https://github.com/alowqie created https://github.com/llvm/llvm-project/pull/186693
Part of https://github.com/llvm/llvm-project/issues/185382 - Moved lowering logic from clangir incubator to upstream - Added tests, partially reusing tests from [neon-intrinsics.c](https://github.com/llvm/llvm-project/blob/main/clang/test/CodeGen/AArch64/neon-intrinsics.c) and [neon.c](https://github.com/llvm/clangir/blob/main/clang/test/CIR/CodeGen/AArch64/neon.c) - Made sure that all intrinsics from [Neon ACLE](https://arm-software.github.io/acle/neon_intrinsics/advsimd.html#vector-shift-right) are implemented and tested @banach-space the work on `vshr_*` seemed to involve moving more code from the incubator than I've seen in other PRs, so I'm a bit unsure if I'm correct with this. Those changes made sense to me in order to cover all of `vshr_*` variants but please let me know if I misunderstood something >From 4a99eda183b19a2672c63be4ef99e3bd05a9a85c Mon Sep 17 00:00:00 2001 From: alowqie <[email protected]> Date: Sun, 15 Mar 2026 17:14:47 +0100 Subject: [PATCH 1/2] [CIR][AArch64] add vshr_* builtins --- clang/lib/CIR/CodeGen/CIRGenBuilder.h | 8 + .../lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp | 94 +++++++- clang/test/CodeGen/AArch64/neon/intrinsics.c | 216 ++++++++++++++++++ 3 files changed, 315 insertions(+), 3 deletions(-) diff --git a/clang/lib/CIR/CodeGen/CIRGenBuilder.h b/clang/lib/CIR/CodeGen/CIRGenBuilder.h index 7cd1bdcf491be..82300e66cd50a 100644 --- a/clang/lib/CIR/CodeGen/CIRGenBuilder.h +++ b/clang/lib/CIR/CodeGen/CIRGenBuilder.h @@ -432,6 +432,14 @@ class CIRGenBuilderTy : public cir::CIRBaseBuilderTy { return cir::ConstantOp::create(*this, loc, getNullMethodAttr(ty)); } + cir::ConstantOp getZero(mlir::Location loc, mlir::Type ty) { + // TODO: dispatch creation for primitive types. + assert((mlir::isa<cir::RecordType>(ty) || mlir::isa<cir::ArrayType>(ty) || + mlir::isa<cir::VectorType>(ty)) && + "NYI for other types"); + return cir::ConstantOp::create(*this, loc, cir::ZeroAttr::get(ty)); + } + //===--------------------------------------------------------------------===// // UnaryOp creation helpers //===--------------------------------------------------------------------===// diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp index 5534e69b5f8bc..12ac46bdd1b55 100644 --- a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp +++ b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp @@ -831,6 +831,69 @@ static cir::VectorType getNeonType(CIRGenFunction *cgf, NeonTypeFlags typeFlags, llvm_unreachable("Unknown vector element type!"); } +// Get integer from a mlir::Value that is an int constant or a constant op. +static int64_t getIntValueFromConstOp(mlir::Value val) { + return val.getDefiningOp<cir::ConstantOp>().getIntValue().getSExtValue(); +} + +// Build a constant shift amount vector of `vecTy` to shift a vector +// Here `shitfVal` is a constant integer that will be splated into a +// a const vector of `vecTy` which is the return of this function +static mlir::Value emitNeonShiftVector(CIRGenBuilderTy &builder, + mlir::Value