Reland "[wasm-simd][arm] Prototype i8x16.popcnt"

This is a reland of e2aa734a

The original change was speculatively reverted in
https://crrev.com/c/2568925 but seems unrelated to the flakey
test failures (which are on x64).

Original change's description:
> [wasm-simd][arm] Prototype i8x16.popcnt
>
> Bug: v8:11002
> Change-Id: Ib97e51ed52249a1af7a4b879396b70a016991719
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2567534
> Commit-Queue: Zhi An Ng <zhin@chromium.org>
> Reviewed-by: Bill Budge <bbudge@chromium.org>
> Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#71552}

Bug: v8:11002
Change-Id: I714918e2640d27d29953f940e3bedeb9ea2f8a08
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2567243
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71585}

src/codegen/arm/assembler-arm.cc

@@ -4799,7 +4799,7 @@ void Assembler::vext(QwNeonRegister dst, QwNeonRegister src1,
        n * B7 | B6 | m * B5 | vm);
 }
 
-enum NeonSizedOp { VZIP, VUZP, VREV16, VREV32, VREV64, VTRN };
+enum NeonSizedOp { VZIP, VUZP, VREV16, VREV32, VREV64, VTRN, VCNT };
 
 static Instr EncodeNeonSizedOp(NeonSizedOp op, NeonRegType reg_type,
                                NeonSize size, int dst_code, int src_code) {
@@ -4823,6 +4823,9 @@ static Instr EncodeNeonSizedOp(NeonSizedOp op, NeonRegType reg_type,
     case VTRN:
       op_encoding = 0x2 * B16 | B7;
       break;
+    case VCNT:
+      op_encoding = 0xA * B7;
+      break;
     default:
       UNREACHABLE();
   }
@@ -4907,6 +4910,13 @@ void Assembler::vtrn(NeonSize size, QwNeonRegister src1, QwNeonRegister src2) {
   emit(EncodeNeonSizedOp(VTRN, NEON_Q, size, src1.code(), src2.code()));
 }
 
+void Assembler::vcnt(QwNeonRegister dst, QwNeonRegister src) {
+  // Qd = vcnt(Qm) SIMD Vector Count Set Bits.
+  // Instruction details available at ARM DDI 0487F.b, F6-5094.
+  DCHECK(IsEnabled(NEON));
+  emit(EncodeNeonSizedOp(VCNT, NEON_Q, Neon8, dst.code(), src.code()));
+}
+
 // Encode NEON vtbl / vtbx instruction.
 static Instr EncodeNeonVTB(DwVfpRegister dst, const NeonListOperand& list,
                            DwVfpRegister index, bool vtbx) {

src/codegen/arm/assembler-arm.h

@@ -966,6 +966,8 @@ class V8_EXPORT_PRIVATE Assembler : public AssemblerBase {
   void vtbx(DwVfpRegister dst, const NeonListOperand& list,
             DwVfpRegister index);
 
+  void vcnt(QwNeonRegister dst, QwNeonRegister src);
+
   // Pseudo instructions
 
   // Different nop operations are used by the code generator to detect certain

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

@@ -1638,6 +1638,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
               i.InputDoubleRegister(0));
       DCHECK_EQ(LeaveCC, i.OutputSBit());
       break;
+    case kArmVcnt: {
+      __ vcnt(i.OutputSimd128Register(), i.InputSimd128Register(0));
+      break;
+    }
     case kArmLdrb:
       __ ldrb(i.OutputRegister(), i.InputOffset());
       DCHECK_EQ(LeaveCC, i.OutputSBit());

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

@@ -110,6 +110,7 @@ namespace compiler {
   V(ArmVst1F64)                    \
   V(ArmVld1S128)                   \
   V(ArmVst1S128)                   \
+  V(ArmVcnt)                       \
   V(ArmFloat32Max)                 \
   V(ArmFloat64Max)                 \
   V(ArmFloat32Min)                 \

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

@@ -103,6 +103,7 @@ int InstructionScheduler::GetTargetInstructionFlags(
     case kArmVmovHighF64U32:
     case kArmVmovF64U32U32:
     case kArmVmovU32U32F64:
+    case kArmVcnt:
     case kArmFloat32Max:
     case kArmFloat64Max:
     case kArmFloat32Min:

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

@@ -2540,6 +2540,7 @@ void InstructionSelector::VisitWord32AtomicPairCompareExchange(Node* node) {
   V(I16x8Abs, kArmI16x8Abs)                             \
   V(I8x16Neg, kArmI8x16Neg)                             \
   V(I8x16Abs, kArmI8x16Abs)                             \
+  V(I8x16Popcnt, kArmVcnt)                              \
   V(S128Not, kArmS128Not)                               \
   V(V32x4AnyTrue, kArmV32x4AnyTrue)                     \
   V(V32x4AllTrue, kArmV32x4AllTrue)                     \

src/compiler/backend/instruction-selector.cc

@@ -2737,6 +2737,11 @@ void InstructionSelector::VisitI64x2Eq(Node* node) { UNIMPLEMENTED(); }
 #endif  // !V8_TARGET_ARCH_X64 && !V8_TARGET_ARCH_S390X && !V8_TARGET_ARCH_ARM64
         // && !V8_TARGET_ARCH_IA32
 
