-
Notifications
You must be signed in to change notification settings - Fork 10.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[AArch64][SME2] Add multi-vector builtins for cvt #74450
Conversation
Adds builtins for: - FCVT - BFCVT - FCVTZS - FCVTZU - SCVTF - UCVTF - BFCVTN - FCVTN - SQCVT - SQCVTU - UQCVT - SQCVTN - SQCVTUN - UQCVTN See ARM-software/acle#217
@llvm/pr-subscribers-clang Author: Matthew Devereau (MDevereau) ChangesAdds builtins for:
Patch is 71.11 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/74450.diff 3 Files Affected:
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 4fcc9327f22fe..a73ac75e3bbc0 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -2149,3 +2149,63 @@ let TargetGuard = "sve2p1" in {
defm SVPMOV_TO_VEC_LANE_S : PMOV_TO_VEC<"svpmov", "iUi", "aarch64_sve_pmov_to_vector_lane", [], ImmCheck1_3>;
defm SVPMOV_TO_VEC_LANE_D : PMOV_TO_VEC<"svpmov", "lUl", "aarch64_sve_pmov_to_vector_lane" ,[], ImmCheck1_7>;
}
+
+//
+// Multi-vector convert to/from floating-point.
+//
+let TargetGuard = "sme2" in {
+ def SVCVT_F16_X2 : SInst<"svcvt_f16[_f32_x2]", "e2", "f", MergeNone, "aarch64_sve_fcvt_x2", [IsStreaming],[]>;
+ def SVCVT_BF16_X2 : SInst<"svcvt_bf16[_f32_x2]", "$2", "f", MergeNone, "aarch64_sve_bfcvt_x2", [IsOverloadNone, IsStreaming],[]>;
+
+ def SVCVT_F32_U32_X2 : SInst<"svcvt_{d}[_u32_x2]", "2.d2.u", "f", MergeNone, "aarch64_sve_fcvtu_x2", [IsStreaming], []>;
+ def SVCVT_U32_F32_X2 : SInst<"svcvt_u32[_{d}_x2]", "2.u2.d", "f", MergeNone, "aarch64_sve_ucvtf_x2", [IsStreaming], []>;
+ def SVCVT_F32_S32_X2 : SInst<"svcvt_{d}[_s32_x2]", "2.d2.x", "f", MergeNone, "aarch64_sve_fcvts_x2", [IsStreaming], []>;
+ def SVCVT_S32_F32_X2 : SInst<"svcvt_s32[_{d}_x2]", "2.x2.d", "f", MergeNone, "aarch64_sve_scvtf_x2", [IsStreaming], []>;
+
+ def SVCVT_F32_U32_X4 : SInst<"svcvt_{d}[_u32_x4]", "4.d4.u", "f", MergeNone, "aarch64_sve_fcvtu_x4", [IsStreaming], []>;
+ def SVCVT_U32_F32_X4 : SInst<"svcvt_u32[_{d}_x4]", "4.u4.d", "f", MergeNone, "aarch64_sve_ucvtf_x4", [IsStreaming], []>;
+ def SVCVT_F32_S32_X4 : SInst<"svcvt_{d}[_s32_x4]", "4.d4.x", "f", MergeNone, "aarch64_sve_fcvts_x4", [IsStreaming], []>;
+ def SVCVT_S32_F32_X4 : SInst<"svcvt_s32[_{d}_x4]", "4.x4.d", "f", MergeNone, "aarch64_sve_scvtf_x4", [IsStreaming], []>;
+}
+
+//
+// Multi-vector floating-point convert from single-precision to interleaved half-precision/BFloat16
+//
+let TargetGuard = "sme2" in {
+ def SVCVTN_F16_X2 : SInst<"svcvtn_f16[_f32_x2]", "e2", "f", MergeNone, "aarch64_sve_fcvtn_x2", [IsStreaming],[]>;
+ def SVCVTN_BF16_X2 : SInst<"svcvtn_bf16[_f32_x2]", "$2", "f", MergeNone, "aarch64_sve_bfcvtn_x2", [IsOverloadNone, IsStreaming],[]>;
+}
+
+//
+// Multi-vector saturating extract narrow
+//
+let TargetGuard = "sme2" in {
+ def SVQCVT_S16_S32_X2 : SInst<"svqcvt_s16[_{d}_x2]", "h2.d", "i", MergeNone, "aarch64_sve_sqcvt_x2", [IsStreaming], []>;
