#368 GBA conditional mul in rasterizer

XProger · Dec 4, 2022 · 5c13524 · 5c13524
1 parent 3ba5ec3
commit 5c13524
Show file tree

Hide file tree

Showing 7 changed files with 67 additions and 71 deletions.
diff --git a/src/platform/gba/asm/common_asm.inc b/src/platform/gba/asm/common_asm.inc
@@ -121,11 +121,11 @@
 .endm
 
 .macro scaleUV uv, tmp, tmp2, f
-    smull \tmp2, \tmp, \uv, \f      // u = (f * uv) >> 32
+    smullne \tmp2, \tmp, \uv, \f      // u = (f * uv) >> 32
 
     lsl \uv, #16
-    asr \uv, #16
-    mul \uv, \f                     // v = f * int16(uv)
+    asrs \uv, #16
+    mulne \uv, \f                     // v = f * int16(uv)
     lsr \uv, #16
 
     orr \uv, \uv, \tmp, lsl #16     // uv = (u & 0xFFFF0000) | (v >> 16)

diff --git a/src/platform/gba/asm/rasterizeF.s b/src/platform/gba/asm/rasterizeF.s
@@ -25,8 +25,6 @@ Lxy     .req tmp
 Ly2     .req Lh
 LMAP    .req Lx
 ptr     .req tmp
-Ltmp    .req N
-Rtmp    .req N
 
 .global rasterizeF_asm
 rasterizeF_asm:
@@ -57,9 +55,9 @@ rasterizeF_asm:
 
         divLUT tmp, Lh              // tmp = FixedInvU(Lh)
 
-        ldrsh Ltmp, [L, #VERTEX_X]
-        sub Ltmp, Lx, asr #16
-        mul Ldx, tmp, Ltmp          // Ldx = tmp * (N->v.x - L->v.x)
+        ldrsh Ldx, [L, #VERTEX_X]
+        subs Ldx, Lx, asr #16
+        mulne Ldx, tmp, Ldx         // Ldx = tmp * (N->v.x - L->v.x)
     .calc_left_end:
 
     cmp Rh, #0
@@ -81,9 +79,9 @@ rasterizeF_asm:
 
         divLUT tmp, Rh              // tmp = FixedInvU(Rh)
 
-        ldrsh Rtmp, [R, #VERTEX_X]
-        sub Rtmp, Rx, asr #16
-        mul Rdx, tmp, Rtmp          // Rdx = tmp * (N->v.x - Rx)
+        ldrsh Rdx, [R, #VERTEX_X]
+        subs Rdx, Rx, asr #16
+        mulne Rdx, tmp, Rdx         // Rdx = tmp * (N->v.x - Rx)
     .calc_right_end:
 
     cmp Rh, Lh              // if (Rh < Lh)

diff --git a/src/platform/gba/asm/rasterizeFT.s b/src/platform/gba/asm/rasterizeFT.s
@@ -95,12 +95,12 @@ rasterizeFT_asm:
 
         divLUT tmp, Lh              // tmp = FixedInvU(Lh)
 
-        ldrsh Ltmp, [L, #VERTEX_X]
-        sub Ltmp, Lx, asr #16
-        mul Ldx, tmp, Ltmp          // Ldx = tmp * (N->v.x - Lx)
+        ldrsh Ldx, [L, #VERTEX_X]
+        subs Ldx, Lx, asr #16
+        mulne Ldx, tmp, Ldx         // Ldx = tmp * (N->v.x - Lx)
 
         ldr Ldt, [L, #VERTEX_T]
-        sub Ldt, Lt                 // Ldt = N->v.t - Lt
+        subs Ldt, Lt                // Ldt = N->v.t - Lt
         scaleUV Ldt, Ltmp, Ltmp2, tmp
     .calc_left_end:
 
@@ -125,12 +125,12 @@ rasterizeFT_asm:
 
         divLUT tmp, Rh              // tmp = FixedInvU(Rh)
 