+#if !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_ARM
+// TODO(v8:11002) Prototype i8x16.popcnt.
+void InstructionSelector::VisitI8x16Popcnt(Node* node) { UNIMPLEMENTED(); }
+#endif  // !V8_TARGET_ARCH_ARM64 && !V8_TARGET_ARCH_ARM
+
 #if !V8_TARGET_ARCH_ARM64
 // TODO(v8:10971) Prototype i16x8.q15mulr_sat_s
 void InstructionSelector::VisitI16x8Q15MulRSatS(Node* node) { UNIMPLEMENTED(); }
@@ -2758,9 +2763,6 @@ void InstructionSelector::VisitI64x2UConvertI32x4High(Node* node) {
   UNIMPLEMENTED();
 }
 
-// TODO(v8:11002) Prototype i8x16.popcnt.
-void InstructionSelector::VisitI8x16Popcnt(Node* node) { UNIMPLEMENTED(); }
-
 // TODO(v8:11008) Prototype extended multiplication.
 void InstructionSelector::VisitI64x2ExtMulLowI32x4S(Node* node) {
   UNIMPLEMENTED();

src/diagnostics/arm/disasm-arm.cc

@@ -2268,6 +2268,9 @@ void Decoder::DecodeAdvancedSIMDTwoOrThreeRegisters(Instruction* instr) {
                                   "vrev%d.%d q%d, q%d", op, esize, Vd, Vm);
     } else if (size == 0 && opc1 == 0b10 && opc2 == 0) {
       Format(instr, q ? "vswp 'Qd, 'Qm" : "vswp 'Dd, 'Dm");
+    } else if (opc1 == 0 && opc2 == 0b1010) {
+      DCHECK_EQ(0, size);
+      Format(instr, q ? "vcnt.8 'Qd, 'Qm" : "vcnt.8 'Dd, 'Dm");
     } else if (opc1 == 0 && opc2 == 0b1011) {
       Format(instr, "vmvn 'Qd, 'Qm");
     } else if (opc1 == 0b01 && (opc2 & 0b0111) == 0b110) {

src/execution/arm/simulator-arm.cc

@@ -4461,6 +4461,17 @@ void Simulator::DecodeAdvancedSIMDTwoOrThreeRegisters(Instruction* instr) {
         set_neon_register(vm, dval);
         set_neon_register(vd, mval);
       }
+    } else if (opc1 == 0 && opc2 == 0b1010) {
+      // vcnt Qd, Qm.
+      DCHECK_EQ(0, size);
+      int vd = instr->VFPDRegValue(q ? kSimd128Precision : kDoublePrecision);
+      int vm = instr->VFPMRegValue(q ? kSimd128Precision : kDoublePrecision);
+      uint8_t q_data[16];
+      get_neon_register(vm, q_data);
+      for (int i = 0; i < 16; i++) {
+        q_data[i] = base::bits::CountPopulation(q_data[i]);
+      }
+      set_neon_register(vd, q_data);
     } else if (opc1 == 0 && opc2 == 0b1011) {
       // vmvn Qd, Qm.
       int vd = instr->VFPDRegValue(kSimd128Precision);

test/cctest/test-disasm-arm.cc

@@ -1323,6 +1323,8 @@ TEST(Neon) {
               "f3b1fa45       vtbx.8 d15, {d1, d2, d3}, d5");
       COMPARE(vtbx(d15, NeonListOperand(d1, 4), d5),
               "f3b1fb45       vtbx.8 d15, {d1, d2, d3, d4}, d5");
+      COMPARE(vcnt(q1, q2),
+              "f3b02544       vcnt.8 q1, q2");
   }
 
   VERIFY_RUN();

test/cctest/wasm/test-run-wasm-simd.cc

@@ -2545,7 +2545,7 @@ WASM_SIMD_TEST(I8x16Abs) {
   RunI8x16UnOpTest(execution_tier, lower_simd, kExprI8x16Abs, Abs);
 }
 
-#if V8_TARGET_ARCH_ARM64
+#if V8_TARGET_ARCH_ARM64 || V8_TARGET_ARCH_ARM
 // TODO(v8:11002) Prototype i8x16.popcnt.
 WASM_SIMD_TEST_NO_LOWERING(I8x16Popcnt) {
   FLAG_SCOPE(wasm_simd_post_mvp);
@@ -2568,7 +2568,7 @@ WASM_SIMD_TEST_NO_LOWERING(I8x16Popcnt) {
     }
   }
 }
-#endif  // V8_TARGET_ARCH_ARM64
+#endif  // V8_TARGET_ARCH_ARM64 || V8_TARGET_ARCH_ARM
 
 // Tests both signed and unsigned conversion from I16x8 (packing).
 WASM_SIMD_TEST(I8x16ConvertI16x8) {