[wasm-simd][liftoff][ia32][x64] Implement bitmask

Implements i8x16 i16x8 i32x4 bitmask.

This was merged into the proposal in
https://github.com/WebAssembly/simd/pull/201/.

Bug: v8:9909,v8:10308
Change-Id: I882f0c2697213cdf593e745112e0897cee252009
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2222607
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#68090}

src/wasm/baseline/arm/liftoff-assembler-arm.h

@@ -2573,6 +2573,11 @@ void LiftoffAssembler::emit_v32x4_alltrue(LiftoffRegister dst,
   mov(dst.gp(), Operand(1), LeaveCC, ne);
 }
 
+void LiftoffAssembler::emit_i32x4_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "i32x4_bitmask");
+}
+
 void LiftoffAssembler::emit_i32x4_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::kLeft, NeonS32, Neon32>(this, dst, lhs, rhs);
@@ -2682,6 +2687,11 @@ void LiftoffAssembler::emit_v16x8_alltrue(LiftoffRegister dst,
   mov(dst.gp(), Operand(1), LeaveCC, ne);
 }
 
+void LiftoffAssembler::emit_i16x8_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "i16x8_bitmask");
+}
+
 void LiftoffAssembler::emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::kLeft, NeonS16, Neon16>(this, dst, lhs, rhs);
@@ -2864,6 +2874,11 @@ void LiftoffAssembler::emit_v8x16_alltrue(LiftoffRegister dst,
   mov(dst.gp(), Operand(1), LeaveCC, ne);
 }
 
+void LiftoffAssembler::emit_i8x16_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "i8x16_bitmask");
+}
+
 void LiftoffAssembler::emit_i8x16_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::kLeft, NeonS8, Neon8>(this, dst, lhs, rhs);

src/wasm/baseline/arm64/liftoff-assembler-arm64.h

@@ -1513,6 +1513,11 @@ void LiftoffAssembler::emit_v32x4_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue(this, dst, src, kFormat4S);
 }
 
+void LiftoffAssembler::emit_i32x4_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "i32x4_bitmask");
+}
+
 void LiftoffAssembler::emit_i32x4_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::ShiftDirection::kLeft>(
@@ -1634,6 +1639,11 @@ void LiftoffAssembler::emit_v16x8_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue(this, dst, src, kFormat8H);
 }
 
+void LiftoffAssembler::emit_i16x8_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "i16x8_bitmask");
+}
+
 void LiftoffAssembler::emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::ShiftDirection::kLeft>(
@@ -1779,6 +1789,11 @@ void LiftoffAssembler::emit_v8x16_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue(this, dst, src, kFormat16B);
 }
 
+void LiftoffAssembler::emit_i8x16_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "i8x16_bitmask");
+}
+
 void LiftoffAssembler::emit_i8x16_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::ShiftDirection::kLeft>(

src/wasm/baseline/ia32/liftoff-assembler-ia32.h

@@ -2552,6 +2552,11 @@ void LiftoffAssembler::emit_v8x16_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqb>(this, dst, src);
 }
 
+void LiftoffAssembler::emit_i8x16_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  Pmovmskb(dst.gp(), src.fp());
+}
+
 void LiftoffAssembler::emit_i8x16_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   static constexpr RegClass tmp_rc = reg_class_for(ValueType::kI32);
@@ -2790,6 +2795,14 @@ void LiftoffAssembler::emit_v16x8_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqw>(this, dst, src);
 }
 
+void LiftoffAssembler::emit_i16x8_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  XMMRegister tmp = liftoff::kScratchDoubleReg;
+  Packsswb(tmp, src.fp());
+  Pmovmskb(dst.gp(), tmp);
+  shr(dst.gp(), 8);
+}
+
 void LiftoffAssembler::emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShiftOp<&Assembler::vpsllw, &Assembler::psllw, 4>(this, dst,
@@ -2924,6 +2937,11 @@ void LiftoffAssembler::emit_v32x4_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqd>(this, dst, src);
 }
 
