[ia32] Unify I32x4SConvertF32x4 SSE and AVX opcodes

Drive-by cleanup: IWYU for macro-assembler-ia32.cc.

IWYU added src/heap/basic-memory-chunk.h which failed a presubmit, so I
updated src/DEPS to allow for including it.

Bug: v8:11217,v8:7490
Change-Id: I63662bfb2b34e354e94f6052edfcb92f1341da58
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2583675Reviewed-by: Bill Budge <bbudge@chromium.org>
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71729}

src/DEPS

@@ -8,6 +8,7 @@ include_rules = [
   "+src/compiler/code-assembler.h",
   "+src/compiler/wasm-compiler.h",
   "-src/heap",
+  "+src/heap/basic-memory-chunk.h",
   "+src/heap/combined-heap.h",
   "+src/heap/embedder-tracing.h",
   "+src/heap/factory.h",

src/codegen/ia32/macro-assembler-ia32.cc

@@ -4,24 +4,58 @@
 
 #if V8_TARGET_ARCH_IA32
 
+#include <stdint.h>
+
+#include "include/v8-internal.h"
 #include "src/base/bits.h"
-#include "src/base/division-by-constant.h"
-#include "src/base/utils/random-number-generator.h"
-#include "src/codegen/callable.h"
+#include "src/base/logging.h"
+#include "src/base/macros.h"
+#include "src/base/platform/platform.h"
+#include "src/builtins/builtins.h"
+#include "src/codegen/assembler.h"
+#include "src/codegen/bailout-reason.h"
 #include "src/codegen/code-factory.h"
 #include "src/codegen/cpu-features.h"
-#include "src/codegen/external-reference-table.h"
-#include "src/codegen/ia32/assembler-ia32-inl.h"
+#include "src/codegen/external-reference.h"
+#include "src/codegen/ia32/assembler-ia32.h"
+#include "src/codegen/ia32/register-ia32.h"
+#include "src/codegen/interface-descriptors.h"
+#include "src/codegen/label.h"
 #include "src/codegen/macro-assembler.h"
-#include "src/debug/debug.h"
+#include "src/codegen/register.h"
+#include "src/codegen/reglist.h"
+#include "src/codegen/reloc-info.h"
+#include "src/codegen/turbo-assembler.h"
+#include "src/common/globals.h"
+#include "src/deoptimizer/deoptimizer.h"
 #include "src/execution/frame-constants.h"
-#include "src/execution/frames-inl.h"
+#include "src/execution/frames.h"
+#include "src/execution/isolate-data.h"
+#include "src/execution/isolate.h"
+#include "src/flags/flags.h"
+#include "src/handles/handles-inl.h"
+#include "src/handles/handles.h"
+#include "src/heap/basic-memory-chunk.h"
+#include "src/heap/factory-inl.h"
+#include "src/heap/factory.h"
 #include "src/heap/memory-chunk.h"
-#include "src/init/bootstrapper.h"
 #include "src/logging/counters.h"
+#include "src/objects/code.h"
+#include "src/objects/contexts.h"
+#include "src/objects/fixed-array.h"
+#include "src/objects/heap-object.h"
+#include "src/objects/js-function.h"
+#include "src/objects/map.h"
+#include "src/objects/objects.h"
+#include "src/objects/oddball.h"
+#include "src/objects/shared-function-info.h"
+#include "src/objects/slots-inl.h"
+#include "src/objects/smi.h"
+#include "src/roots/roots-inl.h"
+#include "src/roots/roots.h"
 #include "src/runtime/runtime.h"
 #include "src/snapshot/embedded/embedded-data.h"
-#include "src/snapshot/snapshot.h"
+#include "src/utils/utils.h"
 
 // Satisfy cpplint check, but don't include platform-specific header. It is
 // included recursively via macro-assembler.h.
@@ -1506,6 +1540,17 @@ void TurboAssembler::Move(XMMRegister dst, uint64_t src) {
   }
 }
 
+void TurboAssembler::Cmpeqps(XMMRegister dst, XMMRegister src1,
+                             XMMRegister src2) {
+  if (CpuFeatures::IsSupported(AVX)) {
+    CpuFeatureScope avx_scope(this, AVX);
+    vcmpeqps(dst, src1, src2);
+  } else {
+    movaps(dst, src1);
+    cmpeqps(dst, src2);
+  }
+}
+
 void TurboAssembler::Pshufhw(XMMRegister dst, Operand src, uint8_t shuffle) {
   if (CpuFeatures::IsSupported(AVX)) {
     CpuFeatureScope scope(this, AVX);

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

@@ -280,6 +280,9 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
   // may be bigger than 2^16 - 1.  Requires a scratch register.
   void Ret(int bytes_dropped, Register scratch);
 