-        ldrsh Rtmp, [R, #VERTEX_X]
-        sub Rtmp, Rx, asr #16
-        mul Rdx, tmp, Rtmp          // Rdx = tmp * (N->v.x - Rx)
+        ldrsh Rdx, [R, #VERTEX_X]
+        subs Rdx, Rx, asr #16
+        mulne Rdx, tmp, Rdx         // Rdx = tmp * (N->v.x - Rx)
 
         ldr Rdt, [R, #VERTEX_T]
-        sub Rdt, Rt                 // Rdt = N->v.t - Rt
+        subs Rdt, Rt                // Rdt = N->v.t - Rt
         scaleUV Rdt, Rtmp, Rtmp2, tmp
     .calc_right_end:
 
@@ -153,7 +153,7 @@ rasterizeFT_asm:
 
     divLUT inv, width               // inv = FixedInvU(width)
 
-    sub dtdx, Rt, Lt                // duv = Rt - Lt
+    subs dtdx, Rt, Lt               // duv = Rt - Lt
     scaleUV dtdx, dtmp, dtmp2, inv
 
     mov t, Lt                       // t = Lt

diff --git a/src/platform/gba/asm/rasterizeFTA.s b/src/platform/gba/asm/rasterizeFTA.s
@@ -96,12 +96,12 @@ rasterizeFTA_asm:
 
         divLUT tmp, Lh              // tmp = FixedInvU(Lh)
 
-        ldrsh Ltmp, [L, #VERTEX_X]
-        sub Ltmp, Lx, asr #16
-        mul Ldx, tmp, Ltmp          // Ldx = tmp * (N->v.x - Lx)
+        ldrsh Ldx, [L, #VERTEX_X]
+        subs Ldx, Lx, asr #16
+        mulne Ldx, tmp, Ldx         // Ldx = tmp * (N->v.x - Lx)
 
         ldr Ldt, [L, #VERTEX_T]
-        sub Ldt, Lt                 // Ldt = N->v.t - Lt
+        subs Ldt, Lt                // Ldt = N->v.t - Lt
         scaleUV Ldt, Ltmp, Ltmp2, tmp
     .calc_left_end:
 
@@ -126,12 +126,12 @@ rasterizeFTA_asm:
 
         divLUT tmp, Rh              // tmp = FixedInvU(Rh)
 
-        ldrsh Rtmp, [R, #VERTEX_X]
-        sub Rtmp, Rx, asr #16
-        mul Rdx, tmp, Rtmp          // Rdx = tmp * (N->v.x - Rx)
+        ldrsh Rdx, [R, #VERTEX_X]
+        subs Rdx, Rx, asr #16
+        mulne Rdx, tmp, Rdx         // Rdx = tmp * (N->v.x - Rx)
 
         ldr Rdt, [R, #VERTEX_T]
-        sub Rdt, Rt                 // Rdt = N->v.t - Rt
+        subs Rdt, Rt                // Rdt = N->v.t - Rt
         scaleUV Rdt, Rtmp, Rtmp2, tmp
     .calc_right_end:
 
@@ -154,7 +154,7 @@ rasterizeFTA_asm:
 
     divLUT inv, width               // inv = FixedInvU(width)
 
-    sub dtdx, Rt, Lt                // duv = Rt - Lt
+    subs dtdx, Rt, Lt               // duv = Rt - Lt
     scaleUV dtdx, dtmp, dtmp2, inv
 
     mov t, Lt                       // t = Lt

diff --git a/src/platform/gba/asm/rasterizeGT.s b/src/platform/gba/asm/rasterizeGT.s
@@ -107,17 +107,17 @@ rasterizeGT_asm:
         divLUT tmp, Lh              // tmp = FixedInvU(Lh)
 
         fiq_on
-        ldrsh Ltmp, [N, #VERTEX_X]
-        sub Ltmp, Lx, asr #16
-        mul Ldx, tmp, Ltmp          // Ldx = tmp * (N->v.x - Lx)
+        ldrsh Ldx, [N, #VERTEX_X]
+        subs Ldx, Lx, asr #16
+        mulne Ldx, tmp, Ldx         // Ldx = tmp * (N->v.x - Lx)
 
