[WebAssembly] Remove experimental SIMD instructions

Removes the instruction definitions, intrinsics, and builtins for qfma/qfms,
signselect, and prefetch instructions, which were not included in the final
WebAssembly SIMD spec.

Depends on D98457.

Differential Revision: https://reviews.llvm.org/D98466

clang/include/clang/Basic/BuiltinsWebAssembly.def

@@ -141,11 +141,6 @@ TARGET_BUILTIN(__builtin_wasm_extadd_pairwise_i16x8_u_i32x4, "V4UiV8Us", "nc", "
 
 TARGET_BUILTIN(__builtin_wasm_bitselect, "V4iV4iV4iV4i", "nc", "simd128")
 
-TARGET_BUILTIN(__builtin_wasm_signselect_i8x16, "V16ScV16ScV16ScV16Sc", "nc", "simd128")
-TARGET_BUILTIN(__builtin_wasm_signselect_i16x8, "V8sV8sV8sV8s", "nc", "simd128")
-TARGET_BUILTIN(__builtin_wasm_signselect_i32x4, "V4iV4iV4iV4i", "nc", "simd128")
-TARGET_BUILTIN(__builtin_wasm_signselect_i64x2, "V2LLiV2LLiV2LLiV2LLi", "nc", "simd128")
-
 TARGET_BUILTIN(__builtin_wasm_shuffle_v8x16, "V16ScV16ScV16ScIiIiIiIiIiIiIiIiIiIiIiIiIiIiIiIi", "nc", "simd128")
 
 TARGET_BUILTIN(__builtin_wasm_any_true_i8x16, "iV16Sc", "nc", "simd128")
@@ -188,11 +183,6 @@ TARGET_BUILTIN(__builtin_wasm_dot_s_i32x4_i16x8, "V4iV8sV8s", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_sqrt_f32x4, "V4fV4f", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_sqrt_f64x2, "V2dV2d", "nc", "simd128")
 
-TARGET_BUILTIN(__builtin_wasm_qfma_f32x4, "V4fV4fV4fV4f", "nc", "simd128")
-TARGET_BUILTIN(__builtin_wasm_qfms_f32x4, "V4fV4fV4fV4f", "nc", "simd128")
-TARGET_BUILTIN(__builtin_wasm_qfma_f64x2, "V2dV2dV2dV2d", "nc", "simd128")
-TARGET_BUILTIN(__builtin_wasm_qfms_f64x2, "V2dV2dV2dV2d", "nc", "simd128")
-
 TARGET_BUILTIN(__builtin_wasm_trunc_saturate_s_i32x4_f32x4, "V4iV4f", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_trunc_saturate_u_i32x4_f32x4, "V4iV4f", "nc", "simd128")
 
@@ -206,9 +196,6 @@ TARGET_BUILTIN(__builtin_wasm_widen_high_s_i32x4_i64x2, "V2LLiV4i", "nc", "simd1
 TARGET_BUILTIN(__builtin_wasm_widen_low_u_i32x4_i64x2, "V2LLUiV4Ui", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_widen_high_u_i32x4_i64x2, "V2LLUiV4Ui", "nc", "simd128")
 
-TARGET_BUILTIN(__builtin_wasm_widen_s_i8x16_i32x4, "V4iV16ScIi", "nc", "simd128")
-TARGET_BUILTIN(__builtin_wasm_widen_u_i8x16_i32x4, "V4UiV16UcIi", "nc", "simd128")
-
 TARGET_BUILTIN(__builtin_wasm_convert_low_s_i32x4_f64x2, "V2dV4i", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_convert_low_u_i32x4_f64x2, "V2dV4Ui", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_trunc_saturate_zero_s_f64x2_i32x4, "V4iV2d", "nc", "simd128")
@@ -230,8 +217,5 @@ TARGET_BUILTIN(__builtin_wasm_store64_lane, "vLLi*V2LLiIi", "n", "simd128")
 
 TARGET_BUILTIN(__builtin_wasm_eq_i64x2, "V2LLiV2LLiV2LLi", "nc", "simd128")
 
