[wasm-simd][ia32] Fix codegen for f32x4 min and max

dst might not be the same as src0 (since we don't define them to be
equals in the instruction-selector if AVX is enabled), so the minps
and maxps comparisons were incorrect.

I found this while trying to run some spec tests, so not adding any
unittest, eventually when the spec tests are enabled, this will be
covered.

Bug: v8:10835
Change-Id: I4fbc1dfe949e4137e057e73c0d5dfb8534a00b8f
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2411484Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#69953}

src/compiler/backend/ia32/code-generator-ia32.cc

@@ -2348,7 +2348,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       Operand src1 = i.InputOperand(1);
       // See comment above for correction of minps.
       __ movups(kScratchDoubleReg, src1);
-      __ vminps(kScratchDoubleReg, kScratchDoubleReg, dst);
+      __ vminps(kScratchDoubleReg, kScratchDoubleReg, src0);
       __ vminps(dst, src0, src1);
       __ vorps(dst, dst, kScratchDoubleReg);
       __ vcmpneqps(kScratchDoubleReg, dst, dst);
@@ -2381,11 +2381,12 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     case kAVXF32x4Max: {
       CpuFeatureScope avx_scope(tasm(), AVX);
       XMMRegister dst = i.OutputSimd128Register();
+      XMMRegister src0 = i.InputSimd128Register(0);
       Operand src1 = i.InputOperand(1);
       // See comment above for correction of maxps.
       __ vmovups(kScratchDoubleReg, src1);
-      __ vmaxps(kScratchDoubleReg, kScratchDoubleReg, dst);
-      __ vmaxps(dst, dst, src1);
+      __ vmaxps(kScratchDoubleReg, kScratchDoubleReg, src0);
+      __ vmaxps(dst, src0, src1);
       __ vxorps(dst, dst, kScratchDoubleReg);
       __ vorps(kScratchDoubleReg, kScratchDoubleReg, dst);
       __ vsubps(kScratchDoubleReg, kScratchDoubleReg, dst);

src/diagnostics/ia32/disasm-ia32.cc

@@ -1074,6 +1074,10 @@ int DisassemblerIA32::AVXInstruction(byte* data) {
     int mod, regop, rm, vvvv = vex_vreg();
     get_modrm(*current, &mod, &regop, &rm);
     switch (opcode) {
+      case 0x10:
+        AppendToBuffer("vmovups %s,", NameOfXMMRegister(regop));
+        current += PrintRightXMMOperand(current);
+        break;
       case 0x28:
         AppendToBuffer("vmovaps %s,", NameOfXMMRegister(regop));
         current += PrintRightXMMOperand(current);

test/cctest/test-disasm-ia32.cc

@@ -687,6 +687,8 @@ TEST(DisasmIa320) {
       __ vsqrtps(xmm1, Operand(ebx, ecx, times_4, 10000));
       __ vrsqrtps(xmm1, xmm0);
       __ vrsqrtps(xmm1, Operand(ebx, ecx, times_4, 10000));
+      __ vmovups(xmm0, xmm1);
+      __ vmovups(xmm0, Operand(edx, 4));
       __ vmovaps(xmm0, xmm1);
       __ vmovapd(xmm0, xmm1);
       __ vmovapd(xmm0, Operand(ebx, ecx, times_4, 10000));