shiftVal, + cir::VectorType vecTy, + mlir::Location loc, bool neg) { + int shiftAmt = getIntValueFromConstOp(shiftVal); + if (neg) + shiftAmt = -shiftAmt; + llvm::SmallVector<mlir::Attribute> vecAttr{ + vecTy.getSize(), + // ConstVectorAttr requires cir::IntAttr + cir::IntAttr::get(vecTy.getElementType(), shiftAmt)}; + cir::ConstVectorAttr constVecAttr = cir::ConstVectorAttr::get( + vecTy, mlir::ArrayAttr::get(builder.getContext(), vecAttr)); + return cir::ConstantOp::create(builder, loc, constVecAttr); +} + +// Build ShiftOp of vector type whose shift amount is a vector built +// from a constant integer using `emitNeonShiftVector` function +static mlir::Value +emitCommonNeonShift(CIRGenBuilderTy &builder, mlir::Location loc, + cir::VectorType resTy, mlir::Value shifTgt, + mlir::Value shiftAmt, bool shiftLeft, bool negAmt = false) { + shiftAmt = emitNeonShiftVector(builder, shiftAmt, resTy, loc, negAmt); + return cir::ShiftOp::create(builder, loc, resTy, + builder.createBitcast(shifTgt, resTy), shiftAmt, + shiftLeft); +} + +// Right-shift a vector by a constant. +static mlir::Value emitNeonRShiftImm(CIRGenFunction &cgf, mlir::Value shiftVec, + mlir::Value shiftVal, + cir::VectorType vecTy, bool usgn, + mlir::Location loc) { + CIRGenBuilderTy &builder = cgf.getBuilder(); + int64_t shiftAmt = getIntValueFromConstOp(shiftVal); + int eltSize = + cgf.cgm.getDataLayout().getTypeSizeInBits(vecTy.getElementType()); + + shiftVec = builder.createBitcast(shiftVec, vecTy); + // lshr/ashr are undefined when the shift amount is equal to the vector + // element size. + if (shiftAmt == eltSize) { + if (usgn) { + // Right-shifting an unsigned value by its size yields 0. + return builder.getZero(loc, vecTy); + } + // Right-shifting a signed value by its size is equivalent + // to a shift of size-1. + --shiftAmt; + shiftVal = builder.getConstInt(loc, vecTy.getElementType(), shiftAmt); + } + return emitCommonNeonShift(builder, loc, vecTy, shiftVec, shiftVal, + false /* right shift */); +} + static mlir::Value emitCommonNeonBuiltinExpr( CIRGenFunction &cgf, unsigned builtinID, unsigned llvmIntrinsic, unsigned altLLVMIntrinsic, const char *nameHint, unsigned modifier, @@ -849,6 +912,7 @@ static mlir::Value emitCommonNeonBuiltinExpr( // Determine the type of this overloaded NEON intrinsic. NeonTypeFlags neonType(neonTypeConst->getZExtValue()); + bool isUnsigned = neonType.isUnsigned(); const bool hasLegalHalfType = cgf.getTarget().hasFastHalfType(); // The value of allowBFloatArgsAndRet is true for AArch64, but it should @@ -1057,8 +1121,13 @@ static mlir::Value emitCommonNeonBuiltinExpr( case NEON::BI__builtin_neon_vshlq_n_v: case NEON::BI__builtin_neon_vshll_n_v: case NEON::BI__builtin_neon_vshrn_n_v: + cgf.cgm.errorNYI(expr->getSourceRange(), + std::string("unimplemented AArch64 builtin call: ") + + ctx.BuiltinInfo.getName(builtinID)); + return mlir::Value{}; case NEON::BI__builtin_neon_vshr_n_v: case NEON::BI__builtin_neon_vshrq_n_v: + return emitNeonRShiftImm(cgf, ops[0], ops[1], vTy, isUnsigned, loc); case NEON::BI__builtin_neon_vst1_v: case