-TARGET_BUILTIN(__builtin_wasm_prefetch_t, "vv*", "n", "simd128")
-TARGET_BUILTIN(__builtin_wasm_prefetch_nt, "vv*", "n", "simd128")
-
 #undef BUILTIN
 #undef TARGET_BUILTIN

clang/lib/CodeGen/CGBuiltin.cpp

@@ -17366,17 +17366,6 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
         CGM.getIntrinsic(Intrinsic::wasm_bitselect, ConvertType(E->getType()));
     return Builder.CreateCall(Callee, {V1, V2, C});
   }
-  case WebAssembly::BI__builtin_wasm_signselect_i8x16:
-  case WebAssembly::BI__builtin_wasm_signselect_i16x8:
-  case WebAssembly::BI__builtin_wasm_signselect_i32x4:
-  case WebAssembly::BI__builtin_wasm_signselect_i64x2: {
-    Value *V1 = EmitScalarExpr(E->getArg(0));
-    Value *V2 = EmitScalarExpr(E->getArg(1));
-    Value *C = EmitScalarExpr(E->getArg(2));
-    Function *Callee =
-        CGM.getIntrinsic(Intrinsic::wasm_signselect, ConvertType(E->getType()));
-    return Builder.CreateCall(Callee, {V1, V2, C});
-  }
   case WebAssembly::BI__builtin_wasm_dot_s_i32x4_i16x8: {
     Value *LHS = EmitScalarExpr(E->getArg(0));
     Value *RHS = EmitScalarExpr(E->getArg(1));
@@ -17444,29 +17433,6 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
     Function *Callee = CGM.getIntrinsic(Intrinsic::sqrt, Vec->getType());
     return Builder.CreateCall(Callee, {Vec});
   }
-  case WebAssembly::BI__builtin_wasm_qfma_f32x4:
-  case WebAssembly::BI__builtin_wasm_qfms_f32x4:
-  case WebAssembly::BI__builtin_wasm_qfma_f64x2:
-  case WebAssembly::BI__builtin_wasm_qfms_f64x2: {
-    Value *A = EmitScalarExpr(E->getArg(0));
-    Value *B = EmitScalarExpr(E->getArg(1));
-    Value *C = EmitScalarExpr(E->getArg(2));
-    unsigned IntNo;
-    switch (BuiltinID) {
-    case WebAssembly::BI__builtin_wasm_qfma_f32x4:
-    case WebAssembly::BI__builtin_wasm_qfma_f64x2:
-      IntNo = Intrinsic::wasm_qfma;
-      break;
-    case WebAssembly::BI__builtin_wasm_qfms_f32x4:
-    case WebAssembly::BI__builtin_wasm_qfms_f64x2:
-      IntNo = Intrinsic::wasm_qfms;
-      break;
-    default:
-      llvm_unreachable("unexpected builtin ID");
-    }
-    Function *Callee = CGM.getIntrinsic(IntNo, A->getType());
-    return Builder.CreateCall(Callee, {A, B, C});
-  }
   case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8:
   case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8:
   case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4:
@@ -17515,26 +17481,6 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
     Function *Callee = CGM.getIntrinsic(IntNo);
     return Builder.CreateCall(Callee, Vec);
   }
-  case WebAssembly::BI__builtin_wasm_widen_s_i8x16_i32x4:
-  case WebAssembly::BI__builtin_wasm_widen_u_i8x16_i32x4: {
-    Value *Vec = EmitScalarExpr(E->getArg(0));
-    llvm::APSInt SubVecConst =
-        *E->getArg(1)->getIntegerConstantExpr(getContext());
-    Value *SubVec = llvm::ConstantInt::get(getLLVMContext(), SubVecConst);
-    unsigned IntNo;
-    switch (BuiltinID) {
-    case WebAssembly::BI__builtin_wasm_widen_s_i8x16_i32x4:
-      IntNo = Intrinsic::wasm_widen_signed;
-      break;
-    case WebAssembly::BI__builtin_wasm_widen_u_i8x16_i32x4:
-      IntNo = Intrinsic::wasm_widen_unsigned;
-      break;
-    default:
-      llvm_unreachable("unexpected builtin ID");
-    }
-    Function *Callee = CGM.getIntrinsic(IntNo);
-    return Builder.CreateCall(Callee, {Vec, SubVec});
-  }
   case WebAssembly::BI__builtin_wasm_convert_low_s_i32x4_f64x2:
   case WebAssembly::BI__builtin_wasm_convert_low_u_i32x4_f64x2: {
     Value *Vec = EmitScalarExpr(E->getArg(0));
@@ -17649,16 +17595,6 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
     Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_shuffle);
     return Builder.CreateCall(Callee, Ops);
   }
-  case WebAssembly::BI__builtin_wasm_prefetch_t: {
-    Value *Ptr = EmitScalarExpr(E->getArg(0));
-    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_prefetch_t);
-    return Builder.CreateCall(Callee, Ptr);
-  }
-  case WebAssembly::BI__builtin_wasm_prefetch_nt: {
-    Value *Ptr = EmitScalarExpr(E->getArg(0));
-    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_prefetch_nt);
-    return Builder.CreateCall(Callee, Ptr);
-  }
   default:
     return nullptr;
   }