+ def SVQCVT_U16_U32_X2 : SInst<"svqcvt_u16[_{d}_x2]", "e2.d", "Ui", MergeNone, "aarch64_sve_uqcvt_x2", [IsStreaming], []>;
+ def SVQCVT_U16_S32_X2 : SInst<"svqcvt_u16[_{d}_x2]", "e2.d", "i", MergeNone, "aarch64_sve_sqcvtu_x2", [IsStreaming], []>;
+
+ def SVQCVT_S8_S32_X4 : SInst<"svqcvt_s8[_{d}_x4]", "q4.d", "i", MergeNone, "aarch64_sve_sqcvt_x4", [IsStreaming], []>;
+ def SVQCVT_U8_U32_X4 : SInst<"svqcvt_u8[_{d}_x4]", "b4.d", "Ui", MergeNone, "aarch64_sve_uqcvt_x4", [IsStreaming], []>;
+ def SVQCVT_U8_S32_X4 : SInst<"svqcvt_u8[_{d}_x4]", "b4.d", "i", MergeNone, "aarch64_sve_sqcvtu_x4", [IsStreaming], []>;
+
+ def SVQCVT_S16_S64_X4 : SInst<"svqcvt_s16[_{d}_x4]", "q4.d", "l", MergeNone, "aarch64_sve_sqcvt_x4", [IsStreaming], []>;
+ def SVQCVT_U16_U64_X4 : SInst<"svqcvt_u16[_{d}_x4]", "b4.d", "Ul", MergeNone, "aarch64_sve_uqcvt_x4", [IsStreaming], []>;
+ def SVQCVT_U16_S64_X4 : SInst<"svqcvt_u16[_{d}_x4]", "b4.d", "l", MergeNone, "aarch64_sve_sqcvtu_x4", [IsStreaming], []>;
+}
+
+//
+// Multi-vector saturating extract narrow and interleave
+//
+let TargetGuard = "sme2" in {
+ def SVQCVTN_S16_S32_X2 : SInst<"svqcvtn_s16[_{d}_x2]", "h2.d", "i", MergeNone, "aarch64_sve_sqcvtn_x2", [IsStreamingCompatible], []>;
+ def SVQCVTN_U16_U32_X2 : SInst<"svqcvtn_u16[_{d}_x2]", "e2.d", "Ui", MergeNone, "aarch64_sve_uqcvtn_x2", [IsStreamingCompatible], []>;
+ def SVQCVTN_U16_S32_X2 : SInst<"svqcvtn_u16[_{d}_x2]", "e2.d", "i", MergeNone, "aarch64_sve_sqcvtun_x2", [IsStreamingCompatible], []>;
+
+ def SVQCVTN_S8_S32_X4 : SInst<"svqcvtn_s8[_{d}_x4]", "q4.d", "i", MergeNone, "aarch64_sve_sqcvtn_x4", [IsStreaming], []>;
+ def SVQCVTN_U8_U32_X4 : SInst<"svqcvtn_u8[_{d}_x4]", "b4.d", "Ui", MergeNone, "aarch64_sve_uqcvtn_x4", [IsStreaming], []>;
+ def SVQCVTN_U8_S32_X4 : SInst<"svqcvtn_u8[_{d}_x4]", "b4.d", "i", MergeNone, "aarch64_sve_sqcvtun_x4", [IsStreaming], []>;
+
+ def SVQCVTN_S16_S64_X4 : SInst<"svqcvtn_s16[_{d}_x4]", "q4.d", "l", MergeNone, "aarch64_sve_sqcvtn_x4", [IsStreaming], []>;
+ def SVQCVTN_U16_U64_X4 : SInst<"svqcvtn_u16[_{d}_x4]", "b4.d", "Ul", MergeNone, "aarch64_sve_uqcvtn_x4", [IsStreaming], []>;
+ def SVQCVTN_U16_S64_X4 : SInst<"svqcvtn_u16[_{d}_x4]", "b4.d", "l", MergeNone, "aarch64_sve_sqcvtun_x4", [IsStreaming], []>;
+}
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvt.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvt.c
new file mode 100644
index 0000000000000..a3ee7d2092f79
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_cvt.c
@@ -0,0 +1,499 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+
+// REQUIRES: aarch64-registered-target
+
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -D__SVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -D__SVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+
+#include <arm_sme_draft_spec_subject_to_change.h>
+
+#ifdef SVE_OVERLOADED_FORMS
+// A simple used,unused... macro, long enough to represent any SVE builtin.
