https://github.com/banach-space created https://github.com/llvm/llvm-project/pull/182307
This PR adds CIR lowering tests for the predicated SVE `svdup` builtins on AArch64. The corresponding ACLE intrinsics are documented at: https://developer.arm.com/architectures/instruction-sets/intrinsics The tests cover the merging-predicated variants (suffix `_m`, e.g. `svdup_n_f32_m`). These forms take an explicit inactive-lane value, which is merged into the result for lanes where the predicate is false. No functional changes are introduced. The existing lowering infrastructure already handles these builtins; this change adds test coverage only. From fd39dc33aa5093a32edb2103bf1bac65cb074e17 Mon Sep 17 00:00:00 2001 From: Andrzej Warzynski <[email protected]> Date: Thu, 19 Feb 2026 15:49:49 +0000 Subject: [PATCH] [CIR][AArch64] Add CIR tests for predicated SVE svdup_m builtins This PR adds CIR lowering tests for the predicated SVE `svdup` builtins on AArch64. The corresponding ACLE intrinsics are documented at: https://developer.arm.com/architectures/instruction-sets/intrinsics The tests cover the merging-predicated variants (suffix `_m`, e.g. `svdup_n_f32_m`). These forms take an explicit inactive-lane value, which is merged into the result for lanes where the predicate is false. No functional changes are introduced. The existing lowering infrastructure already handles these builtins; this change adds test coverage only. --- .../CodeGenBuiltins/AArch64/acle_sve_dup.c | 162 ++++++++++++++++++ 1 file changed, 162 insertions(+) diff --git a/clang/test/CIR/CodeGenBuiltins/AArch64/acle_sve_dup.c b/clang/test/CIR/CodeGenBuiltins/AArch64/acle_sve_dup.c index 8944b7400d3a3..8697b2121fb3e 100644 --- a/clang/test/CIR/CodeGenBuiltins/AArch64/acle_sve_dup.c +++ b/clang/test/CIR/CodeGenBuiltins/AArch64/acle_sve_dup.c @@ -323,3 +323,165 @@ svfloat64_t test_svdup_n_f64_z(svbool_t pg, float64_t op) MODE_ATTR // LLVM_OGCG_CIR: ret {{.*}} [[RES]] return SVE_ACLE_FUNC(svdup,_n,_f64_z,)(pg, op); } + +//===------------------------------------------------------===// +// 2. PREDICATED MERGING-ING SVDUP +//===------------------------------------------------------===// +// ALL-LABEL: @test_svdup_n_s8_m( +svint8_t test_svdup_n_s8_m(svint8_t inactive, svbool_t pg, int8_t op) MODE_ATTR +{ +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %{{.*}}, %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !s8i>, !cir.vector<[16] x !cir.int<u, 1>>, !s8i) -> !cir.vector<[16] x !s8i> + +// LLVM_OGCG_CIR-SAME: <vscale x 16 x i8> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], i8{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[TMP0:%.*]] = call <vscale x 16 x i8> @llvm.aarch64.sve.dup.nxv16i8(<vscale x 16 x i8> [[INACTIVE]], <vscale x 16 x i1> [[PG]], i8 [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 16 x i8> [[TMP0]] + return SVE_ACLE_FUNC(svdup,_n,_s8_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_s16_m( +svint16_t test_svdup_n_s16_m(svint16_t inactive, svbool_t pg, int16_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[8] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[8] x !s16i>, !cir.vector<[8] x !cir.int<u, 1>>, !s16i) -> !cir.vector<[8] x !s16i> + +// LLVM_OGCG_CIR-SAME: <vscale x 8 x i16> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], i16{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 8 x i16> @llvm.aarch64.sve.dup.nxv8i16(<vscale x 8 x i16> [[INACTIVE]], <vscale x 8 x i1> [[PG_CONVERTED]], i16 [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 8 x i16> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_s16_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_s32_m( +svint32_t test_svdup_n_s32_m(svint32_t inactive, svbool_t pg, int32_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[4] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[4] x !s32i>, !cir.vector<[4] x !cir.int<u, 1>>, !s32i) -> !cir.vector<[4] x !s32i> + +// LLVM_OGCG_CIR-SAME: <vscale x 4 x i32> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], i32{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.dup.nxv4i32(<vscale x 4 x i32> [[INACTIVE]], <vscale x 4 x i1> [[PG_CONVERTED]], i32 [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 4 x i32> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_s32_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_s64_m( +svint64_t test_svdup_n_s64_m(svint64_t inactive, svbool_t pg, int64_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[2] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[2] x !s64i>, !cir.vector<[2] x !cir.int<u, 1>>, !s64i) -> !cir.vector<[2] x !s64i> + +// LLVM_OGCG_CIR-SAME: <vscale x 2 x i64> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], i64{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.nxv2i64(<vscale x 2 x i64> [[INACTIVE]], <vscale x 2 x i1> [[PG_CONVERTED]], i64 [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 2 x i64> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_s64_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_u8_m( +svuint8_t test_svdup_n_u8_m(svuint8_t inactive, svbool_t pg, uint8_t op) MODE_ATTR +{ +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %{{.