Index: include/llvm/IR/IntrinsicsWebAssembly.td =================================================================== --- include/llvm/IR/IntrinsicsWebAssembly.td +++ include/llvm/IR/IntrinsicsWebAssembly.td @@ -87,4 +87,17 @@ [IntrInaccessibleMemOnly, NoCapture<0>, IntrHasSideEffects], "", [SDNPMemOperand]>; -} +//===----------------------------------------------------------------------===// +// SIMD intrinsics +//===----------------------------------------------------------------------===// + +def int_wasm_anytrue : + Intrinsic<[llvm_i32_ty], + [llvm_anyvector_ty], + [IntrNoMem, IntrSpeculatable]>; +def int_wasm_alltrue : + Intrinsic<[llvm_i32_ty], + [llvm_anyvector_ty], + [IntrNoMem, IntrSpeculatable]>; + +} // TargetPrefix = "wasm" Index: lib/Target/WebAssembly/WebAssemblyISD.def =================================================================== --- lib/Target/WebAssembly/WebAssemblyISD.def +++ lib/Target/WebAssembly/WebAssemblyISD.def @@ -22,5 +22,7 @@ HANDLE_NODETYPE(BR_IF) HANDLE_NODETYPE(BR_TABLE) HANDLE_NODETYPE(SHUFFLE) +HANDLE_NODETYPE(ANYTRUE) +HANDLE_NODETYPE(ALLTRUE) // add memory opcodes starting at ISD::FIRST_TARGET_MEMORY_OPCODE here... Index: lib/Target/WebAssembly/WebAssemblyISelLowering.cpp =================================================================== --- lib/Target/WebAssembly/WebAssemblyISelLowering.cpp +++ lib/Target/WebAssembly/WebAssemblyISelLowering.cpp @@ -959,7 +959,13 @@ switch (IntNo) { default: return {}; // Don't custom lower most intrinsics. - + case Intrinsic::wasm_anytrue: + case Intrinsic::wasm_alltrue: { + unsigned OpCode = IntNo == Intrinsic::wasm_anytrue + ? WebAssemblyISD::ANYTRUE + : WebAssemblyISD::ALLTRUE; + return DAG.getNode(OpCode, DL, Op.getValueType(), Op.getOperand(1)); + } case Intrinsic::wasm_lsda: // TODO For now, just return 0 not to crash return DAG.getConstant(0, DL, Op.getValueType()); Index: lib/Target/WebAssembly/WebAssemblyInstrSIMD.td =================================================================== --- lib/Target/WebAssembly/WebAssemblyInstrSIMD.td +++ lib/Target/WebAssembly/WebAssemblyInstrSIMD.td @@ -18,6 +18,13 @@ foreach SIZE = [2, 4, 8, 16, 32] in def LaneIdx#SIZE : ImmLeaf; +// Custom nodes for custom operations +def wasm_shuffle_t : SDTypeProfile<1, 18, []>; +def wasm_reduce_t : SDTypeProfile<1, 1, [SDTCisVT<0, i32>, SDTCisVec<1>]>; +def wasm_shuffle : SDNode<"WebAssemblyISD::SHUFFLE", wasm_shuffle_t>; +def wasm_anytrue : SDNode<"WebAssemblyISD::ANYTRUE", wasm_reduce_t>; +def wasm_alltrue : SDNode<"WebAssemblyISD::ALLTRUE", wasm_reduce_t>; + multiclass ConstVec { let isMoveImm = 1, isReMaterializable = 1 in defm CONST_V128_#vec_t : SIMD_I<(outs V128:$dst), ops, (outs), ops, @@ -186,6 +193,18 @@ )], "v128.not\t$dst, $vec", "v128.not", 63>; } +multiclass SIMDReduceVec simdop> { + defm _#vec_t : SIMD_I<(outs I32:$dst), (ins V128:$vec), (outs), (ins), + [(set I32:$dst, (i32 (op (vec_t V128:$vec))))], + vec#"."#name#"\t$dst, $vec", vec#"."