[wasm-relaxed-simd] Add liftoff min/max operations on Arm and Arm64

Instruction lowering on ARM uses pmin/pmax as there is no hardware support.
More details here: https://github.com/WebAssembly/relaxed-simd/issues/33

Bug: v8:12284
Change-Id: I075d4a6660afc4418fcc91c5d48e8e9096536dc9
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3646075
Commit-Queue: Deepti Gandluri <gdeepti@chromium.org>
Reviewed-by: Thibaud Michaud <thibaudm@chromium.org>
Cr-Commit-Position: refs/heads/main@{#80621}

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

@@ -2639,6 +2639,18 @@ void LiftoffAssembler::emit_f64x2_pmax(LiftoffRegister dst, LiftoffRegister lhs,
   vmov(dest.high(), right.high(), gt);
 }
 
+void LiftoffAssembler::emit_f64x2_relaxed_min(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  emit_f64x2_pmin(dst, lhs, rhs);
+}
+
+void LiftoffAssembler::emit_f64x2_relaxed_max(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  emit_f64x2_pmax(dst, lhs, rhs);
+}
+
 void LiftoffAssembler::emit_f64x2_convert_low_i32x4_s(LiftoffRegister dst,
                                                       LiftoffRegister src) {
   F64x2ConvertLowI32x4S(liftoff::GetSimd128Register(dst),
@@ -2800,6 +2812,20 @@ void LiftoffAssembler::emit_f32x4_max(LiftoffRegister dst, LiftoffRegister lhs,
        liftoff::GetSimd128Register(rhs));
 }
 
+void LiftoffAssembler::emit_f32x4_relaxed_min(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  vmin(liftoff::GetSimd128Register(dst), liftoff::GetSimd128Register(lhs),
+       liftoff::GetSimd128Register(rhs));
+}
+
+void LiftoffAssembler::emit_f32x4_relaxed_max(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  vmax(liftoff::GetSimd128Register(dst), liftoff::GetSimd128Register(lhs),
+       liftoff::GetSimd128Register(rhs));
+}
+
 void LiftoffAssembler::emit_f32x4_pmin(LiftoffRegister dst, LiftoffRegister lhs,
                                        LiftoffRegister rhs) {
   UseScratchRegisterScope temps(this);

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

@@ -1912,6 +1912,18 @@ void LiftoffAssembler::emit_f64x2_pmax(LiftoffRegister dst, LiftoffRegister lhs,
   }
 }
 
+void LiftoffAssembler::emit_f64x2_relaxed_min(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  Fmin(dst.fp().V2D(), lhs.fp().V2D(), rhs.fp().V2D());
+}
+
+void LiftoffAssembler::emit_f64x2_relaxed_max(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  Fmax(dst.fp().V2D(), lhs.fp().V2D(), rhs.fp().V2D());
+}
+
 void LiftoffAssembler::emit_f64x2_convert_low_i32x4_s(LiftoffRegister dst,
                                                       LiftoffRegister src) {
   Sxtl(dst.fp().V2D(), src.fp().V2S());
@@ -2019,6 +2031,18 @@ void LiftoffAssembler::emit_f32x4_max(LiftoffRegister dst, LiftoffRegister lhs,
   Fmax(dst.fp().V4S(), lhs.fp().V4S(), rhs.fp().V4S());
 }
 
+void LiftoffAssembler::emit_f32x4_relaxed_min(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  Fmin(dst.fp().V4S(), lhs.fp().V4S(), rhs.fp().V4S());
+}
+
+void LiftoffAssembler::emit_f32x4_relaxed_max(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  Fmax(dst.fp().V4S(), lhs.fp().V4S(), rhs.fp().V4S());
+}
+
 void LiftoffAssembler::emit_f32x4_pmin(LiftoffRegister dst, LiftoffRegister lhs,
                                        LiftoffRegister rhs) {
   UseScratchRegisterScope temps(this);

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

@@ -4035,6 +4035,18 @@ void LiftoffAssembler::emit_f32x4_pmax(LiftoffRegister dst, LiftoffRegister lhs,
       this, dst, rhs, lhs);
 }
 
+void LiftoffAssembler::emit_f32x4_relaxed_min(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  bailout(kSimd, "emit_f32x4_relaxed_min");
+}
+
+void LiftoffAssembler::emit_f32x4_relaxed_max(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  bailout(kSimd, "emit_f32x4_relaxed_max");
+}
+
 void LiftoffAssembler::emit_f64x2_abs(LiftoffRegister dst,
                                       LiftoffRegister src) {
   Register tmp = GetUnusedRegister(kGpReg, {}).gp();
@@ -4128,6 +4140,18 @@ void LiftoffAssembler::emit_f64x2_pmax(LiftoffRegister dst, LiftoffRegister lhs,
       this, dst, rhs, lhs);
 }
 