*}}, %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !u8i>, !cir.vector<[16] x !cir.int<u, 1>>, !u8i) -> !cir.vector<[16] x !u8i> + +// LLVM_OGCG_CIR-SAME: <vscale x 16 x i8> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], i8{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 16 x i8> @llvm.aarch64.sve.dup.nxv16i8(<vscale x 16 x i8> [[INACTIVE]], <vscale x 16 x i1> [[PG]], i8 [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 16 x i8> [[PG_CONVERTED]] + return SVE_ACLE_FUNC(svdup,_n,_u8_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_u16_m( +svuint16_t test_svdup_n_u16_m(svuint16_t inactive, svbool_t pg, uint16_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[8] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[8] x !u16i>, !cir.vector<[8] x !cir.int<u, 1>>, !u16i) -> !cir.vector<[8] x !u16i> + +// LLVM_OGCG_CIR-SAME: <vscale x 8 x i16> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], i16{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 8 x i16> @llvm.aarch64.sve.dup.nxv8i16(<vscale x 8 x i16> [[INACTIVE]], <vscale x 8 x i1> [[PG_CONVERTED]], i16 [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 8 x i16> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_u16_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_u32_m( +svuint32_t test_svdup_n_u32_m(svuint32_t inactive, svbool_t pg, uint32_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[4] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[4] x !u32i>, !cir.vector<[4] x !cir.int<u, 1>>, !u32i) -> !cir.vector<[4] x !u32i> + +// LLVM_OGCG_CIR-SAME: <vscale x 4 x i32> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], i32{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.dup.nxv4i32(<vscale x 4 x i32> [[INACTIVE]], <vscale x 4 x i1> [[PG_CONVERTED]], i32 [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 4 x i32> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_u32_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_u64_m( +svuint64_t test_svdup_n_u64_m(svuint64_t inactive, svbool_t pg, uint64_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[2] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[2] x !u64i>, !cir.vector<[2] x !cir.int<u, 1>>, !u64i) -> !cir.vector<[2] x !u64i> + +// LLVM_OGCG_CIR-SAME: <vscale x 2 x i64> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], i64{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.nxv2i64(<vscale x 2 x i64> [[INACTIVE]], <vscale x 2 x i1> [[PG_CONVERTED]], i64 [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 2 x i64> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_u64_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_f16_m( +svfloat16_t test_svdup_n_f16_m(svfloat16_t inactive, svbool_t pg, float16_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[8] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[8] x !cir.f16>, !cir.vector<[8] x !cir.int<u, 1>>, !cir.f16) -> !cir.vector<[8] x !cir.f16> + +// LLVM_OGCG_CIR-SAME: <vscale x 8 x half> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], half{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 8 x half> @llvm.aarch64.sve.dup.nxv8f16(<vscale x 8 x half> [[INACTIVE]], <vscale x 8 x i1> [[PG_CONVERTED]], half [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 8 x half> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_f16_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_f32_m( +svfloat32_t test_svdup_n_f32_m(svfloat32_t inactive, svbool_t pg, float32_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[4] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[4] x !cir.float>, !cir.vector<[4] x !cir.int<u, 1>>, !cir.float) -> !cir.vector<[4] x !cir.float> + +// LLVM_OGCG_CIR-SAME: <vscale x 4 x float> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], float{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.dup.nxv4f32(<vscale x 4 x float> [[INACTIVE]], <vscale x 4 x i1> [[PG_CONVERTED]], float [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 4 x float> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_f32_m,)(inactive, pg, op); +} + +// ALL-LABEL: @test_svdup_n_f64_m( +svfloat64_t test_svdup_n_f64_m(svfloat64_t inactive, svbool_t pg, float64_t op) MODE_ATTR +{ +// CIR: %[[CONVERT_PG:.*]] = cir.call_llvm_intrinsic "aarch64.sve.convert.from.svbool" %{{.*}} : +// CIR-SAME: (!cir.vector<[16] x !cir.int<u, 1>>) -> !cir.vector<[2] x !cir.int<u, 1>> +// CIR: cir.call_llvm_intrinsic "aarch64.sve.dup" %{{.*}}, %[[CONVERT_PG]], %{{.*}} : +// CIR-SAME: (!cir.vector<[2] x !cir.double>, !cir.vector<[2] x !cir.int<u, 1>>, !cir.double) -> !cir.vector<[2] x !cir.double> + +// LLVM_OGCG_CIR-SAME: <vscale x 2 x double> [[INACTIVE:%.*]], <vscale x 16 x i1> [[PG:%.*]], double{{.*}} [[OP:%.*]]) +// LLVM_OGCG_CIR: [[PG_CONVERTED:%.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PG]]) +// LLVM_OGCG_CIR-NEXT: [[RES:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.dup.nxv2f64(<vscale x 2 x double> [[INACTIVE]], <vscale x 2 x i1> [[PG_CONVERTED]], double [[OP]]) +// LLVM_OGCG_CIR-NEXT: ret <vscale x 2 x double> [[RES]] + return SVE_ACLE_FUNC(svdup,_n,_f64_m,)(inactive, pg, op); +} _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