#name, simdop>; +} +multiclass SIMDReduce baseInst> { + defm "" : SIMDReduceVec; + defm "" : SIMDReduceVec; + defm "" : SIMDReduceVec; + defm "" : SIMDReduceVec; +} multiclass SIMDCondition simdop> { defm _#vec_t : @@ -361,6 +380,9 @@ defm "" : SIMDNot; defm "" : SIMDNot; +defm ANYTRUE : SIMDReduce<"any_true", wasm_anytrue, 65>; +defm ALLTRUE : SIMDReduce<"all_true", wasm_alltrue, 69>; + let isCommutable = 1 in { defm EQ : SIMDConditionInt<"eq", SETEQ, 73>; defm EQ : SIMDConditionFP<"eq", SETOEQ, 77>; @@ -457,8 +479,6 @@ (v2i64 (shifts[1] (v2i64 V128:$vec), (I32_WRAP_I64 I64:$x)))>; // Shuffles after custom lowering -def wasm_shuffle_t : SDTypeProfile<1, 18, []>; -def wasm_shuffle : SDNode<"WebAssemblyISD::SHUFFLE", wasm_shuffle_t>; foreach vec_t = [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64] in { def : Pat<(v16i8 (wasm_shuffle (vec_t V128:$x), (vec_t V128:$y), (i32 LaneIdx32:$m0), (i32 LaneIdx32:$m1), Index: test/CodeGen/WebAssembly/simd-intrinsics.ll =================================================================== --- /dev/null +++ test/CodeGen/WebAssembly/simd-intrinsics.ll @@ -0,0 +1,117 @@ +; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -wasm-enable-unimplemented-simd -mattr=+simd128 | FileCheck %s --check-prefixes CHECK,SIMD128 +; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -wasm-enable-unimplemented-simd -mattr=+simd128 -fast-isel | FileCheck %s --check-prefixes CHECK,SIMD128 + +; Test that SIMD128 intrinsics lower as expected. These intrinsics are +; only expected to lower successfully if the simd128 attribute is +; enabled and legal types are used. + +target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128" +target triple = "wasm32-unknown-unknown" + +; ============================================================================== +; 16 x i8 +; ============================================================================== +; CHECK-LABEL: any_v16i8: +; NO-SIMD128-NOT: i8x16 +; SIMD128-NEXT: .param v128{{$}} +; SIMD128-NEXT: .result i32{{$}} +; SIMD128-NEXT: i8x16.any_true $push[[R:[0-9]+]]=, $0{{$}} +; SIMD128-NEXT: return $pop[[R]]{{$}} +declare i32 @llvm.wasm.anytrue.v16i8(<16 x i8>) +define i32 @any_v16i8(<16 x i8> %x) { + %a = call i32 @llvm.wasm.anytrue.v16i8(<16 x i8> %x) + ret i32 %a +} + +; CHECK-LABEL: all_v16i8: +; NO-SIMD128-NOT: i8x16 +; SIMD128-NEXT: .param v128{{$}} +; SIMD128-NEXT: .result i32{{$}} +; SIMD128-NEXT: i8x16.all_true $push[[R:[0-9]+]]=, $0{{$}} +; SIMD128-NEXT: return $pop[[R]]{{$}} +declare i32 @llvm.wasm.alltrue.v16i8(<16 x i8>) +define i32 @all_v16i8(<16 x i8> %x) { + %a = call i32 @llvm.wasm.alltrue.v16i8(<16 x i8> %x) + ret i32 %a +} + +; ============================================================================== +; 8 x i16 +; ============================================================================== +; CHECK-LABEL: any_v8i16: +; NO-SIMD128-NOT: i16x8 +; SIMD128-NEXT: .param v128{{$}} +; SIMD128-NEXT: .result i32{{$}} +; SIMD128-NEXT: i16x8.any_true $push[[R:[0-9]+]]=, $0{{$}} +; SIMD128-NEXT: return $pop[[R]]{{$}} +declare i32 @llvm.wasm.anytrue.v8i16(<8 x i16>) +define i32 @any_v8i16(<8 x i16> %x) { + %a = call i32 @llvm.wasm.anytrue.v8i16(<8 x i16> %x) + ret i32 %a +} + +; CHECK-LABEL: all_v8i16: +; NO-SIMD128-NOT: i16x8 +; SIMD128-NEXT: .param v128{{$}} +; SIMD128-NEXT: .result i32{{$}} +; SIMD128-NEXT: i16x8.all_true $push[[R:[0-9]+]]=, $0{{$}} +; SIMD128-NEXT: return $pop[[R]]{{$}} +declare i32 @llvm.wasm.alltrue.v8i16(<8 x i16>) +define i32 @all_v8i16(<8 x i16> %x) { + %a = call i32 @llvm.wasm.alltrue.v8i16(<8 x i16> %x) + ret i32 %a +} + +; ============================================================================== +; 4 x i32 +; ============================================================================== +; CHECK-LABEL: any_v4i32: +; NO-SIMD128-NOT: i32x4 +; SIMD128-NEXT: .param v128{{$}} +; SIMD128-NEXT: .result i32{{$}} +; SIMD128-NEXT: i32x4.any_true $push[[R:[0-9]+]]=, $0{{$}} +; SIMD128-NEXT: return $pop[[R]]{{$}} +declare i32 @llvm.wasm.anytrue.v4i32(<4 x i32>) +define i32 @any_v4i32(<4 x i32> %x) { + %a = call i32 @llvm.wasm.anytrue.v4i32(<4 x i32> %x) + ret i32 %a +} + +; CHECK-LABEL: all_v4i32: +; NO-SIMD128-NOT: i32x4 +; SIMD128-NEXT: .param v128{{$}} +; SIMD128-NEXT: .result i32{{$}} +; SIMD128-NEXT: i32x4.all_true $push[[R:[0-9]+]]=, $0{{$}} +; SIMD128-NEXT: return $pop[[R]]{{$}} +declare i32 @llvm.wasm.alltrue.v4i32(<4 x i32>) +define i32 @all_v4i32(<4 x i32> %x) { + %a = call i32 @llvm.wasm.alltrue.v4i32(<4 x i32> %x) + ret i32 %a +} + +; ============================================================================== +; 2 x i64 +; ============================================================================== +; CHECK-LABEL: any_v2i64: +; NO-SIMD128-NOT: i64x2 +; SIMD128-NEXT: .param v128{{$}} +; SIMD128-NEXT: .result i32{{$}} +; SIMD128-NEXT: i64x2.any_true $push[[R:[0-9]+]]=, $0{{$}} +; SIMD128-NEXT: return $pop[[R]]{{$}} +declare i32 @llvm.wasm.anytrue.v2i64(<2 x i64>) +define i32 @any_v2i64(<2 x i64> %x) { + %a = call i32 @llvm.wasm.anytrue.v2i64(<2 x i64> %x) + ret i32 %a +} + +; CHECK-LABEL: all_v2i64: +; NO-SIMD128-NOT: i64x2 +; SIMD128-NEXT: .param v128{{$}} +; SIMD128-NEXT: .result i32{{$}} +; SIMD128-NEXT: i64x2.all_true $push[[R:[0-9]+]]=, $0{{$}} +; SIMD128-NEXT: return $pop[[R]]{{$}} +declare i32 @llvm.wasm.alltrue.v2i64(<2 x i64>) +define i32 @all_v2i64(<2 x i64> %x) { + %a = call i32 @llvm.wasm.alltrue.v2i64(<2 x i64> %x) + ret i32 %a +} Index: test/MC/WebAssembly/simd-encodings.s =================================================================== --- test/MC/WebAssembly/simd-encodings.s +++ test/MC/WebAssembly/simd-encodings.s @@ -193,6 +193,30 @@ # CHECK: v128.not # encoding: [0xfd,0x3f] v128.not + # CHECK: i8x16.any_true # encoding: [0xfd,0x41] + i8x16.any_true + + # CHECK: i16x8.any_true # encoding: [0xfd,0x42] + i16x8.any_true + + # CHECK: i32x4.any_true # encoding: [0xfd,0x43] + i32x4.any_true + + # CHECK: i64x2.any_true # encoding: [0xfd,0x44] + i64x2.any_true + + # CHECK: i8x16.all_true # encoding: [0xfd,0x45] + i8x16.all_true + + # CHECK: i16x8.all_true # encoding: [0xfd,0x46] + i16x8.all_true + + # CHECK: i32x4.all_true # encoding: [0xfd,0x47] + i32x4.all_true + + # CHECK: i64x2.all_true # encoding: [0xfd,0x48] + i64x2.all_true + # CHECK: i8x16.eq # encoding: [0xfd,0x49] i8x16.eq