clang/test/CodeGen/builtins-wasm.c

@@ -644,34 +644,6 @@ i32x4 bitselect(i32x4 x, i32x4 y, i32x4 c) {
   // WEBASSEMBLY-NEXT: ret
 }
 
-i8x16 signselect_i8x16(i8x16 x, i8x16 y, i8x16 c) {
-  return __builtin_wasm_signselect_i8x16(x, y, c);
-  // WEBASSEMBLY: call <16 x i8> @llvm.wasm.signselect.v16i8(
-  // WEBASSEMBLY-SAME: <16 x i8> %x, <16 x i8> %y, <16 x i8> %c)
-  // WEBASSEMBLY-NEXT: ret
-}
-
-i16x8 signselect_i16x8(i16x8 x, i16x8 y, i16x8 c) {
-  return __builtin_wasm_signselect_i16x8(x, y, c);
-  // WEBASSEMBLY: call <8 x i16> @llvm.wasm.signselect.v8i16(
-  // WEBASSEMBLY-SAME: <8 x i16> %x, <8 x i16> %y, <8 x i16> %c)
-  // WEBASSEMBLY-NEXT: ret
-}
-
-i32x4 signselect_i32x4(i32x4 x, i32x4 y, i32x4 c) {
-  return __builtin_wasm_signselect_i32x4(x, y, c);
-  // WEBASSEMBLY: call <4 x i32> @llvm.wasm.signselect.v4i32(
-  // WEBASSEMBLY-SAME: <4 x i32> %x, <4 x i32> %y, <4 x i32> %c)
-  // WEBASSEMBLY-NEXT: ret
-}
-
-i64x2 signselect_i64x2(i64x2 x, i64x2 y, i64x2 c) {
-  return __builtin_wasm_signselect_i64x2(x, y, c);
-  // WEBASSEMBLY: call <2 x i64> @llvm.wasm.signselect.v2i64(
-  // WEBASSEMBLY-SAME: <2 x i64> %x, <2 x i64> %y, <2 x i64> %c)
-  // WEBASSEMBLY-NEXT: ret
-}
-
 i8x16 popcnt(i8x16 x) {
   return __builtin_wasm_popcnt_i8x16(x);
   // WEBASSEMBLY: call <16 x i8> @llvm.wasm.popcnt(<16 x i8> %x)
@@ -884,34 +856,6 @@ f64x2 sqrt_f64x2(f64x2 x) {
   // WEBASSEMBLY: ret
 }
 