NEON::BI__builtin_neon_vst1q_v: case NEON::BI__builtin_neon_vst2_v: @@ -2786,9 +2855,28 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned builtinID, const CallExpr *expr, case NEON::BI__builtin_neon_vrsrad_n_u64: case NEON::BI__builtin_neon_vrsrad_n_s64: case NEON::BI__builtin_neon_vshld_n_s64: - case NEON::BI__builtin_neon_vshld_n_u64: - case NEON::BI__builtin_neon_vshrd_n_s64: - case NEON::BI__builtin_neon_vshrd_n_u64: + case NEON::BI__builtin_neon_vshld_n_u64: { + cgm.errorNYI(expr->getSourceRange(), + std::string("unimplemented AArch64 builtin call: ") + + getContext().BuiltinInfo.getName(builtinID)); + return mlir::Value{}; + } + case NEON::BI__builtin_neon_vshrd_n_s64: { + std::optional<llvm::APSInt> amt = + expr->getArg(1)->getIntegerConstantExpr(getContext()); + assert(amt && "Expected argument to be a constant"); + uint64_t bits = std::min(static_cast<uint64_t>(63), amt->getZExtValue()); + return builder.createShiftRight(loc, ops[0], bits); + } + case NEON::BI__builtin_neon_vshrd_n_u64: { + std::optional<llvm::APSInt> amt = + expr->getArg(1)->getIntegerConstantExpr(getContext()); + assert(amt && "Expected argument to be a constant"); + uint64_t shiftAmt = amt->getZExtValue(); + if (shiftAmt == 64) + return builder.getConstInt(loc, builder.getUInt64Ty(), 0); + return builder.createShiftRight(loc, ops[0], shiftAmt); + } case NEON::BI__builtin_neon_vsrad_n_s64: case NEON::BI__builtin_neon_vsrad_n_u64: case NEON::BI__builtin_neon_vqdmlalh_lane_s16: diff --git a/clang/test/CodeGen/AArch64/neon/intrinsics.c b/clang/test/CodeGen/AArch64/neon/intrinsics.c index b740c3b5b2310..be4412f99ab8a 100644 --- a/clang/test/CodeGen/AArch64/neon/intrinsics.c +++ b/clang/test/CodeGen/AArch64/neon/intrinsics.c @@ -936,3 +936,219 @@ uint32x4_t test_vabaq_u32(uint32x4_t v1, uint32x4_t v2, uint32x4_t v3) { // LLVM-NEXT: ret <4 x i32> [[ADD_I]] return vabaq_u32(v1, v2, v3); } + +//===------------------------------------------------------===// +// 2.1.3.2.1 Vector shift right +//===------------------------------------------------------===// + +// LLVM-LABEL: @test_vshr_n_s8( +// CIR-LABEL: @test_vshr_n_s8( +int8x8_t test_vshr_n_s8(int8x8_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, +// CIR-SAME: #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i]> : !cir.vector<8 x !s8i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<8 x !s8i>, [[AMT]] : !cir.vector<8 x !s8i>) -> !cir.vector<8 x !s8i> + +// LLVM: [[VSHR_N:%.*]] = ashr <8 x i8> {{.*}}, splat (i8 3) +// LLVM: ret <8 x i8> [[VSHR_N]] + return vshr_n_s8(a, 3); +} + +// LLVM-LABEL: @test_vshr_n_s16( +// CIR-LABEL: @test_vshr_n_s16( +int16x4_t test_vshr_n_s16(int16x4_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !s16i, #cir.int<3> : !s16i, +// CIR-SAME: #cir.int<3> : !s16i, #cir.int<3> : !s16i]> : !cir.vector<4 x !s16i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<4 x !s16i>, [[AMT]] : !cir.vector<4 x !s16i>) -> !cir.vector<4 x !s16i> + +// LLVM: [[VSHR_N:%.*]] = ashr <4 x i16> {{.