-        ldrb Ltmp, [N, #VERTEX_G]
-        sub Ltmp, Lg, lsr #(8 + G_EXTRA)
-        mul Ldg, tmp, Ltmp          // Ldg = tmp * (N->v.g - Lg)
+        ldrb Ldg, [N, #VERTEX_G]
+        subs Ldg, Lg, lsr #(8 + G_EXTRA)
+        mulne Ldg, tmp, Ldg         // Ldg = tmp * (N->v.g - Lg)
         asr Ldg, #(8 - G_EXTRA)     // (8 + G_EXTRA)-bit for fractional part
 
         ldr Ldt, [N, #VERTEX_T]
-        sub Ldt, Lt                 // Ldt = N->v.t - Lt
+        subs Ldt, Lt                // Ldt = N->v.t - Lt
         scaleUV Ldt, Ltmp, Ltmp2, tmp
         fiq_off
     .calc_left_end:
@@ -146,17 +146,17 @@ rasterizeGT_asm:
         divLUT tmp, Rh              // tmp = FixedInvU(Rh)
 
         fiq_on
-        ldrsh Rtmp, [N, #VERTEX_X]
-        sub Rtmp, Rx, asr #16
-        mul Rdx, tmp, Rtmp          // Rdx = tmp * (N->v.x - Rx)
+        ldrsh Rdx, [N, #VERTEX_X]
+        subs Rdx, Rx, asr #16
+        mulne Rdx, tmp, Rdx         // Rdx = tmp * (N->v.x - Rx)
 
-        ldrb Rtmp, [N, #VERTEX_G]
-        sub Rtmp, Rg, lsr #(8 + G_EXTRA)
-        mul Rdg, tmp, Rtmp          // Rdg = tmp * (N->v.g - Rg)
+        ldrb Rdg, [N, #VERTEX_G]
+        subs Rdg, Rg, lsr #(8 + G_EXTRA)
+        mulne Rdg, tmp, Rdg         // Rdg = tmp * (N->v.g - Rg)
         asr Rdg, #(8 - G_EXTRA)     // (8 + G_EXTRA)-bit for fractional part
 
         ldr Rdt, [N, #VERTEX_T]
-        sub Rdt, Rt                 // Rdt = N->v.t - Rt
+        subs Rdt, Rt                // Rdt = N->v.t - Rt
         scaleUV Rdt, Rtmp, Rtmp2, tmp
         fiq_off
     .calc_right_end:
@@ -183,12 +183,12 @@ rasterizeGT_asm:
 
     divLUT inv, width               // inv = FixedInvU(width)
 
-    sub dtdx, Rt, Lt                // dtdx = Rt - Lt
+    subs dtdx, Rt, Lt               // dtdx = Rt - Lt
     scaleUV dtdx, dtmp, dtmp2, inv
     // t == Lt (alias)
 
-    sub dgdx, Rg, Lg                // dgdx = Rg - Lg
-    mul dgdx, inv                   // dgdx *= FixedInvU(width)
+    subs dgdx, Rg, Lg               // dgdx = Rg - Lg
+    mulne dgdx, inv                 // dgdx *= FixedInvU(width)
     asr dgdx, #16                   // dgdx >>= 16
     // g == Lg (alias)
 

diff --git a/src/platform/gba/asm/rasterizeGTA.s b/src/platform/gba/asm/rasterizeGTA.s
@@ -106,17 +106,17 @@ rasterizeGTA_asm:
         divLUT tmp, Lh              // tmp = FixedInvU(Lh)
 
         fiq_on
-        ldrsh Ltmp, [N, #VERTEX_X]
-        sub Ltmp, Lx, asr #16
-        mul Ldx, tmp, Ltmp          // Ldx = tmp * (N->v.x - Lx)
+        ldrsh Ldx, [N, #VERTEX_X]
+        subs Ldx, Lx, asr #16
+        mulne Ldx, tmp, Ldx         // Ldx = tmp * (N->v.x - Lx)
 
