#368 GBA audio processing optimizations, enable VRAM meshes, cleanup …

…rendering constants
XProger · Dec 1, 2022 · b1a5590 · b1a5590
1 parent b70eade
commit b1a5590
Show file tree

Hide file tree

Showing 7 changed files with 149 additions and 107 deletions.
diff --git a/src/fixed/common.h b/src/fixed/common.h
@@ -66,6 +66,7 @@
     #define FRAME_HEIGHT 160
 
     #define USE_FMT     (LVL_FMT_PKD)
+    #define USE_VRAM_MESH // experimental
 
     #include <tonc.h>
 #elif defined(__NDS__)
@@ -189,8 +190,6 @@
 #include <math.h>
 #include <limits.h>
 
-#define VRAM_WIDTH   (FRAME_WIDTH/2)    // in shorts
-
 #ifndef USE_FMT
     #define USE_FMT (LVL_FMT_PHD | LVL_FMT_PSX | LVL_FMT_SAT | LVL_FMT_TR2 | LVL_FMT_TR4)
 #endif
@@ -343,13 +342,13 @@ X_INLINE int32 abs(int32 x) {
 #endif
 
 #if defined(__GBA_WIN__)
-    extern uint16 fb[VRAM_WIDTH * FRAME_HEIGHT];
+    extern uint16 fb[FRAME_WIDTH * FRAME_HEIGHT];
 #elif defined(__GBA__)
     extern uint32 fb;
 #elif defined(__TNS__)
-    extern uint16 fb[VRAM_WIDTH * FRAME_HEIGHT];
+    extern uint16 fb[FRAME_WIDTH * FRAME_HEIGHT];
 #elif defined(__DOS__)
-    extern uint16 fb[VRAM_WIDTH * FRAME_HEIGHT];
+    extern uint16 fb[FRAME_WIDTH * FRAME_HEIGHT];
 #endif
 
 #define STATIC_MESH_FLAG_NO_COLLISION   1
@@ -2839,7 +2838,7 @@ int32 doTutorial(ItemObj* lara, int32 track);
 void sndInit();
 void sndInitSamples();
 void sndFreeSamples();
-void sndFill(int8* buffer, int32 count);
+void sndFill(int8* buffer);
 void* sndPlaySample(int32 index, int32 volume, int32 pitch, int32 mode);
 void sndPlayTrack(int32 track);
 bool sndTrackIsPlaying();

diff --git a/src/platform/gba/asm/sndIMA.s b/src/platform/gba/asm/sndIMA.s
@@ -10,49 +10,53 @@ stepLUT .req r6
 step    .req r7
 n       .req r8
 index   .req r9
+mask    .req r10
 out     .req r12
 tmp     .req out
+diff    .req step
 
 IMA_STEP_SIZE = 88
 
-.macro decode4 n, out
+.macro ima_decode
     ldr step, [stepLUT, idx, lsl #2]
 
-    and index, \n, #7
-    mov tmp, step, lsl #1
-    mla step, index, tmp, step
-    tst \n, #8
-    subne smp, smp, step, lsr #3
-    addeq smp, smp, step, lsr #3
+    mul tmp, step, index
+    add diff, tmp, lsl #1
+
+    subne smp, diff, lsr #3
+    addeq smp, diff, lsr #3
 
     subs index, #3
-    suble idx, idx, #1
-    bicle idx, idx, idx, asr #31
-    addgt idx, idx, index, lsl #1
-    cmpgt idx, #IMA_STEP_SIZE
+    suble idx, #1
+    addgt idx, index, lsl #1
+
+    // clamp 0..88
+    bic idx, idx, asr #31
+    cmp idx, #IMA_STEP_SIZE
     movgt idx, #IMA_STEP_SIZE
 
-    mov \out, smp, asr #(2 + SND_VOL_SHIFT)
+    mov out, smp, asr #(2 + SND_VOL_SHIFT)
+    strb out, [buffer], #1
 .endm
 
-.global sndIMA_asm
-sndIMA_asm:
+.global sndIMA_fill_asm
+sndIMA_fill_asm:
     stmfd sp!, {r4-r9}
 
     ldmia state, {smp, idx}
-
     ldr stepLUT, =IMA_STEP
 
+    mov mask, #7
 .loop:
     ldrb n, [data], #1
 
-    decode4 n, out
-    strb out, [buffer], #1
-
-    mov n, n, lsr #4
+    and index, mask, n
+    tst n, #8
+    ima_decode
 