*}}, splat (i16 3) +// LLVM: ret <4 x i16> [[VSHR_N]] + return vshr_n_s16(a, 3); +} + +// LLVM-LABEL: @test_vshr_n_s32( +// CIR-LABEL: @test_vshr_n_s32( +int32x2_t test_vshr_n_s32(int32x2_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !s32i, #cir.int<3> : !s32i]> : !cir.vector<2 x !s32i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<2 x !s32i>, [[AMT]] : !cir.vector<2 x !s32i>) -> !cir.vector<2 x !s32i> + +// LLVM: [[VSHR_N:%.*]] = ashr <2 x i32> {{.*}}, splat (i32 3) +// LLVM: ret <2 x i32> [[VSHR_N]] + return vshr_n_s32(a, 3); +} + +// LLVM-LABEL: @test_vshr_n_s64( +// CIR-LABEL: @test_vshr_n_s64( +int64x1_t test_vshr_n_s64(int64x1_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !s64i]> : !cir.vector<1 x !s64i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<1 x !s64i>, [[AMT]] : !cir.vector<1 x !s64i>) -> !cir.vector<1 x !s64i> + +// LLVM: [[VSHR_N:%.*]] = ashr <1 x i64> {{.*}}, splat (i64 3) +// LLVM: ret <1 x i64> [[VSHR_N]] + return vshr_n_s64(a, 3); +} + +// LLVM-LABEL: @test_vshrq_n_s8( +// CIR-LABEL: @test_vshrq_n_s8( +int8x16_t test_vshrq_n_s8(int8x16_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, +// CIR-SAME: #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, +// CIR-SAME: #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i, +// CIR-SAME: #cir.int<3> : !s8i, #cir.int<3> : !s8i, #cir.int<3> : !s8i]> : !cir.vector<16 x !s8i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<16 x !s8i>, [[AMT]] : !cir.vector<16 x !s8i>) -> !cir.vector<16 x !s8i> + +// LLVM: [[VSHR_N:%.*]] = ashr <16 x i8> {{.*}}, splat (i8 3) +// LLVM: ret <16 x i8> [[VSHR_N]] + return vshrq_n_s8(a, 3); +} + +// LLVM-LABEL: @test_vshrq_n_s16( +// CIR-LABEL: @test_vshrq_n_s16( +int16x8_t test_vshrq_n_s16(int16x8_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !s16i, #cir.int<3> : !s16i, #cir.int<3> : !s16i, +// CIR-SAME: #cir.int<3> : !s16i, #cir.int<3> : !s16i, #cir.int<3> : !s16i, #cir.int<3> : !s16i, +// CIR-SAME: #cir.int<3> : !s16i]> : !cir.vector<8 x !s16i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<8 x !s16i>, [[AMT]] : !cir.vector<8 x !s16i>) -> !cir.vector<8 x !s16i> + +// LLVM: [[VSHR_N:%.*]] = ashr <8 x i16> {{.*}}, splat (i16 3) +// LLVM: ret <8 x i16> [[VSHR_N]] + return vshrq_n_s16(a, 3); +} + +// LLVM-LABEL: @test_vshrq_n_s32( +// CIR-LABEL: @test_vshrq_n_s32( +int32x4_t test_vshrq_n_s32(int32x4_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !s32i, #cir.int<3> : !s32i, +// CIR-SAME: #cir.int<3> : !s32i, #cir.int<3> : !s32i]> : !cir.vector<4 x !s32i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<4 x !s32i>, [[AMT]] : !cir.vector<4 x !s32i>) -> !cir.vector<4 x !s32i> + +// LLVM: [[VSHR_N:%.*]] = ashr <4 x i32> {{.*}}, splat (i32 3) +// LLVM: ret <4 x i32> [[VSHR_N]] + return vshrq_n_s32(a, 3); +} + +// LLVM-LABEL: @test_vshrq_n_s64( +// CIR-LABEL: @test_vshrq_n_s64( +int64x2_t test_vshrq_n_s64(int64x2_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !s64i, #cir.int<3> : !s64i]> : !cir.vector<2 x !s64i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<2 x !s64i>, [[AMT]] : !cir.vector<2 x !s64i>) -> !cir.vector<2 x !s64i> + +// LLVM: [[VSHR_N:%.*]] = ashr <2 x i64> {{.