+#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED) A1##A3
+#else
+#define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
+#endif
+
+// CHECK-LABEL: @test_cvt_f16_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvt.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT: ret <vscale x 8 x half> [[TMP2]]
+//
+// CPP-CHECK-LABEL: @_Z15test_cvt_f16_x213svfloat32x2_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.fcvt.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT: ret <vscale x 8 x half> [[TMP2]]
+//
+svfloat16_t test_cvt_f16_x2(svfloat32x2_t zn) __arm_streaming {
+ return SVE_ACLE_FUNC(svcvt_f16,_f32_x2,,)(zn);
+}
+
+// CHECK-LABEL: @test_cvt_bf16_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvt.x2(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT: ret <vscale x 8 x bfloat> [[TMP2]]
+//
+// CPP-CHECK-LABEL: @_Z16test_cvt_bf16_x213svfloat32x2_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.bfcvt.x2(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT: ret <vscale x 8 x bfloat> [[TMP2]]
+//
+svbfloat16_t test_cvt_bf16_x2(svfloat32x2_t zn) __arm_streaming {
+ return SVE_ACLE_FUNC(svcvt_bf16,_f32_x2,,)(zn);
+}
+
+
+// x2
+// CHECK-LABEL: @test_svcvt_f32_u32_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvtu.x2.nxv4f32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
+// CHECK-NEXT: ret <vscale x 8 x float> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svcvt_f32_u32_x212svuint32x2_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvtu.x2.nxv4f32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT: ret <vscale x 8 x float> [[TMP6]]
+//
+svfloat32x2_t test_svcvt_f32_u32_x2(svuint32x2_t zn) __arm_streaming {
+ return SVE_ACLE_FUNC(svcvt_f32,_u32_x2,,)(zn);
+}
+
+// CHECK-LABEL: @test_svcvt_f32_s32_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvts.x2.nxv4f32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
+// CHECK-NEXT: ret <vscale x 8 x float> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svcvt_f32_s32_x211svint32x2_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[ZN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvts.x2.nxv4f32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]])
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT: ret <vscale x 8 x float> [[TMP6]]
+//
+svfloat32x2_t test_svcvt_f32_s32_x2(svint32x2_t zn) __arm_streaming {
+ return SVE_ACLE_FUNC(svcvt_f32,_s32_x2,,)(zn);
+}
+
+// CHECK-LABEL: @test_svcvt_u32_f32_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ucvtf.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
+// CHECK-NEXT: ret <vscale x 8 x i32> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svcvt_u32_f32_x213svfloat32x2_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ucvtf.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT: ret <vscale x 8 x i32> [[TMP6]]
+//
+svuint32x2_t test_svcvt_u32_f32_x2(svfloat32x2_t zn) __arm_streaming {
+ return SVE_ACLE_FUNC(svcvt_u32,_f32_x2,,)(zn);
+}
+
+// CHECK-LABEL: @test_svcvt_s32_f32_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.scvtf.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
+// CHECK-NEXT: ret <vscale x 8 x i32> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svcvt_s32_f32_x213svfloat32x2_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.scvtf.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]])
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 0
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], 1
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], i64 4)
+// CPP-CHECK-NEXT: ret <vscale x 8 x i32> [[TMP6]]
+//
+svint32x2_t test_svcvt_s32_f32_x2(svfloat32x2_t zn) __arm_streaming {
+ return SVE_ACLE_FUNC(svcvt_s32,_f32_x2,,)(zn);
+}
+
+// x4
+// CHECK-LABEL: @test_svcvt_f32_u32_x4(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 8)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[ZN]], i64 12)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.fcvtu.x4.nxv4f32(<vscale x 4 x i32> [[TMP0]], <vscale x 4 x i32> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
+// CHECK-NEXT: [[TMP9:%.*]] = extr...
[truncated]
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM!
|
||
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s | ||
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK | ||
// RUN: %clang_cc1 -D__SVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I used -target-feature +b16b16 for BF16 support, not sure now how correct now, I am waiting for Carol to resolve a correct nameing after that I can rebase my change.
Adds builtins for:
See ARM-software/acle#217