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

Implement all v8x16 v16x8 v32x4 all_true on ia32 and x64. arm and arm64
bailout for now, will be implemented later.

Bug: v8:9909
Change-Id: Icea38c00e33b387191e000b64100251525d8f8da
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2220446
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#68077}

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

@@ -2562,6 +2562,11 @@ void LiftoffAssembler::emit_v32x4_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v32x4_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "v32x4_alltrue");
+}
+
 void LiftoffAssembler::emit_i32x4_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::kLeft, NeonS32, Neon32>(this, dst, lhs, rhs);
@@ -2659,6 +2664,11 @@ void LiftoffAssembler::emit_v16x8_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v16x8_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "v16x8_alltrue");
+}
+
 void LiftoffAssembler::emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::kLeft, NeonS16, Neon16>(this, dst, lhs, rhs);
@@ -2828,6 +2838,11 @@ void LiftoffAssembler::emit_v8x16_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v8x16_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "v8x16_alltrue");
+}
+
 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

@@ -1498,6 +1498,11 @@ void LiftoffAssembler::emit_v32x4_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v32x4_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "v32x4_alltrue");
+}
+
 void LiftoffAssembler::emit_i32x4_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::ShiftDirection::kLeft>(
@@ -1614,6 +1619,11 @@ void LiftoffAssembler::emit_v16x8_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v16x8_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "v16x8_alltrue");
+}
+
 void LiftoffAssembler::emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::ShiftDirection::kLeft>(
@@ -1754,6 +1764,11 @@ void LiftoffAssembler::emit_v8x16_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v8x16_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  bailout(kSimd, "v8x16_alltrue");
+}
+
 void LiftoffAssembler::emit_i8x16_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShift<liftoff::ShiftDirection::kLeft>(

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

@@ -2136,6 +2136,20 @@ inline void EmitAnyTrue(LiftoffAssembler* assm, LiftoffRegister dst,
   assm->cmov(zero, dst.gp(), tmp);
 }
 
+template <void (TurboAssembler::*pcmp)(XMMRegister, XMMRegister)>
+inline void EmitAllTrue(LiftoffAssembler* assm, LiftoffRegister dst,
+                        LiftoffRegister src) {
+  Register tmp =
+      assm->GetUnusedRegister(kGpReg, LiftoffRegList::ForRegs(dst)).gp();
+  XMMRegister tmp_simd = liftoff::kScratchDoubleReg;
+  assm->mov(tmp, Immediate(1));
+  assm->xor_(dst.gp(), dst.gp());
+  assm->Pxor(tmp_simd, tmp_simd);
+  (assm->*pcmp)(tmp_simd, src.fp());
+  assm->Ptest(tmp_simd, tmp_simd);
+  assm->cmov(zero, dst.gp(), tmp);
+}
+
 }  // namespace liftoff
 
 void LiftoffAssembler::LoadTransform(LiftoffRegister dst, Register src_addr,
@@ -2533,6 +2547,11 @@ void LiftoffAssembler::emit_v8x16_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v8x16_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqb>(this, dst, src);
+}
+
 void LiftoffAssembler::emit_i8x16_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   static constexpr RegClass tmp_rc = reg_class_for(ValueType::kI32);
@@ -2766,6 +2785,11 @@ void LiftoffAssembler::emit_v16x8_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v16x8_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqw>(this, dst, src);
+}
+
 void LiftoffAssembler::emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShiftOp<&Assembler::vpsllw, &Assembler::psllw, 4>(this, dst,
@@ -2895,6 +2919,11 @@ void LiftoffAssembler::emit_v32x4_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v32x4_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqd>(this, dst, src);
+}
+
 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

@@ -817,6 +817,7 @@ class LiftoffAssembler : public TurboAssembler {
                                LiftoffRegister src2, LiftoffRegister mask);
   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_shl(LiftoffRegister dst, LiftoffRegister lhs,
                              LiftoffRegister rhs);
   inline void emit_i8x16_shli(LiftoffRegister dst, LiftoffRegister lhs,
@@ -857,6 +858,7 @@ class LiftoffAssembler : public TurboAssembler {
                                LiftoffRegister rhs);
   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_shl(LiftoffRegister dst, LiftoffRegister lhs,
                              LiftoffRegister rhs);
   inline void emit_i16x8_shli(LiftoffRegister dst, LiftoffRegister lhs,
@@ -897,6 +899,7 @@ class LiftoffAssembler : public TurboAssembler {
                                LiftoffRegister rhs);
   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_shl(LiftoffRegister dst, LiftoffRegister lhs,
                              LiftoffRegister rhs);
   inline void emit_i32x4_shli(LiftoffRegister dst, LiftoffRegister lhs,