*}}, splat (i64 3) +// LLVM: ret <2 x i64> [[VSHR_N]] + return vshrq_n_s64(a, 3); +} + +// LLVM-LABEL: @test_vshr_n_u8( +// CIR-LABEL: @test_vshr_n_u8( +uint8x8_t test_vshr_n_u8(uint8x8_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, +// CIR-SAME: #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i]> : !cir.vector<8 x !u8i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<8 x !u8i>, [[AMT]] : !cir.vector<8 x !u8i>) -> !cir.vector<8 x !u8i> + +// LLVM: [[VSHR_N:%.*]] = lshr <8 x i8> {{.*}}, splat (i8 3) +// LLVM: ret <8 x i8> [[VSHR_N]] + return vshr_n_u8(a, 3); +} + +// LLVM-LABEL: @test_vshr_n_u16( +// CIR-LABEL: @test_vshr_n_u16( +uint16x4_t test_vshr_n_u16(uint16x4_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !u16i, #cir.int<3> : !u16i, +// CIR-SAME: #cir.int<3> : !u16i, #cir.int<3> : !u16i]> : !cir.vector<4 x !u16i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<4 x !u16i>, [[AMT]] : !cir.vector<4 x !u16i>) -> !cir.vector<4 x !u16i> + +// LLVM: [[VSHR_N:%.*]] = lshr <4 x i16> {{.*}}, splat (i16 3) +// LLVM: ret <4 x i16> [[VSHR_N]] + return vshr_n_u16(a, 3); +} + +// LLVM-LABEL: @test_vshr_n_u32( +// CIR-LABEL: @test_vshr_n_u32( +uint32x2_t test_vshr_n_u32(uint32x2_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !u32i, #cir.int<3> : !u32i]> : !cir.vector<2 x !u32i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<2 x !u32i>, [[AMT]] : !cir.vector<2 x !u32i>) -> !cir.vector<2 x !u32i> + +// LLVM: [[VSHR_N:%.*]] = lshr <2 x i32> {{.*}}, splat (i32 3) +// LLVM: ret <2 x i32> [[VSHR_N]] + return vshr_n_u32(a, 3); +} + +// LLVM-LABEL: @test_vshr_n_u64( +// CIR-LABEL: @test_vshr_n_u64( +uint64x1_t test_vshr_n_u64(uint64x1_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !u64i]> : !cir.vector<1 x !u64i> + // CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<1 x !u64i>, [[AMT]] : !cir.vector<1 x !u64i>) -> !cir.vector<1 x !u64i> + +// LLVM: [[VSHR_N:%.*]] = lshr <1 x i64> {{.*}}, splat (i64 3) +// ret <1 x i64> [[VSHR_N]] + return vshr_n_u64(a, 3); +} + +// LLVM-LABEL: @test_vshrq_n_u8( +// CIR-LABEL: @test_vshrq_n_u8( +uint8x16_t test_vshrq_n_u8(uint8x16_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, +// CIR-SAME: #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, +// CIR-SAME: #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, +// CIR-SAME: #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i, #cir.int<3> : !u8i]> : !cir.vector<16 x !u8i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<16 x !u8i>, [[AMT]] : !cir.vector<16 x !u8i>) -> !cir.vector<16 x !u8i> + +// LLVM: [[VSHR_N:%.*]] = lshr <16 x i8> {{.*}}, splat (i8 3) +// LLVM: ret <16 x i8> [[VSHR_N]] + return vshrq_n_u8(a, 3); +} + +// LLVM-LABEL: @test_vshrq_n_u16( +// CIR-LABEL: @test_vshrq_n_u16( +uint16x8_t test_vshrq_n_u16(uint16x8_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !u16i, #cir.int<3> : !u16i, #cir.int<3> : !u16i, +// CIR-SAME: #cir.int<3> : !u16i, #cir.int<3> : !u16i, #cir.int<3> : !u16i, #cir.int<3> : !u16i, +// CIR-SAME: #cir.int<3> : !u16i]> : !cir.vector<8 x !u16i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<8 x !u16i>, [[AMT]] : !cir.vector<8 x !u16i>) -> !cir.vector<8 x !u16i> + +// LLVM: [[VSHR_N:%.