deps: V8: cherry-pick ed3eedae33d0

Original commit message:

    Merged: [ia32][wasm-simd] Fix aligned moves in codegen

    For SIMD instructions that use aligned moves (like movaps or movapd), we
    don't have correct memory alignment for SIMD moves yet. Switch to to
    movupd.

    [email protected],[email protected]
    Bug: v8:9198
    Bug: v8:10831
    Bug: chromium:1134039
    (cherry picked from commit ab23ff3c0eed141361365241d13e3211efd608cf)

    Change-Id: Icc038b4a32364b8bc66b723403ccc11f954b080d
    No-Try: true
    No-Presubmit: true
    No-Tree-Checks: true
    Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2469600
    Commit-Queue: Zhi An Ng <[email protected]>
    Reviewed-by: Zhi An Ng <[email protected]>
    Cr-Commit-Position: refs/branch-heads/8.6@{#30}
    Cr-Branched-From: a64aed2333abf49e494d2a5ce24bbd14fff19f60-refs/heads/8.6.395@{#1}
    Cr-Branched-From: a626bc036236c9bf92ac7b87dc40c9e538b087e3-refs/heads/master@{#69472}

Refs: v8/v8@ed3eeda

PR-URL: #38275
Reviewed-By: Matteo Collina <[email protected]>
Reviewed-By: Jiawen Geng <[email protected]>
Reviewed-By: Shelley Vohr <[email protected]>

common.gypi

@@ -36,7 +36,7 @@
 
     # Reset this number to 0 on major V8 upgrades.
     # Increment by one for each non-official patch applied to deps/v8.
-    'v8_embedder_string': '-node.35',
+    'v8_embedder_string': '-node.36',
 
     ##### V8 defaults for Node.js #####
 

deps/v8/src/codegen/ia32/assembler-ia32.h

@@ -959,6 +959,9 @@ class V8_EXPORT_PRIVATE Assembler : public AssemblerBase {
   void movapd(XMMRegister dst, Operand src) {
     sse2_instr(dst, src, 0x66, 0x0F, 0x28);
   }
+  void movupd(XMMRegister dst, Operand src) {
+    sse2_instr(dst, src, 0x66, 0x0F, 0x10);
+  }
 
   void movmskpd(Register dst, XMMRegister src);
   void movmskps(Register dst, XMMRegister src);
@@ -1331,6 +1334,7 @@ class V8_EXPORT_PRIVATE Assembler : public AssemblerBase {
   void vmovapd(XMMRegister dst, Operand src) { vpd(0x28, dst, xmm0, src); }
   void vmovups(XMMRegister dst, XMMRegister src) { vmovups(dst, Operand(src)); }
   void vmovups(XMMRegister dst, Operand src) { vps(0x10, dst, xmm0, src); }
+  void vmovupd(XMMRegister dst, Operand src) { vpd(0x10, dst, xmm0, src); }
   void vshufps(XMMRegister dst, XMMRegister src1, XMMRegister src2, byte imm8) {
     vshufps(dst, src1, Operand(src2), imm8);
   }

deps/v8/src/codegen/ia32/macro-assembler-ia32.h

@@ -292,6 +292,7 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
   AVX_OP2_WITH_TYPE(Movaps, movaps, XMMRegister, XMMRegister)
   AVX_OP2_WITH_TYPE(Movapd, movapd, XMMRegister, XMMRegister)
   AVX_OP2_WITH_TYPE(Movapd, movapd, XMMRegister, const Operand&)
+  AVX_OP2_WITH_TYPE(Movupd, movupd, XMMRegister, const Operand&)
   AVX_OP2_WITH_TYPE(Pmovmskb, pmovmskb, Register, XMMRegister)
   AVX_OP2_WITH_TYPE(Movmskps, movmskps, Register, XMMRegister)
 

deps/v8/src/compiler/backend/ia32/code-generator-ia32.cc

@@ -1966,7 +1966,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
                   tmp = i.TempSimd128Register(0);
       // The minpd instruction doesn't propagate NaNs and +0's in its first
       // operand. Perform minpd in both orders, merge the resuls, and adjust.
-      __ Movapd(tmp, src1);
+      __ Movupd(tmp, src1);
       __ Minpd(tmp, tmp, src);
       __ Minpd(dst, src, src1);
       // propagate -0's and NaNs, which may be non-canonical.
@@ -1985,7 +1985,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
                   tmp = i.TempSimd128Register(0);
       // The maxpd instruction doesn't propagate NaNs and +0's in its first
       // operand. Perform maxpd in both orders, merge the resuls, and adjust.
-      __ Movapd(tmp, src1);
+      __ Movupd(tmp, src1);
       __ Maxpd(tmp, tmp, src);
       __ Maxpd(dst, src, src1);
       // Find discrepancies.
@@ -2375,7 +2375,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       XMMRegister dst = i.OutputSimd128Register();
       Operand src1 = i.InputOperand(1);
       // See comment above for correction of maxps.
-      __ movaps(kScratchDoubleReg, src1);
+      __ vmovups(kScratchDoubleReg, src1);
       __ vmaxps(kScratchDoubleReg, kScratchDoubleReg, dst);
       __ vmaxps(dst, dst, src1);
       __ vxorps(dst, dst, kScratchDoubleReg);

deps/v8/src/diagnostics/ia32/disasm-ia32.cc

@@ -1161,6 +1161,10 @@ int DisassemblerIA32::AVXInstruction(byte* data) {
     int mod, regop, rm, vvvv = vex_vreg();
     get_modrm(*current, &mod, &regop, &rm);
     switch (opcode) {
+      case 0x10:
+        AppendToBuffer("vmovupd %s,", NameOfXMMRegister(regop));
+        current += PrintRightXMMOperand(current);
+        break;
       case 0x28:
         AppendToBuffer("vmovapd %s,", NameOfXMMRegister(regop));
         current += PrintRightXMMOperand(current);
@@ -2090,7 +2094,13 @@ int DisassemblerIA32::InstructionDecode(v8::internal::Vector<char> out_buffer,
           data += 2;
         } else if (*data == 0x0F) {
           data++;
-          if (*data == 0x28) {
+          if (*data == 0x10) {
+            data++;
+            int mod, regop, rm;
+            get_modrm(*data, &mod, &regop, &rm);
+            AppendToBuffer("movupd %s,", NameOfXMMRegister(regop));
+            data += PrintRightXMMOperand(data);
+          } else if (*data == 0x28) {
             data++;
             int mod, regop, rm;
             get_modrm(*data, &mod, &regop, &rm);

deps/v8/test/cctest/test-disasm-ia32.cc

@@ -473,6 +473,7 @@ TEST(DisasmIa320) {
 
     __ movapd(xmm0, xmm1);
     __ movapd(xmm0, Operand(edx, 4));
+    __ movupd(xmm0, Operand(edx, 4));
 
     __ movd(xmm0, edi);
     __ movd(xmm0, Operand(ebx, ecx, times_4, 10000));
@@ -689,6 +690,7 @@ TEST(DisasmIa320) {
       __ vmovaps(xmm0, xmm1);
       __ vmovapd(xmm0, xmm1);
       __ vmovapd(xmm0, Operand(ebx, ecx, times_4, 10000));
+      __ vmovupd(xmm0, Operand(ebx, ecx, times_4, 10000));
       __ vshufps(xmm0, xmm1, xmm2, 3);
       __ vshufps(xmm0, xmm1, Operand(edx, 4), 3);
       __ vhaddps(xmm0, xmm1, xmm2);