-    decode4 n, out
-    strb out, [buffer], #1
+    and index, mask, n, lsr #4
+    tst n, #(8 << 4)
+    ima_decode
 
     subs size, #1
     bne .loop

diff --git a/src/platform/gba/asm/sndPCM.s b/src/platform/gba/asm/sndPCM.s
@@ -7,61 +7,87 @@ volume  .req r3
 
 data    .req r4
 buffer  .req r5
-count   .req r6
-ampA    .req r7
-ampB    .req r8
-outA    .req r9
-outB    .req r12
-last    .req count
-tmpSP   .req outB
-tmp     .req ampA
-
-.macro clamp amp
+tmp     .req r6
+last    .req r12
+tmpSP   .req last
+out     .req size
+
+.macro clamp
     // Vanadium's clamp trick (-128..127)
-    mov tmp, \amp, asr #31  // tmp <- 0xffffffff
-    cmp tmp, \amp, asr #7   // not equal
-    eorne \amp, tmp, #0x7F  // amp <- 0xffffff80
+    mov tmp, out, asr #31  // tmp <- 0xffffffff
+    cmp tmp, out, asr #7   // not equal
+    eorne out, tmp, #0x7F  // out <- 0xffffff80
+.endm
+
+.macro calc_last
+    // last = pos + inc * SND_SAMPLES (176)
+    add last, inc, inc, lsl #2      // last = inc * 5
+    add last, inc, last, lsl #1     // last = inc * 11
+    add last, pos, last, lsl #4     // last = pos + (inc * 11) * 16
+.endm
+
+.macro pcm_sample_fetch
+    ldrb out, [data, pos, lsr #SND_FIXED_SHIFT]
+    add pos, inc
+    sub out, #128
+    mul out, volume
+.endm
+
+.macro pcm_sample_fill
+    pcm_sample_fetch
+    asr out, #SND_VOL_SHIFT
+    strb out, [buffer], #1
+.endm
+
+.macro pcm_sample_mix
+    pcm_sample_fetch
+    ldrsb tmp, [buffer]
+    add out, tmp, out, asr #SND_VOL_SHIFT
+    clamp
+    strb out, [buffer], #1
 .endm
 
-.global sndPCM_asm
-sndPCM_asm:
+.global sndPCM_fill_asm
+sndPCM_fill_asm:
     mov tmpSP, sp
-    stmfd sp!, {r4-r9}
+    stmfd sp!, {r4-r5}
 
-    ldmia tmpSP, {data, buffer, count}
+    ldmia tmpSP, {data, buffer}
+
+    calc_last
 
-    mla last, inc, count, pos
     cmp last, size
     movgt last, size
 
-.loop:
-    ldrb ampA, [data, pos, lsr #SND_FIXED_SHIFT]
-    add pos, pos, inc
-    ldrb ampB, [data, pos, lsr #SND_FIXED_SHIFT]
-    add pos, pos, inc
+.loop_fill:
+    pcm_sample_fill
+    pcm_sample_fill
 
-    // can't use signed PCM because of LDRSB restrictions
-    sub ampA, ampA, #128
-    sub ampB, ampB, #128
+    cmp pos, last
+    blt .loop_fill
 
-    mul ampA, volume
-    mul ampB, volume
+    ldmfd sp!, {r4-r5}
+    bx lr
 
-    ldrsb outA, [buffer, #0]
-    ldrsb outB, [buffer, #1]
 
-    add outA, ampA, asr #SND_VOL_SHIFT
-    add outB, ampB, asr #SND_VOL_SHIFT
+.global sndPCM_mix_asm
+sndPCM_mix_asm:
+    mov tmpSP, sp
+    stmfd sp!, {r4-r6} // tmp reg required
 
-    clamp outA
-    clamp outB
+    ldmia tmpSP, {data, buffer}
 
-    strb outA, [buffer], #1
-    strb outB, [buffer], #1
+    calc_last
+
+    cmp last, size
+    movgt last, size
+
+.loop_mix:
+    pcm_sample_mix
+    pcm_sample_mix
 
     cmp pos, last
-    blt .loop
+    blt .loop_mix
 
-.done:
-    ldmfd sp!, {r4-r9}
-    bx lr
+    ldmfd sp!, {r4-r6}
+    bx lr
diff --git a/src/platform/gba/main.cpp b/src/platform/gba/main.cpp
@@ -123,7 +123,7 @@ void soundFill()
 {
     WAVEHDR *waveHdr = waveBuf + curSoundBuffer;
     waveOutUnprepareHeader(waveOut, waveHdr, sizeof(WAVEHDR));
-    sndFill((int8*)waveHdr->lpData, SND_SAMPLES);
+    sndFill((int8*)waveHdr->lpData);
     waveOutPrepareHeader(waveOut, waveHdr, sizeof(WAVEHDR));
     waveOutWrite(waveOut, waveHdr, sizeof(WAVEHDR));
     curSoundBuffer ^= 1;
@@ -503,7 +503,7 @@ void soundFill()
         REG_DMA1CNT = DMA_DST_FIXED | DMA_REPEAT | DMA_16 | DMA_AT_FIFO | DMA_ENABLE;
     }
 