*]] = lshr <8 x i16> {{.*}}, splat (i16 3) +// LLVM: ret <8 x i16> [[VSHR_N]] + return vshrq_n_u16(a, 3); +} + +// LLVM-LABEL: @test_vshrq_n_u32( +// CIR-LABEL: @test_vshrq_n_u32( +uint32x4_t test_vshrq_n_u32(uint32x4_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !u32i, #cir.int<3> : !u32i, +// CIR-SAME: #cir.int<3> : !u32i, #cir.int<3> : !u32i]> : !cir.vector<4 x !u32i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<4 x !u32i>, [[AMT]] : !cir.vector<4 x !u32i>) -> !cir.vector<4 x !u32i> + +// LLVM: [[VSHR_N:%.*]] = lshr <4 x i32> {{.*}}, splat (i32 3) +// LLVM: ret <4 x i32> [[VSHR_N]] + return vshrq_n_u32(a, 3); +} + +// LLVM-LABEL: @test_vshrq_n_u64( +// CIR-LABEL: @test_vshrq_n_u64( +uint64x2_t test_vshrq_n_u64(uint64x2_t a) { +// CIR: [[AMT:%.*]] = cir.const #cir.const_vector<[#cir.int<3> : !u64i, #cir.int<3> : !u64i]> : !cir.vector<2 x !u64i> +// CIR: {{%.*}} = cir.shift(right, {{%.*}} : !cir.vector<2 x !u64i>, [[AMT]] : !cir.vector<2 x !u64i>) -> !cir.vector<2 x !u64i> + +// LLVM: [[VSHR_N:%.*]] = lshr <2 x i64> {{.*}}, splat (i64 3) +// LLVM: ret <2 x i64> [[VSHR_N]] + return vshrq_n_u64(a, 3); +} + +// LLVM-LABEL: @test_vshrd_n_s64( +// CIR-LABEL: @test_vshrd_n_s64( +int64_t test_vshrd_n_s64(int64_t a) { + // CIR: {{%.*}} = cir.shift(right, {{%.*}} : !s64i, {{%.*}} : !s64i) -> !s64i + + // LLVM: [[SHRD_N:%.*]] = ashr i64 {{.*}}, 1 + // LLVM: ret i64 [[SHRD_N]] + return (int64_t)vshrd_n_s64(a, 1); +} + +// LLVM-LABEL: @test_vshrd_n_u64( +// CIR-LABEL: @test_vshrd_n_u64( +uint64_t test_vshrd_n_u64(uint64_t a) { + // CIR: {{.*}} = cir.const #cir.int<0> : !u64i + // CIR: cir.return {{.*}} : !u64i + + // LLVM: ret i64 0 + return (uint64_t)vshrd_n_u64(a, 64); +} >From 6939f20107c5252e1bd3ebbd256131b2ec6f4a1a Mon Sep 17 00:00:00 2001 From: alowqie <[email protected]> Date: Sun, 15 Mar 2026 18:26:49 +0100 Subject: [PATCH 2/2] Remove tests from neon-intrinsics.c, moved to neon/intrinsics.c --- clang/test/CodeGen/AArch64/neon-intrinsics.c | 191 ------------------- 1 file changed, 191 deletions(-) diff --git a/clang/test/CodeGen/AArch64/neon-intrinsics.c b/clang/test/CodeGen/AArch64/neon-intrinsics.c index bfaea2b8ae909..5a95133dd7429 100644 --- a/clang/test/CodeGen/AArch64/neon-intrinsics.c +++ b/clang/test/CodeGen/AArch64/neon-intrinsics.c @@ -6770,166 +6770,6 @@ uint64x2_t test_vshlq_n_u64(uint64x2_t a) { return vshlq_n_u64(a, 3); } -// CHECK-LABEL: define dso_local <8 x i8> @test_vshr_n_s8( -// CHECK-SAME: <8 x i8> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[VSHR_N:%.*]] = ashr <8 x i8> [[A]], splat (i8 3) -// CHECK-NEXT: ret <8 x i8> [[VSHR_N]] -// -int8x8_t test_vshr_n_s8(int8x8_t a) { - return vshr_n_s8(a, 3); -} - -// CHECK-LABEL: define dso_local <4 x i16> @test_vshr_n_s16( -// CHECK-SAME: <4 x i16> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i16> [[A]] to <8 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16> -// CHECK-NEXT: [[VSHR_N:%.