-        ldrb Ltmp, [N, #VERTEX_G]
-        sub Ltmp, Lg, lsr #8
-        mul Ldg, tmp, Ltmp          // Ldg = tmp * (N->v.g - Lg)
+        ldrb Ldg, [N, #VERTEX_G]
+        subs Ldg, Lg, lsr #8
+        mulne Ldg, tmp, Ldg         // Ldg = tmp * (N->v.g - Lg)
         asr Ldg, #8                 // 8-bit for fractional part
 
         ldr Ldt, [N, #VERTEX_T]
-        sub Ldt, Lt                 // Ldt = N->v.t - Lt
+        subs Ldt, Lt                // Ldt = N->v.t - Lt
         scaleUV Ldt, Ltmp, Ltmp2, tmp
         fiq_off
     .calc_left_end:
@@ -145,17 +145,17 @@ rasterizeGTA_asm:
         divLUT tmp, Rh              // tmp = FixedInvU(Rh)
 
         fiq_on
-        ldrsh Rtmp, [N, #VERTEX_X]
-        sub Rtmp, Rx, asr #16
-        mul Rdx, tmp, Rtmp          // Rdx = tmp * (N->v.x - Rx)
+        ldrsh Rdx, [N, #VERTEX_X]
+        subs Rdx, Rx, asr #16
+        mulne Rdx, tmp, Rdx         // Rdx = tmp * (N->v.x - Rx)
 
-        ldrb Rtmp, [N, #VERTEX_G]
-        sub Rtmp, Rg, lsr #8
-        mul Rdg, tmp, Rtmp          // Rdg = tmp * (N->v.g - Rg)
+        ldrb Rdg, [N, #VERTEX_G]
+        subs Rdg, Rg, lsr #8
+        mulne Rdg, tmp, Rdg         // Rdg = tmp * (N->v.g - Rg)
         asr Rdg, #8                 // 8-bit for fractional part
 
         ldr Rdt, [N, #VERTEX_T]
-        sub Rdt, Rt                 // Rdt = N->v.t - Rt
+        subs Rdt, Rt                // Rdt = N->v.t - Rt
         scaleUV Rdt, Rtmp, Rtmp2, tmp
         fiq_off
     .calc_right_end:
@@ -182,12 +182,12 @@ rasterizeGTA_asm:
 
     divLUT inv, width               // inv = FixedInvU(width)
 
-    sub dtdx, Rt, Lt                // dtdx = Rt - Lt
+    subs dtdx, Rt, Lt               // dtdx = Rt - Lt
     scaleUV dtdx, dtmp, dtmp2, inv
     // t == Lt (alias)
 
-    sub dgdx, Rg, Lg                // dgdx = Rg - Lg
-    mul dgdx, inv                   // dgdx *= FixedInvU(width)
+    subs dgdx, Rg, Lg               // dgdx = Rg - Lg
+    mulne dgdx, inv                 // dgdx *= FixedInvU(width)
     asr dgdx, #16                   // dgdx >>= 16
     // g == Lg (alias)
 

diff --git a/src/platform/gba/asm/rasterizeS.s b/src/platform/gba/asm/rasterizeS.s
@@ -23,8 +23,6 @@ Ry2     .req Rh
 Lxy     .req tmp
 Ly2     .req Lh
 indexB  .req pair
-Ltmp    .req N
-Rtmp    .req N
 
 .global rasterizeS_asm
 rasterizeS_asm:
@@ -52,9 +50,9 @@ rasterizeS_asm:
 
         divLUT tmp, Lh              // tmp = FixedInvU(Lh)
 
-        ldrsh Ltmp, [L, #VERTEX_X]
-        sub Ltmp, Lx, asr #16
-        mul Ldx, tmp, Ltmp          // Ldx = tmp * (N->v.x - Lx)
+        ldrsh Ldx, [L, #VERTEX_X]
+        subs Ldx, Lx, asr #16
+        mulne Ldx, tmp, Ldx         // Ldx = tmp * (N->v.x - Lx)
     .calc_left_end:
 
     cmp Rh, #0
@@ -76,9 +74,9 @@ rasterizeS_asm:
 
         divLUT tmp, Rh              // tmp = FixedInvU(Rh)
 
-        ldrsh Rtmp, [R, #VERTEX_X]
-        sub Rtmp, Rx, asr #16
-        mul Rdx, tmp, Rtmp          // Rdx = tmp * (N->v.x - Rx)
+        ldrsh Rdx, [R, #VERTEX_X]
+        subs Rdx, Rx, asr #16
+        mulne Rdx, tmp, Rdx         // Rdx = tmp * (N->v.x - Rx)
     .calc_right_end:
 
     cmp Rh, Lh              // if (Rh < Lh)