+void LiftoffAssembler::emit_f64x2_relaxed_min(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  bailout(kSimd, "emit_f64x2_relaxed_min");
+}
+
+void LiftoffAssembler::emit_f64x2_relaxed_max(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  bailout(kSimd, "emit_f64x2_relaxed_min");
+}
+
 void LiftoffAssembler::emit_f64x2_convert_low_i32x4_s(LiftoffRegister dst,
                                                       LiftoffRegister src) {
   Cvtdq2pd(dst.fp(), src.fp());

src/wasm/baseline/liftoff-assembler.h

@@ -1362,6 +1362,10 @@ class LiftoffAssembler : public TurboAssembler {
                               LiftoffRegister rhs);
   inline void emit_f32x4_pmax(LiftoffRegister dst, LiftoffRegister lhs,
                               LiftoffRegister rhs);
+  inline void emit_f32x4_relaxed_min(LiftoffRegister dst, LiftoffRegister lhs,
+                                     LiftoffRegister rhs);
+  inline void emit_f32x4_relaxed_max(LiftoffRegister dst, LiftoffRegister lhs,
+                                     LiftoffRegister rhs);
   inline void emit_f64x2_abs(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_f64x2_neg(LiftoffRegister dst, LiftoffRegister src);
   inline void emit_f64x2_sqrt(LiftoffRegister dst, LiftoffRegister src);
@@ -1385,6 +1389,10 @@ class LiftoffAssembler : public TurboAssembler {
                               LiftoffRegister rhs);
   inline void emit_f64x2_pmax(LiftoffRegister dst, LiftoffRegister lhs,
                               LiftoffRegister rhs);
+  inline void emit_f64x2_relaxed_min(LiftoffRegister dst, LiftoffRegister lhs,
+                                     LiftoffRegister rhs);
+  inline void emit_f64x2_relaxed_max(LiftoffRegister dst, LiftoffRegister lhs,
+                                     LiftoffRegister rhs);
   inline void emit_f64x2_convert_low_i32x4_s(LiftoffRegister dst,
                                              LiftoffRegister src);
   inline void emit_f64x2_convert_low_i32x4_u(LiftoffRegister dst,

src/wasm/baseline/liftoff-compiler.cc

@@ -4068,6 +4068,18 @@ class LiftoffCompiler {
       case wasm::kExprI32x4RelaxedLaneSelect:
       case wasm::kExprI64x2RelaxedLaneSelect:
         return EmitRelaxedLaneSelect();
+      case wasm::kExprF32x4RelaxedMin:
+        return EmitBinOp<kS128, kS128, false, kF32>(
+            &LiftoffAssembler::emit_f32x4_relaxed_min);
+      case wasm::kExprF32x4RelaxedMax:
+        return EmitBinOp<kS128, kS128, false, kF32>(
+            &LiftoffAssembler::emit_f32x4_relaxed_max);
+      case wasm::kExprF64x2RelaxedMin:
+        return EmitBinOp<kS128, kS128, false, kF64>(
+            &LiftoffAssembler::emit_f64x2_relaxed_min);
+      case wasm::kExprF64x2RelaxedMax:
+        return EmitBinOp<kS128, kS128, false, kF64>(
+            &LiftoffAssembler::emit_f64x2_relaxed_max);
       case wasm::kExprI16x8RelaxedQ15MulRS:
         return EmitBinOp<kS128, kS128>(
             &LiftoffAssembler::emit_i16x8_relaxed_q15mulr_s);

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

@@ -3598,6 +3598,18 @@ void LiftoffAssembler::emit_f32x4_pmax(LiftoffRegister dst, LiftoffRegister lhs,
       this, dst, rhs, lhs);
 }
 
+void LiftoffAssembler::emit_f32x4_relaxed_min(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  bailout(kSimd, "emit_f32x4_relaxed_min");
+}
+
+void LiftoffAssembler::emit_f32x4_relaxed_max(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  bailout(kSimd, "emit_f32x4_relaxed_max");
+}
+
 void LiftoffAssembler::emit_f64x2_abs(LiftoffRegister dst,
                                       LiftoffRegister src) {
   Abspd(dst.fp(), src.fp(), kScratchRegister);
@@ -3675,6 +3687,18 @@ void LiftoffAssembler::emit_f64x2_max(LiftoffRegister dst, LiftoffRegister lhs,
   F64x2Max(dst.fp(), lhs.fp(), rhs.fp(), kScratchDoubleReg);
 }
 
+void LiftoffAssembler::emit_f64x2_relaxed_min(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  bailout(kSimd, "emit_f64x2_relaxed_min");
+}
+
+void LiftoffAssembler::emit_f64x2_relaxed_max(LiftoffRegister dst,
+                                              LiftoffRegister lhs,
+                                              LiftoffRegister rhs) {
+  bailout(kSimd, "emit_f64x2_relaxed_max");
+}
+
 void LiftoffAssembler::emit_f64x2_pmin(LiftoffRegister dst, LiftoffRegister lhs,
                                        LiftoffRegister rhs) {
   // Due to the way minpd works, pmin(a, b) = minpd(b, a).

test/cctest/cctest.status

@@ -1194,6 +1194,14 @@
   'test-run-wasm-relaxed-simd/RunWasm_F64x2Qfma_liftoff': [SKIP],
   'test-run-wasm-relaxed-simd/RunWasm_F64x2Qfms_liftoff': [SKIP],
   'test-run-wasm-relaxed-simd/RunWasm_I8x16RelaxedSwizzle_liftoff': [SKIP],
+}],
+
+['arch != arm and arch != arm64', {
+  # TODO(12284): Implement relaxed SIMD in Liftoff.
+  'test-run-wasm-relaxed-simd/RunWasm_F32x4RelaxedMin_liftoff': [SKIP],
+  'test-run-wasm-relaxed-simd/RunWasm_F32x4RelaxedMax_liftoff': [SKIP],
+  'test-run-wasm-relaxed-simd/RunWasm_F64x2RelaxedMin_liftoff': [SKIP],
+  'test-run-wasm-relaxed-simd/RunWasm_F64x2RelaxedMax_liftoff': [SKIP],
 }]
 
 ]