add a x*y+0 -> x*y peephole

This is mostly to test how easy rebaselining SkVMTest is.

Change-Id: I27ab6f6bb8b7e126327374009783afd86d416f55
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/218039
Reviewed-by: Brian Osman <brianosman@google.com>

resources/SkVMTest.expected

@@ -16,74 +16,72 @@ r0 = to_i32 r0
 store8 arg(1) r0
 
 A8 over G8
-r0 = splat 0 (0)
-r1 = load8 arg(0)
-r2 = splat 3B808081 (0.0039215689)
-r1 = to_f32 r1
+r0 = load8 arg(0)
+r1 = splat 3B808081 (0.0039215689)
+r0 = to_f32 r0
+r0 = mul_f32 r1 r0
+r2 = load8 arg(1)
+r2 = to_f32 r2
+r2 = mul_f32 r1 r2
+r1 = splat 3F800000 (1)
+r1 = sub_f32 r1 r0
 r1 = mul_f32 r2 r1
-r3 = load8 arg(1)
-r3 = to_f32 r3
-r3 = mul_f32 r2 r3
-r2 = splat 3F800000 (1)
-r2 = sub_f32 r2 r1
-r2 = mad_f32 r3 r2 r0
-r3 = splat 3E59B3D0 (0.21259999)
+r2 = splat 3E59B3D0 (0.21259999)
 r0 = splat 3F371759 (0.71520001)
-r1 = splat 3D93DD98 (0.0722)
-r1 = mul_f32 r2 r1
-r1 = mad_f32 r2 r0 r1
-r1 = mad_f32 r2 r3 r1
-r3 = splat 437F0000 (255)
-r2 = splat 3F000000 (0.5)
-r2 = mad_f32 r1 r3 r2
-r2 = to_i32 r2
-store8 arg(1) r2
+r3 = splat 3D93DD98 (0.0722)
+r3 = mul_f32 r1 r3
+r3 = mad_f32 r1 r0 r3
+r3 = mad_f32 r1 r2 r3
+r2 = splat 437F0000 (255)
+r1 = splat 3F000000 (0.5)
+r1 = mad_f32 r3 r2 r1
+r1 = to_i32 r1
+store8 arg(1) r1
 
 A8 over RGBA_8888
-r0 = splat 0 (0)
-r1 = load8 arg(0)
-r2 = splat 3B808081 (0.0039215689)
-r1 = to_f32 r1
-r1 = mul_f32 r2 r1
-r3 = load32 arg(1)
-r4 = splat FF (3.5733111e-43)
-r5 = bit_and r3 r4
+r0 = load8 arg(0)
+r1 = splat 3B808081 (0.0039215689)
+r0 = to_f32 r0
+r0 = mul_f32 r1 r0
+r2 = load32 arg(1)
+r3 = splat FF (3.5733111e-43)
+r4 = bit_and r2 r3
+r4 = to_f32 r4
+r4 = mul_f32 r1 r4
+r5 = shr r2 8 (1.1210388e-44)
+r5 = bit_and r5 r3
 r5 = to_f32 r5
-r5 = mul_f32 r2 r5
-r6 = shr r3 8 (1.1210388e-44)
-r6 = bit_and r6 r4
+r5 = mul_f32 r1 r5
+r6 = shr r2 10 (2.2420775e-44)
+r6 = bit_and r6 r3
 r6 = to_f32 r6
-r6 = mul_f32 r2 r6
-r7 = shr r3 10 (2.2420775e-44)
-r7 = bit_and r7 r4
-r7 = to_f32 r7
-r7 = mul_f32 r2 r7
-r3 = shr r3 18 (3.3631163e-44)
-r3 = to_f32 r3
-r3 = mul_f32 r2 r3
-r2 = splat 3F800000 (1)
-r2 = sub_f32 r2 r1
+r6 = mul_f32 r1 r6
+r2 = shr r2 18 (3.3631163e-44)
+r2 = to_f32 r2
+r2 = mul_f32 r1 r2
+r1 = splat 3F800000 (1)
+r1 = sub_f32 r1 r0
+r4 = mul_f32 r4 r1
+r5 = mul_f32 r5 r1
+r6 = mul_f32 r6 r1
+r1 = mad_f32 r2 r1 r0
+r2 = splat 437F0000 (255)
+r0 = splat 3F000000 (0.5)
+r4 = mad_f32 r4 r2 r0
+r4 = to_i32 r4
 r5 = mad_f32 r5 r2 r0
-r6 = mad_f32 r6 r2 r0
-r7 = mad_f32 r7 r2 r0
-r2 = mad_f32 r3 r2 r1
-r3 = splat 437F0000 (255)
-r1 = splat 3F000000 (0.5)
-r5 = mad_f32 r5 r3 r1
 r5 = to_i32 r5
-r6 = mad_f32 r6 r3 r1
+r5 = shl r5 8 (1.1210388e-44)
+r6 = mad_f32 r6 r2 r0
 r6 = to_i32 r6
-r6 = shl r6 8 (1.1210388e-44)
-r7 = mad_f32 r7 r3 r1
-r7 = to_i32 r7
-r7 = shl r7 10 (2.2420775e-44)
-r1 = mad_f32 r2 r3 r1
-r1 = to_i32 r1
-r1 = shl r1 18 (3.3631163e-44)
-r6 = bit_or r5 r6
-r6 = bit_or r6 r7
-r6 = bit_or r6 r1
-store32 arg(1) r6
+r6 = shl r6 10 (2.2420775e-44)
+r0 = mad_f32 r1 r2 r0
+r0 = to_i32 r0
+r0 = shl r0 18 (3.3631163e-44)
+r5 = bit_or r4 r5
+r5 = bit_or r5 r6
+r5 = bit_or r5 r0
+store32 arg(1) r5
 
 G8 over A8
 r0 = splat 3B808081 (0.0039215689)

src/core/SkVM.cpp

@@ -114,6 +114,17 @@ namespace skvm {
     ID Builder::push(Op op, ID x=NA, ID y=NA, ID z=NA, int imm=0) {
         Instruction inst{op, /*life=*/NA, x, y, z, imm};
 
+        // Simple peepholes that come up fairly often.
+
+        auto is_zero = [&](ID id) {
+            return fProgram[id].op  == Op::splat
+                && fProgram[id].imm == 0;
+        };
+
+        // x*y+0 --> x*y
+        if (op == Op::mad_f32 && is_zero(z)) { inst = { Op::mul_f32, NA, x,y,NA, 0 }; }
+
+
         // Basic common subexpression elimination:
         // if we've already seen this exact Instruction, use it instead of creating a new one.
         auto lookup = fIndex.find(inst);