-f32x4 qfma_f32x4(f32x4 a, f32x4 b, f32x4 c) {
-  return __builtin_wasm_qfma_f32x4(a, b, c);
-  // WEBASSEMBLY: call <4 x float> @llvm.wasm.qfma.v4f32(
-  // WEBASSEMBLY-SAME: <4 x float> %a, <4 x float> %b, <4 x float> %c)
-  // WEBASSEMBLY-NEXT: ret
-}
-
-f32x4 qfms_f32x4(f32x4 a, f32x4 b, f32x4 c) {
-  return __builtin_wasm_qfms_f32x4(a, b, c);
-  // WEBASSEMBLY: call <4 x float> @llvm.wasm.qfms.v4f32(
-  // WEBASSEMBLY-SAME: <4 x float> %a, <4 x float> %b, <4 x float> %c)
-  // WEBASSEMBLY-NEXT: ret
-}
-
-f64x2 qfma_f64x2(f64x2 a, f64x2 b, f64x2 c) {
-  return __builtin_wasm_qfma_f64x2(a, b, c);
-  // WEBASSEMBLY: call <2 x double> @llvm.wasm.qfma.v2f64(
-  // WEBASSEMBLY-SAME: <2 x double> %a, <2 x double> %b, <2 x double> %c)
-  // WEBASSEMBLY-NEXT: ret
-}
-
-f64x2 qfms_f64x2(f64x2 a, f64x2 b, f64x2 c) {
-  return __builtin_wasm_qfms_f64x2(a, b, c);
-  // WEBASSEMBLY: call <2 x double> @llvm.wasm.qfms.v2f64(
-  // WEBASSEMBLY-SAME: <2 x double> %a, <2 x double> %b, <2 x double> %c)
-  // WEBASSEMBLY-NEXT: ret
-}
-
 i32x4 trunc_saturate_s_i32x4_f32x4(f32x4 f) {
   return __builtin_wasm_trunc_saturate_s_i32x4_f32x4(f);
   // WEBASSEMBLY: call <4 x i32> @llvm.wasm.trunc.saturate.signed.v4i32.v4f32(<4 x float> %f)
@@ -976,18 +920,6 @@ u64x2 widen_high_u_i32x4_i64x2(u32x4 x) {
   // WEBASSEMBLY: ret
 }
 
-i32x4 widen_s_i8x16_i32x4(i8x16 x) {
-  return __builtin_wasm_widen_s_i8x16_i32x4(x, 3);
-  // WEBASSEMBLY: call <4 x i32> @llvm.wasm.widen.signed(<16 x i8> %x, i32 3)
-  // WEBASSEMBLY: ret
-}
-
-u32x4 widen_u_i8x16_i32x4(u8x16 x) {
-  return __builtin_wasm_widen_u_i8x16_i32x4(x, 3);
-  // WEBASSEMBLY: call <4 x i32> @llvm.wasm.widen.unsigned(<16 x i8> %x, i32 3)
-  // WEBASSEMBLY: ret
-}
-
 f64x2 convert_low_s_i32x4_f64x2(i32x4 x) {
   return __builtin_wasm_convert_low_s_i32x4_f64x2(x);
   // WEBASSEMBLY: call <2 x double> @llvm.wasm.convert.low.signed(<4 x i32> %x)
@@ -1050,13 +982,3 @@ i8x16 shuffle(i8x16 x, i8x16 y) {
   // WEBASSEMBLY-SAME: i32 15
   // WEBASSEMBLY-NEXT: ret
 }
-
-void prefetch_t(void *p) {
-  return __builtin_wasm_prefetch_t(p);
-  // WEBASSEMBLY: call void @llvm.wasm.prefetch.t(i8* %p)
-}
-
-void prefetch_nt(void *p) {
-  return __builtin_wasm_prefetch_nt(p);
-  // WEBASSEMBLY: call void @llvm.wasm.prefetch.nt(i8* %p)
-}

llvm/include/llvm/IR/IntrinsicsWebAssembly.td

@@ -143,14 +143,6 @@ def int_wasm_bitmask :
   Intrinsic<[llvm_i32_ty],
             [llvm_anyvector_ty],
             [IntrNoMem, IntrSpeculatable]>;
-def int_wasm_qfma :
-  Intrinsic<[llvm_anyvector_ty],
-            [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
-            [IntrNoMem, IntrSpeculatable]>;
-def int_wasm_qfms :
-  Intrinsic<[llvm_anyvector_ty],
-            [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
-            [IntrNoMem, IntrSpeculatable]>;
 def int_wasm_dot :
   Intrinsic<[llvm_v4i32_ty],
             [llvm_v8i16_ty, llvm_v8i16_ty],
@@ -302,32 +294,13 @@ def int_wasm_extadd_pairwise_unsigned :
             [LLVMSubdivide2VectorType<0>],
             [IntrNoMem, IntrSpeculatable]>;
 