*]] = ashr <4 x i16> [[TMP1]], splat (i16 3) -// CHECK-NEXT: ret <4 x i16> [[VSHR_N]] -// -int16x4_t test_vshr_n_s16(int16x4_t a) { - return vshr_n_s16(a, 3); -} - -// CHECK-LABEL: define dso_local <2 x i32> @test_vshr_n_s32( -// CHECK-SAME: <2 x i32> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x i32> [[A]] to <8 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32> -// CHECK-NEXT: [[VSHR_N:%.*]] = ashr <2 x i32> [[TMP1]], splat (i32 3) -// CHECK-NEXT: ret <2 x i32> [[VSHR_N]] -// -int32x2_t test_vshr_n_s32(int32x2_t a) { - return vshr_n_s32(a, 3); -} - -// CHECK-LABEL: define dso_local <16 x i8> @test_vshrq_n_s8( -// CHECK-SAME: <16 x i8> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[VSHR_N:%.*]] = ashr <16 x i8> [[A]], splat (i8 3) -// CHECK-NEXT: ret <16 x i8> [[VSHR_N]] -// -int8x16_t test_vshrq_n_s8(int8x16_t a) { - return vshrq_n_s8(a, 3); -} - -// CHECK-LABEL: define dso_local <8 x i16> @test_vshrq_n_s16( -// CHECK-SAME: <8 x i16> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x i16> [[A]] to <16 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16> -// CHECK-NEXT: [[VSHR_N:%.*]] = ashr <8 x i16> [[TMP1]], splat (i16 3) -// CHECK-NEXT: ret <8 x i16> [[VSHR_N]] -// -int16x8_t test_vshrq_n_s16(int16x8_t a) { - return vshrq_n_s16(a, 3); -} - -// CHECK-LABEL: define dso_local <4 x i32> @test_vshrq_n_s32( -// CHECK-SAME: <4 x i32> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i32> [[A]] to <16 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32> -// CHECK-NEXT: [[VSHR_N:%.*]] = ashr <4 x i32> [[TMP1]], splat (i32 3) -// CHECK-NEXT: ret <4 x i32> [[VSHR_N]] -// -int32x4_t test_vshrq_n_s32(int32x4_t a) { - return vshrq_n_s32(a, 3); -} - -// CHECK-LABEL: define dso_local <2 x i64> @test_vshrq_n_s64( -// CHECK-SAME: <2 x i64> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x i64> [[A]] to <16 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <2 x i64> -// CHECK-NEXT: [[VSHR_N:%.*]] = ashr <2 x i64> [[TMP1]], splat (i64 3) -// CHECK-NEXT: ret <2 x i64> [[VSHR_N]] -// -int64x2_t test_vshrq_n_s64(int64x2_t a) { - return vshrq_n_s64(a, 3); -} - -// CHECK-LABEL: define dso_local <8 x i8> @test_vshr_n_u8( -// CHECK-SAME: <8 x i8> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[VSHR_N:%.*]] = lshr <8 x i8> [[A]], splat (i8 3) -// CHECK-NEXT: ret <8 x i8> [[VSHR_N]] -// -uint8x8_t test_vshr_n_u8(uint8x8_t a) { - return vshr_n_u8(a, 3); -} - -// CHECK-LABEL: define dso_local <4 x i16> @test_vshr_n_u16( -// CHECK-SAME: <4 x i16> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i16> [[A]] to <8 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16> -// CHECK-NEXT: [[VSHR_N:%.*]] = lshr <4 x i16> [[TMP1]], splat (i16 3) -// CHECK-NEXT: ret <4 x i16> [[VSHR_N]] -// -uint16x4_t test_vshr_n_u16(uint16x4_t a) { - return vshr_n_u16(a, 3); -} - -// CHECK-LABEL: define dso_local <2 x i32> @test_vshr_n_u32( -// CHECK-SAME: <2 x i32> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x i32> [[A]] to <8 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32> -// CHECK-NEXT: [[VSHR_N:%.