src/wasm/baseline/liftoff-compiler.cc

@@ -2577,6 +2577,8 @@ class LiftoffCompiler {
         return EmitUnOp<kS128, kS128>(&LiftoffAssembler::emit_i8x16_neg);
       case wasm::kExprV8x16AnyTrue:
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v8x16_anytrue);
+      case wasm::kExprV8x16AllTrue:
+        return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v8x16_alltrue);
       case wasm::kExprI8x16Shl:
         return EmitSimdShiftOp(&LiftoffAssembler::emit_i8x16_shl,
                                &LiftoffAssembler::emit_i8x16_shli);
@@ -2616,6 +2618,8 @@ class LiftoffCompiler {
         return EmitUnOp<kS128, kS128>(&LiftoffAssembler::emit_i16x8_neg);
       case wasm::kExprV16x8AnyTrue:
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v16x8_anytrue);
+      case wasm::kExprV16x8AllTrue:
+        return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v16x8_alltrue);
       case wasm::kExprI16x8Shl:
         return EmitSimdShiftOp(&LiftoffAssembler::emit_i16x8_shl,
                                &LiftoffAssembler::emit_i16x8_shli);
@@ -2655,6 +2659,8 @@ class LiftoffCompiler {
         return EmitUnOp<kS128, kS128>(&LiftoffAssembler::emit_i32x4_neg);
       case wasm::kExprV32x4AnyTrue:
         return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v32x4_anytrue);
+      case wasm::kExprV32x4AllTrue:
+        return EmitUnOp<kS128, kI32>(&LiftoffAssembler::emit_v32x4_alltrue);
       case wasm::kExprI32x4Shl:
         return EmitSimdShiftOp(&LiftoffAssembler::emit_i32x4_shl,
                                &LiftoffAssembler::emit_i32x4_shli);

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

@@ -2178,6 +2178,17 @@ inline void EmitAnyTrue(LiftoffAssembler* assm, LiftoffRegister dst,
   assm->setcc(not_equal, dst.gp());
 }
 
+template <void (TurboAssembler::*pcmp)(XMMRegister, XMMRegister)>
+inline void EmitAllTrue(LiftoffAssembler* assm, LiftoffRegister dst,
+                        LiftoffRegister src) {
+  XMMRegister tmp = kScratchDoubleReg;
+  assm->xorq(dst.gp(), dst.gp());
+  assm->Pxor(tmp, tmp);
+  (assm->*pcmp)(tmp, src.fp());
+  assm->Ptest(tmp, tmp);
+  assm->setcc(equal, dst.gp());
+}
+
 }  // namespace liftoff
 
 void LiftoffAssembler::LoadTransform(LiftoffRegister dst, Register src_addr,
@@ -2575,6 +2586,11 @@ void LiftoffAssembler::emit_v8x16_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v8x16_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqb>(this, dst, src);
+}
+
 void LiftoffAssembler::emit_i8x16_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   static constexpr RegClass tmp_simd_rc = reg_class_for(ValueType::kS128);
@@ -2809,6 +2825,11 @@ void LiftoffAssembler::emit_v16x8_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v16x8_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqw>(this, dst, src);
+}
+
 void LiftoffAssembler::emit_i16x8_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShiftOp<&Assembler::vpsllw, &Assembler::psllw, 4>(this, dst,
@@ -2938,6 +2959,11 @@ void LiftoffAssembler::emit_v32x4_anytrue(LiftoffRegister dst,
   liftoff::EmitAnyTrue(this, dst, src);
 }
 
+void LiftoffAssembler::emit_v32x4_alltrue(LiftoffRegister dst,
+                                          LiftoffRegister src) {
+  liftoff::EmitAllTrue<&TurboAssembler::Pcmpeqd>(this, dst, src);
+}
+
 void LiftoffAssembler::emit_i32x4_shl(LiftoffRegister dst, LiftoffRegister lhs,
                                       LiftoffRegister rhs) {
   liftoff::EmitSimdShiftOp<&Assembler::vpslld, &Assembler::pslld, 5>(this, dst,