-def int_wasm_signselect :
-  Intrinsic<[llvm_anyvector_ty],
-            [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
-            [IntrNoMem, IntrSpeculatable]>;
-
 // TODO: Remove this intrinsic and the associated builtin if i64x2.eq gets
 // merged to the proposal.
 def int_wasm_eq :
   Intrinsic<[llvm_v2i64_ty],
             [llvm_v2i64_ty, llvm_v2i64_ty],
             [IntrNoMem, IntrSpeculatable]>;
 
-// TODO: Remove this after experiments have been run. Use the target-agnostic
-// int_prefetch if this becomes specified at some point.
-def int_wasm_prefetch_t :
-  Intrinsic<[], [llvm_ptr_ty],
-            [IntrInaccessibleMemOrArgMemOnly, IntrWillReturn,
-             ReadOnly<ArgIndex<0>>, NoCapture<ArgIndex<0>>],
-            "", [SDNPMemOperand]>;
-
-def int_wasm_prefetch_nt :
-  Intrinsic<[], [llvm_ptr_ty],
-            [IntrInaccessibleMemOrArgMemOnly, IntrWillReturn,
-             ReadOnly<ArgIndex<0>>, NoCapture<ArgIndex<0>>],
-            "", [SDNPMemOperand]>;
-
 // TODO: Remove these if possible if they are merged to the spec.
 def int_wasm_convert_low_signed :
   Intrinsic<[llvm_v2f64_ty], [llvm_v4i32_ty],
@@ -348,14 +321,6 @@ def int_wasm_promote_low :
   Intrinsic<[llvm_v2f64_ty], [llvm_v4f32_ty],
             [IntrNoMem, IntrSpeculatable]>;
 
-// TODO: Remove these if possible if they are merged to the spec.
-def int_wasm_widen_signed :
-  Intrinsic<[llvm_v4i32_ty], [llvm_v16i8_ty, llvm_i32_ty],
-            [IntrNoMem, IntrSpeculatable, ImmArg<ArgIndex<1>>]>;
-def int_wasm_widen_unsigned :
-  Intrinsic<[llvm_v4i32_ty], [llvm_v16i8_ty, llvm_i32_ty],
-            [IntrNoMem, IntrSpeculatable, ImmArg<ArgIndex<1>>]>;
-
 //===----------------------------------------------------------------------===//
 // Thread-local storage intrinsics
 //===----------------------------------------------------------------------===//

llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h

@@ -196,8 +196,6 @@ inline unsigned GetDefaultP2AlignAny(unsigned Opc) {
   WASM_LOAD_STORE(LOAD8_SPLAT)
   WASM_LOAD_STORE(LOAD_LANE_I8x16)
   WASM_LOAD_STORE(STORE_LANE_I8x16)
-  WASM_LOAD_STORE(PREFETCH_T)
-  WASM_LOAD_STORE(PREFETCH_NT)
   return 0;
   WASM_LOAD_STORE(LOAD16_S_I32)
   WASM_LOAD_STORE(LOAD16_U_I32)

llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp

@@ -761,16 +761,6 @@ bool WebAssemblyTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
     Info.align = MemAlign;
     return true;
   }
-  case Intrinsic::wasm_prefetch_t:
-  case Intrinsic::wasm_prefetch_nt: {
-    Info.opc = ISD::INTRINSIC_VOID;
-    Info.memVT = MVT::i8;
-    Info.ptrVal = I.getArgOperand(0);
-    Info.offset = 0;
-    Info.align = Align(1);
-    Info.flags = MachineMemOperand::MOLoad;
-    return true;
-  }
   default:
     return false;
   }