*]] = lshr <2 x i32> [[TMP1]], splat (i32 3) -// CHECK-NEXT: ret <2 x i32> [[VSHR_N]] -// -uint32x2_t test_vshr_n_u32(uint32x2_t a) { - return vshr_n_u32(a, 3); -} - -// CHECK-LABEL: define dso_local <16 x i8> @test_vshrq_n_u8( -// CHECK-SAME: <16 x i8> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[VSHR_N:%.*]] = lshr <16 x i8> [[A]], splat (i8 3) -// CHECK-NEXT: ret <16 x i8> [[VSHR_N]] -// -uint8x16_t test_vshrq_n_u8(uint8x16_t a) { - return vshrq_n_u8(a, 3); -} - -// CHECK-LABEL: define dso_local <8 x i16> @test_vshrq_n_u16( -// CHECK-SAME: <8 x i16> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <8 x i16> [[A]] to <16 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16> -// CHECK-NEXT: [[VSHR_N:%.*]] = lshr <8 x i16> [[TMP1]], splat (i16 3) -// CHECK-NEXT: ret <8 x i16> [[VSHR_N]] -// -uint16x8_t test_vshrq_n_u16(uint16x8_t a) { - return vshrq_n_u16(a, 3); -} - -// CHECK-LABEL: define dso_local <4 x i32> @test_vshrq_n_u32( -// CHECK-SAME: <4 x i32> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <4 x i32> [[A]] to <16 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32> -// CHECK-NEXT: [[VSHR_N:%.*]] = lshr <4 x i32> [[TMP1]], splat (i32 3) -// CHECK-NEXT: ret <4 x i32> [[VSHR_N]] -// -uint32x4_t test_vshrq_n_u32(uint32x4_t a) { - return vshrq_n_u32(a, 3); -} - -// CHECK-LABEL: define dso_local <2 x i64> @test_vshrq_n_u64( -// CHECK-SAME: <2 x i64> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <2 x i64> [[A]] to <16 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <2 x i64> -// CHECK-NEXT: [[VSHR_N:%.*]] = lshr <2 x i64> [[TMP1]], splat (i64 3) -// CHECK-NEXT: ret <2 x i64> [[VSHR_N]] -// -uint64x2_t test_vshrq_n_u64(uint64x2_t a) { - return vshrq_n_u64(a, 3); -} - // CHECK-LABEL: define dso_local <8 x i8> @test_vsra_n_s8( // CHECK-SAME: <8 x i8> noundef [[A:%.*]], <8 x i8> noundef [[B:%.*]]) #[[ATTR0]] { // CHECK-NEXT: [[ENTRY:.*:]] @@ -17677,37 +17517,6 @@ uint64_t test_vcaltd_f64(float64_t a, float64_t b) { return (uint64_t)vcaltd_f64(a, b); } -// CHECK-LABEL: define dso_local i64 @test_vshrd_n_s64( -// CHECK-SAME: i64 noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[SHRD_N:%.*]] = ashr i64 [[A]], 1 -// CHECK-NEXT: ret i64 [[SHRD_N]] -// -int64_t test_vshrd_n_s64(int64_t a) { - return (int64_t)vshrd_n_s64(a, 1); -} - -// CHECK-LABEL: define dso_local <1 x i64> @test_vshr_n_s64( -// CHECK-SAME: <1 x i64> noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[TMP0:%.*]] = bitcast <1 x i64> [[A]] to <8 x i8> -// CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x i64> -// CHECK-NEXT: [[VSHR_N:%.*]] = ashr <1 x i64> [[TMP1]], splat (i64 1) -// CHECK-NEXT: ret <1 x i64> [[VSHR_N]] -// -int64x1_t test_vshr_n_s64(int64x1_t a) { - return vshr_n_s64(a, 1); -} - -// CHECK-LABEL: define dso_local i64 @test_vshrd_n_u64( -// CHECK-SAME: i64 noundef [[A:%.*]]) #[[ATTR0]] { -// CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: ret i64 0 -// -uint64_t test_vshrd_n_u64(uint64_t a) { - return (uint64_t)vshrd_n_u64(a, 64); -} - // CHECK-LABEL: define dso_local i64 @test_vshrd_n_u64_2( // CHECK-SAME: ) #[[ATTR0]] { // CHECK-NEXT: [[ENTRY:.*:]] _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
