Add render tile visualisations

NOTES.md

@@ -1,9 +1,18 @@
 
+# Things that are interesting about IpuDoom
+
+ - Fixed precision int math
+    - 64-bit int division
+        - wasn't supported in the supervisor context, drop into a worker thread for each individual division, do double division
+- Avoid recursive algorithms, e.g. make a non-recursive BSP
+- Split column rendering over 32 tiles
+- Memory
+    - Save 50-60k of lookup tabes for transcendentals by using float functions
+    - Store textures on other tiles, JIT-fetch as needed (by live patching exchange programs to select a tile)
+        - Can call exchanges from deep in the call stack, don't need to exit the vertex
+
 
-Things I don't support, to make my life easier:
+#### Things I don't support, to make my life easier:
  - gamemodes other than shareware
  - multiplayer
  - cheating
-
-What about:
- -precomputed trancendental takes like 32kb on its own, recompute live?

src/ipu/ipu_texturetiles.cpp

@@ -78,6 +78,8 @@ void IPU_R_FulfilColumnRequest(unsigned* progBuf, unsigned char* textureBuf, uns
     auto sendProgram = &progBuf[progBuf[1]];
     auto aggrProgram = &progBuf[progBuf[2]];
 
+    int noise = 0;
+
     while (1) {
 
       XCOM_Execute(recvProgram, NULL, textureBuf);
@@ -91,20 +93,37 @@ void IPU_R_FulfilColumnRequest(unsigned* progBuf, unsigned char* textureBuf, uns
           textureNum >= tileLocalTextureRange[0] && 
           textureNum < tileLocalTextureRange[1]) {
         col = &textureBuf[tileLocalTextureOffsets[textureNum] + columnOffset];
-        // if (firstPrint && tileID >= 32 && tileID < 36) printf("Accepted texture %d\n", (int)textureNum);
-      } else {
-        // if (firstPrint && tileID >= 32 && tileID < 36) printf("Rejected texture %d\n", (int)textureNum);
       }
       
-      byte c1 = (17 * 9) % 256;
-      byte c2 = (c1 + 1) % 256;
-      byte c3 = (c1 + 2) % 256;
-      byte c4 = (c1 + 1) % 256;
-      unsigned colour = c1 | (c2 << 8) | (c3 << 16) | (c4 << 24);
-      for (int i = 0; i < IPUTEXTURECACHELINESIZE; i++) {
-        ((unsigned*)textureBuf)[i] = colour;
-      }
+
+    //   noise = (noise + 1) % 4;
+    //   unsigned renderStripe = (tileID - IPUFIRSTTEXTURETILE) / IPUTEXTURETILESPERRENDERTILE;
+    //   {
+    //     // unsigned colour = (renderStripe * 9) % 256;
+    //     colour = colour | (colour << 8) | (colour << 16) | (colour << 24);
+    //     for (int i = 0; i < IPUTEXTURECACHELINESIZE; i++) {
+    //         if (noise == 0) {
+    //             ((unsigned*)col)[i] += 0x01010101;
+    //         }
+    //         // ((unsigned*)col)[i] = colour;
+    //     }
+    //   }
+      
+    //   {
+    //     // Render Tile Stripes
+    //     unsigned colour = renderStripe * 16;
+    //     for (int i = 0; i < IPUTEXTURECACHELINESIZE * sizeof(int); i++) {
+    //         if (renderStripe % 2) {
+    //             textureBuf[i] = (col[i] % 4); //(col[i] % 30) + 20;
+    //         } else {
+    //             textureBuf[i] = (col[i] % 30) + 80; //(col[i] % 4);
+    //         }
+    //     }
+    //     col = textureBuf;
+    //   }
+
       XCOM_Execute(sendProgram, col, NULL);
+      // XCOM_Execute(sendProgram, col, NULL);
 
       XCOM_Execute(aggrProgram, NULL, commsBuf);
       if (commsBuf[0]) 

src/ipu/m_fixed.cpp

@@ -55,25 +55,17 @@ fixed_t FixedMul(fixed_t a, fixed_t b)
 struct DivWorker : public poplar::Vertex {
     fixed_t a, b;
     void compute() {
-        // Using int64_ caused some weird errors. So use double instead...?
+        // Using int64_t caused some weird errors. So use double instead...?
         // a = (((int64_t) a) << FRACBITS) / b;
         a = ((double) a * FRACUNIT) / (double) b; 
     }
 };
 
-//
-// FixedDiv, C version.
-//
-extern "C"
-__SUPER__ 
 fixed_t FixedDiv(fixed_t a, fixed_t b)
 {
     if ((abs(a) >> 14) >= abs(b)) {
     	return (a^b) < 0 ? INT_MIN : INT_MAX;
     } else {
-        // int64_t result; // JOSEF
-        // result = ((int64_t) a << FRACBITS) / b;
-        // return (fixed_t) result;
 
         DivWorker workerArgs;
         unsigned workerArgsPtr = (unsigned)&workerArgs - TMEM_BASE_ADDR;
@@ -91,4 +83,3 @@ fixed_t FixedDiv(fixed_t a, fixed_t b)
         return workerArgs.a;
     }
 }
-

src/ipu/r_codelets.cpp

@@ -88,6 +88,9 @@ R_RenderPlayerView_Vertex : public poplar::SupervisorVertex {
       (R_RenderPlayerView_MiscValues_t*) &miscValues[0]
     );
 
+    // For visualisation of progress
+    // memset(I_VideoBuffer, 0, IPUCOLSPERRENDERTILE * SCREENHEIGHT);
+
     R_RenderPlayerView(&players[displayplayer]);
     IPU_R_RenderTileDone();
     return;

src/r_data.c

@@ -333,7 +333,7 @@ void R_GenerateLookup(int texnum) {
 //
 // R_GetColumn
 //
-byte *R_GetColumn_Original(int tex, int col) { // JOSEF: Renamed to `_Original`
+byte *R_GetColumn(int tex, int col) { // JOSEF: Renamed to `_Original`
   int lump;
   int ofs;
 
@@ -382,7 +382,7 @@ void GenerateIPUTextureBlob(void) {
     ipuTextureBlobOffsets[t] = pos;
     byte* dst = &ipuTextureBlob[tile * IPUTEXTURETILEBUFSIZE + pos];
     for (int x = 0; x < tex_width; x += 1, dst += tex_height) {
-      byte* col = R_GetColumn_Original(t, x);
+      byte* col = R_GetColumn(t, x);
       memcpy(dst, col, tex_height);
     }
     pos += tex_size;
@@ -393,8 +393,7 @@ void GenerateIPUTextureBlob(void) {
 //
 // R_GetColumn : JOSEF: The version that uses the IPU texture dump
 //
-byte *R_GetColumn(int tex, int col) {
-  // return R_GetColumn_Original(tex, col); // For checking old behaviour
+byte *R_GetColumn_LikeIPU(int tex, int col) {
   int tile;
   col &= texturewidthmask[tex];
   for (tile = 0; tile < IPUTEXTURETILESPERRENDERTILE; ++tile) {