+  // Three-operand cmpeqps that moves src1 to dst if AVX is not supported.
+  void Cmpeqps(XMMRegister dst, XMMRegister src1, XMMRegister src2);
+
   void Pshufhw(XMMRegister dst, XMMRegister src, uint8_t shuffle) {
     Pshufhw(dst, Operand(src), shuffle);
   }
@@ -435,6 +438,7 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
   AVX_PACKED_OP3(Pmuludq, pmuludq)
   AVX_PACKED_OP3(Pavgb, pavgb)
   AVX_PACKED_OP3(Pavgw, pavgw)
+  AVX_PACKED_OP3(Pand, pand)
 #undef AVX_PACKED_OP3
 
   AVX_PACKED_OP3_WITH_TYPE(Psllw, psllw, XMMRegister, uint8_t)

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

@@ -2640,40 +2640,21 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ Pextrd(i.OutputRegister(), i.InputSimd128Register(0), i.InputInt8(1));
       break;
     }
-    case kSSEI32x4SConvertF32x4: {
-      DCHECK_EQ(i.OutputSimd128Register(), i.InputSimd128Register(0));
-      XMMRegister dst = i.OutputSimd128Register();
-      // NAN->0
-      __ movaps(kScratchDoubleReg, dst);
-      __ cmpeqps(kScratchDoubleReg, kScratchDoubleReg);
-      __ pand(dst, kScratchDoubleReg);
-      // Set top bit if >= 0 (but not -0.0!)
-      __ pxor(kScratchDoubleReg, dst);
-      // Convert
-      __ cvttps2dq(dst, dst);
-      // Set top bit if >=0 is now < 0
-      __ pand(kScratchDoubleReg, dst);
-      __ psrad(kScratchDoubleReg, 31);
-      // Set positive overflow lanes to 0x7FFFFFFF
-      __ pxor(dst, kScratchDoubleReg);
-      break;
-    }
-    case kAVXI32x4SConvertF32x4: {
-      CpuFeatureScope avx_scope(tasm(), AVX);
+    case kIA32I32x4SConvertF32x4: {
       XMMRegister dst = i.OutputSimd128Register();
       XMMRegister src = i.InputSimd128Register(0);
       // NAN->0
-      __ vcmpeqps(kScratchDoubleReg, src, src);
-      __ vpand(dst, src, kScratchDoubleReg);
+      __ Cmpeqps(kScratchDoubleReg, src, src);
+      __ Pand(dst, src, kScratchDoubleReg);
       // Set top bit if >= 0 (but not -0.0!)
-      __ vpxor(kScratchDoubleReg, kScratchDoubleReg, dst);
+      __ Pxor(kScratchDoubleReg, dst);
       // Convert
-      __ vcvttps2dq(dst, dst);
+      __ Cvttps2dq(dst, dst);
       // Set top bit if >=0 is now < 0
-      __ vpand(kScratchDoubleReg, kScratchDoubleReg, dst);
-      __ vpsrad(kScratchDoubleReg, kScratchDoubleReg, 31);
+      __ Pand(kScratchDoubleReg, dst);
+      __ Psrad(kScratchDoubleReg, kScratchDoubleReg, 31);
       // Set positive overflow lanes to 0x7FFFFFFF
-      __ vpxor(dst, dst, kScratchDoubleReg);
+      __ Pxor(dst, kScratchDoubleReg);
       break;
     }
     case kIA32I32x4SConvertI16x8Low: {

src/compiler/backend/ia32/instruction-codes-ia32.h

@@ -194,8 +194,7 @@ namespace compiler {
   V(IA32F32x4Round)                \
   V(IA32I32x4Splat)                \
   V(IA32I32x4ExtractLane)          \
-  V(SSEI32x4SConvertF32x4)         \
-  V(AVXI32x4SConvertF32x4)         \
+  V(IA32I32x4SConvertF32x4)        \
   V(IA32I32x4SConvertI16x8Low)     \
   V(IA32I32x4SConvertI16x8High)    \
   V(IA32I32x4Neg)                  \

src/compiler/backend/ia32/instruction-scheduler-ia32.cc

@@ -173,8 +173,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kIA32F32x4Round:
     case kIA32I32x4Splat:
     case kIA32I32x4ExtractLane:
-    case kSSEI32x4SConvertF32x4:
-    case kAVXI32x4SConvertF32x4:
+    case kIA32I32x4SConvertF32x4:
     case kIA32I32x4SConvertI16x8Low:
     case kIA32I32x4SConvertI16x8High:
     case kIA32I32x4Neg:

src/compiler/backend/ia32/instruction-selector-ia32.cc

@@ -2439,7 +2439,7 @@ void InstructionSelector::VisitF32x4UConvertI32x4(Node* node) {
 }
 
 void InstructionSelector::VisitI32x4SConvertF32x4(Node* node) {
-  VisitRRSimd(this, node, kAVXI32x4SConvertF32x4, kSSEI32x4SConvertF32x4);
+  VisitRRSimd(this, node, kIA32I32x4SConvertF32x4);
 }
 
 void InstructionSelector::VisitI32x4UConvertF32x4(Node* node) {