-    sndFill(soundBuffer + curSoundBuffer, SND_SAMPLES);
+    sndFill(soundBuffer + curSoundBuffer);
     curSoundBuffer ^= SND_SAMPLES;
 }
 

diff --git a/src/platform/gba/rasterizer.h b/src/platform/gba/rasterizer.h
@@ -143,7 +143,7 @@ void rasterizeS_c(uint16* pixel, const VertexLink* L, const VertexLink* R)
                 }
             }
 
-            pixel += VRAM_WIDTH;
+            pixel += (FRAME_WIDTH >> 1);
 
             Lx += Ldx;
             Rx += Rdx;
@@ -253,7 +253,7 @@ void rasterizeF_c(uint16* pixel, const VertexLink* L, const VertexLink* R)
                 }
             }
 
-            pixel += VRAM_WIDTH;
+            pixel += (FRAME_WIDTH >> 1);
 
             Lx += Ldx;
             Rx += Rdx;
@@ -377,7 +377,7 @@ void rasterizeFT_c(uint16* pixel, const VertexLink* L, const VertexLink* R)
                 }
             }
 
-            pixel += VRAM_WIDTH;
+            pixel += (FRAME_WIDTH >> 1);
 
             Lx += Ldx;
             Rx += Rdx;
@@ -533,7 +533,7 @@ void rasterizeGT_c(uint16* pixel, const VertexLink* L, const VertexLink* R)
                 }
             }
 
-            pixel += VRAM_WIDTH;
+            pixel += (FRAME_WIDTH >> 1);
 
             Lx += Ldx;
             Rx += Rdx;
@@ -672,7 +672,7 @@ void rasterizeFTA_c(uint16* pixel, const VertexLink* L, const VertexLink* R)
                 }
             }
 
-            pixel += VRAM_WIDTH;
+            pixel += (FRAME_WIDTH >> 1);
 
             Lx += Ldx;
             Rx += Rdx;
@@ -846,7 +846,7 @@ void rasterizeGTA_c(uint16* pixel, const VertexLink* L, const VertexLink* R)
                 }
             }
 
-            pixel += VRAM_WIDTH;
+            pixel += (FRAME_WIDTH >> 1);
 
             Lx += Ldx;
             Rx += Rdx;
@@ -970,7 +970,7 @@ extern "C" X_NOINLINE void rasterizeSprite_c(uint16* pixel, const VertexLink* L,
 
     if (L->v.y < 0)
     {
-        pixel -= L->v.y * VRAM_WIDTH;
+        pixel -= L->v.y * (FRAME_WIDTH >> 1);
         v -= L->v.y * dv;
         h += L->v.y;
     }

diff --git a/src/platform/gba/render.iwram.cpp b/src/platform/gba/render.iwram.cpp
@@ -26,13 +26,13 @@ struct ViewportRel {
 ViewportRel viewportRel;
 
 #if defined(__GBA_WIN__)
-    uint16 fb[VRAM_WIDTH * FRAME_HEIGHT];
+    uint16 fb[FRAME_WIDTH * FRAME_HEIGHT];
 #elif defined(__GBA__)
     uint32 fb = MEM_VRAM;
 #elif defined(__TNS__)
-    uint16 fb[VRAM_WIDTH * FRAME_HEIGHT];
+    uint16 fb[FRAME_WIDTH * FRAME_HEIGHT];
 #elif defined(__DOS__)
-    uint16 fb[VRAM_WIDTH * FRAME_HEIGHT];
+    uint16 fb[FRAME_WIDTH * FRAME_HEIGHT];
 #endif
 
 enum FaceType {
@@ -150,7 +150,7 @@ extern "C" {
     #define faceAddMeshTriangles    faceAddMeshTriangles_c
     #define rasterize               rasterize_c
 
-X_INLINE bool checkBackface(const Vertex *a, const Vertex *b, const Vertex *c)
+X_INLINE bool checkBackface(const Vertex* a, const Vertex* b, const Vertex* c)
 {
     return (b->x - a->x) * (c->y - a->y) <= (c->x - a->x) * (b->y - a->y);
 }
@@ -803,7 +803,7 @@ void faceAddMesh(const MeshQuad* quads, const MeshTriangle* triangles, int32 qCo
 
 void clear()
 {
-    dmaFill((void*)fb, 0, VRAM_WIDTH * FRAME_HEIGHT * 2);
+    dmaFill((void*)fb, 0, FRAME_WIDTH * FRAME_HEIGHT);
 }
 
 void renderRoom(const Room* room)