1 // RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +bf16 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
2 // RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +bf16 -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
3
4 #include <arm_sve.h>
5
6 #ifdef SVE_OVERLOADED_FORMS
7 // A simple used,unused... macro, long enough to represent any SVE builtin.
8 #define SVE_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED) A1##A3
9 #else
10 #define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
11 #endif
12
test_svld3_bf16(svbool_t pg,const bfloat16_t * base)13 svbfloat16x3_t test_svld3_bf16(svbool_t pg, const bfloat16_t *base)
14 {
15 // CHECK-LABEL: test_svld3_bf16
16 // CHECK: %[[PG:.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %pg)
17 // CHECK: %[[LOAD:.*]] = call <vscale x 24 x bfloat> @llvm.aarch64.sve.ld3.nxv24bf16.nxv8i1(<vscale x 8 x i1> %[[PG]], bfloat* %base)
18 // CHECK-NEXT: ret <vscale x 24 x bfloat> %[[LOAD]]
19 return SVE_ACLE_FUNC(svld3,_bf16,,)(pg, base);
20 }
21
test_svld3_vnum_bf16(svbool_t pg,const bfloat16_t * base,int64_t vnum)22 svbfloat16x3_t test_svld3_vnum_bf16(svbool_t pg, const bfloat16_t *base, int64_t vnum)
23 {
24 // CHECK-LABEL: test_svld3_vnum_bf16
25 // CHECK-DAG: %[[PG:.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %pg)
26 // CHECK-DAG: %[[BASE:.*]] = bitcast bfloat* %base to <vscale x 8 x bfloat>*
27 // CHECK-DAG: %[[GEP:.*]] = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* %[[BASE]], i64 %vnum, i64 0
28 // CHECK: %[[LOAD:.*]] = call <vscale x 24 x bfloat> @llvm.aarch64.sve.ld3.nxv24bf16.nxv8i1(<vscale x 8 x i1> %[[PG]], bfloat* %[[GEP]])
29 // CHECK-NEXT: ret <vscale x 24 x bfloat> %[[LOAD]]
30 return SVE_ACLE_FUNC(svld3_vnum,_bf16,,)(pg, base, vnum);
31 }
32