+        fmov.s  @r15+,fr15
+        fmov.s  @r15+,fr14
+        fmov.s  @r15+,fr13
+        fmov.s  @r15+,fr12
+
         mov.l @r15+,r11
         mov.l @r15+,r10
         mov.l @r15+,r9