+void LiftoffAssembler::emit_i32x4_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  Movmskps(dst.gp(), src.fp());
+}
+
 void LiftoffAssembler::emit_i32x4_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShiftOp<&Assembler::vpslld, &Assembler::pslld, 5>(this, dst,

src/wasm/baseline/liftoff-assembler.h

@@ -818,6 +818,7 @@ class LiftoffAssembler : public TurboAssembler {
   inline void emit_i8x16_neg(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_v8x16_anytrue(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_v8x16_alltrue(LiftoffRegister dst, LiftoffRegister src);
+  inline void emit_i8x16_bitmask(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_i8x16_shl(LiftoffRegister dst, LiftoffRegister lhs,
                              LiftoffRegister rhs);
   inline void emit_i8x16_shli(LiftoffRegister dst, LiftoffRegister lhs,
@@ -859,6 +860,7 @@ class LiftoffAssembler : public TurboAssembler {
   inline void emit_i16x8_neg(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_v16x8_anytrue(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_v16x8_alltrue(LiftoffRegister dst, LiftoffRegister src);
+  inline void emit_i16x8_bitmask(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                              LiftoffRegister rhs);
   inline void emit_i16x8_shli(LiftoffRegister dst, LiftoffRegister lhs,
@@ -900,6 +902,7 @@ class LiftoffAssembler : public TurboAssembler {
   inline void emit_i32x4_neg(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_v32x4_anytrue(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_v32x4_alltrue(LiftoffRegister dst, LiftoffRegister src);
+  inline void emit_i32x4_bitmask(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_i32x4_shl(LiftoffRegister dst, LiftoffRegister lhs,
                              LiftoffRegister rhs);
   inline void emit_i32x4_shli(LiftoffRegister dst, LiftoffRegister lhs,

src/wasm/baseline/liftoff-compiler.cc

@@ -2579,6 +2579,8 @@ class LiftoffCompiler {
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v8x16_anytrue);
       case wasm::kExprV8x16AllTrue:
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v8x16_alltrue);
+      case wasm::kExprI8x16BitMask:
+        return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_i8x16_bitmask);
       case wasm::kExprI8x16Shl:
         return EmitSimdShiftOp(&LiftoffAssembler::emit_i8x16_shl,
                                &LiftoffAssembler::emit_i8x16_shli);
@@ -2620,6 +2622,8 @@ class LiftoffCompiler {
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v16x8_anytrue);
       case wasm::kExprV16x8AllTrue:
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v16x8_alltrue);
+      case wasm::kExprI16x8BitMask:
+        return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_i16x8_bitmask);
       case wasm::kExprI16x8Shl:
         return EmitSimdShiftOp(&LiftoffAssembler::emit_i16x8_shl,
                                &LiftoffAssembler::emit_i16x8_shli);
@@ -2661,6 +2665,8 @@ class LiftoffCompiler {
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v32x4_anytrue);
       case wasm::kExprV32x4AllTrue:
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v32x4_alltrue);
+      case wasm::kExprI32x4BitMask:
+        return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_i32x4_bitmask);
       case wasm::kExprI32x4Shl:
         return EmitSimdShiftOp(&LiftoffAssembler::emit_i32x4_shl,
                                &LiftoffAssembler::emit_i32x4_shli);

src/wasm/baseline/x64/liftoff-assembler-x64.h

@@ -2591,6 +2591,11 @@ void LiftoffAssembler::emit_v8x16_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqb>(this, dst, src);
 }
 
+void LiftoffAssembler::emit_i8x16_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  Pmovmskb(dst.gp(), src.fp());
+}
+
 void LiftoffAssembler::emit_i8x16_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   static constexpr RegClass tmp_simd_rc = reg_class_for(ValueType::kS128);
@@ -2830,6 +2835,14 @@ void LiftoffAssembler::emit_v16x8_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqw>(this, dst, src);
 }
 
+void LiftoffAssembler::emit_i16x8_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  XMMRegister tmp = kScratchDoubleReg;
+  Packsswb(tmp, src.fp());
+  Pmovmskb(dst.gp(), tmp);
+  shrq(dst.gp(), Immediate(8));
+}
+
 void LiftoffAssembler::emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShiftOp<&Assembler::vpsllw, &Assembler::psllw, 4>(this, dst,
@@ -2964,6 +2977,11 @@ void LiftoffAssembler::emit_v32x4_alltrue(LiftoffRegister dst,
   liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqd>(this, dst, src);
 }
 
+void LiftoffAssembler::emit_i32x4_bitmask(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  Movmskps(dst.gp(), src.fp());
+}
+
 void LiftoffAssembler::emit_i32x4_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShiftOp<&Assembler::vpslld, &Assembler::pslld, 5>(this, dst,