Skip to content

Commit 2c20a9b

Browse files
RKSimonRKSimon
committed
[X86] avx512bf16-builtins.c / avx512vlbf16-builtins.c - add C/C++ and 32/64-bit test coverage
1 parent 3a8f579 commit 2c20a9b

File tree

2 files changed

+90
-127
lines changed

2 files changed

+90
-127
lines changed

‎clang/test/CodeGen/X86/avx512bf16-builtins.c‎

Lines changed: 29 additions & 40 deletions
Original file line numberDiff line numberDiff line change
@@ -1,107 +1,96 @@
1-
// RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin \
2-
// RUN: -target-feature +avx512bf16 -emit-llvm -o - -Wall -Werror \
3-
// RUN: | FileCheck %s
1+
// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512bf16 -emit-llvm -o - -Wall -Werror | FileCheck %s
2+
// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512bf16 -emit-llvm -o - -Wall -Werror | FileCheck %s
3+
// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512bf16 -emit-llvm -o - -Wall -Werror | FileCheck %s
4+
// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512bf16 -emit-llvm -o - -Wall -Werror | FileCheck %s
45

56
#include <immintrin.h>
67

78
float test_mm_cvtsbh_ss(__bf16 A) {
8-
// CHECK-LABEL: @test_mm_cvtsbh_ss
9+
// CHECK-LABEL: test_mm_cvtsbh_ss
910
// CHECK: fpext bfloat %{{.*}} to float
1011
// CHECK: ret float %{{.*}}
1112
return _mm_cvtsbh_ss(A);
1213
}
1314

1415
__m512bh test_mm512_cvtne2ps_pbh(__m512 A, __m512 B) {
15-
// CHECK-LABEL: @test_mm512_cvtne2ps_pbh
16-
// CHECK: @llvm.x86.avx512bf16.cvtne2ps2bf16.512
17-
// CHECK: ret <32 x bfloat> %{{.*}}
16+
// CHECK-LABEL: test_mm512_cvtne2ps_pbh
17+
// CHECK: call {{.*}}<32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> %{{.*}}, <16 x float> %{{.*}})
1818
return _mm512_cvtne2ps_pbh(A, B);
1919
}
2020

2121
__m512bh test_mm512_maskz_cvtne2ps_pbh(__m512 A, __m512 B, __mmask32 U) {
22-
// CHECK-LABEL: @test_mm512_maskz_cvtne2ps_pbh
23-
// CHECK: @llvm.x86.avx512bf16.cvtne2ps2bf16.512
22+
// CHECK-LABEL: test_mm512_maskz_cvtne2ps_pbh
23+
// CHECK: call {{.*}}<32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> %{{.*}}, <16 x float> %{{.*}})
2424
// CHECK: select <32 x i1> %{{.*}}, <32 x bfloat> %{{.*}}, <32 x bfloat> %{{.*}}
25-
// CHECK: ret <32 x bfloat> %{{.*}}
2625
return _mm512_maskz_cvtne2ps_pbh(U, A, B);
2726
}
2827

2928
__m512bh test_mm512_mask_cvtne2ps_pbh(__m512bh C, __mmask32 U, __m512 A, __m512 B) {
30-
// CHECK-LABEL: @test_mm512_mask_cvtne2ps_pbh
31-
// CHECK: @llvm.x86.avx512bf16.cvtne2ps2bf16.512
29+
// CHECK-LABEL: test_mm512_mask_cvtne2ps_pbh
30+
// CHECK: call {{.*}}<32 x bfloat> @llvm.x86.avx512bf16.cvtne2ps2bf16.512(<16 x float> %{{.*}}, <16 x float> %{{.*}})
3231
// CHECK: select <32 x i1> %{{.*}}, <32 x bfloat> %{{.*}}, <32 x bfloat> %{{.*}}
33-
// CHECK: ret <32 x bfloat> %{{.*}}
3432
return _mm512_mask_cvtne2ps_pbh(C, U, A, B);
3533
}
3634

3735
__m256bh test_mm512_cvtneps_pbh(__m512 A) {
38-
// CHECK-LABEL: @test_mm512_cvtneps_pbh
39-
// CHECK: @llvm.x86.avx512bf16.cvtneps2bf16.512
40-
// CHECK: ret <16 x bfloat> %{{.*}}
36+
// CHECK-LABEL: test_mm512_cvtneps_pbh
37+
// CHECK: call {{.*}}<16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %{{.*}})
4138
return _mm512_cvtneps_pbh(A);
4239
}
4340

4441
__m256bh test_mm512_mask_cvtneps_pbh(__m256bh C, __mmask16 U, __m512 A) {
45-
// CHECK-LABEL: @test_mm512_mask_cvtneps_pbh
46-
// CHECK: @llvm.x86.avx512bf16.cvtneps2bf16.512
42+
// CHECK-LABEL: test_mm512_mask_cvtneps_pbh
43+
// CHECK: call {{.*}}<16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %{{.*}})
4744
// CHECK: select <16 x i1> %{{.*}}, <16 x bfloat> %{{.*}}, <16 x bfloat> %{{.*}}
48-
// CHECK: ret <16 x bfloat> %{{.*}}
4945
return _mm512_mask_cvtneps_pbh(C, U, A);
5046
}
5147

5248
__m256bh test_mm512_maskz_cvtneps_pbh(__m512 A, __mmask16 U) {
53-
// CHECK-LABEL: @test_mm512_maskz_cvtneps_pbh
54-
// CHECK: @llvm.x86.avx512bf16.cvtneps2bf16.512
49+
// CHECK-LABEL: test_mm512_maskz_cvtneps_pbh
50+
// CHECK: call {{.*}}<16 x bfloat> @llvm.x86.avx512bf16.cvtneps2bf16.512(<16 x float> %{{.*}})
5551
// CHECK: select <16 x i1> %{{.*}}, <16 x bfloat> %{{.*}}, <16 x bfloat> %{{.*}}
56-
// CHECK: ret <16 x bfloat> %{{.*}}
5752
return _mm512_maskz_cvtneps_pbh(U, A);
5853
}
5954

6055
__m512 test_mm512_dpbf16_ps(__m512 D, __m512bh A, __m512bh B) {
61-
// CHECK-LABEL: @test_mm512_dpbf16_ps
62-
// CHECK: @llvm.x86.avx512bf16.dpbf16ps.512
63-
// CHECK: ret <16 x float> %{{.*}}
56+
// CHECK-LABEL: test_mm512_dpbf16_ps
57+
// CHECK: call {{.*}}<16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> %{{.*}}, <32 x bfloat> %{{.*}}, <32 x bfloat> %{{.*}})
6458
return _mm512_dpbf16_ps(D, A, B);
6559
}
6660

6761
__m512 test_mm512_maskz_dpbf16_ps(__m512 D, __m512bh A, __m512bh B, __mmask16 U) {
68-
// CHECK-LABEL: @test_mm512_maskz_dpbf16_ps
69-
// CHECK: @llvm.x86.avx512bf16.dpbf16ps.512
62+
// CHECK-LABEL: test_mm512_maskz_dpbf16_ps
63+
// CHECK: call {{.*}}<16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> %{{.*}}, <32 x bfloat> %{{.*}}, <32 x bfloat> %{{.*}})
7064
// CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
71-
// CHECK: ret <16 x float> %{{.*}}
7265
return _mm512_maskz_dpbf16_ps(U, D, A, B);
7366
}
7467

7568
__m512 test_mm512_mask_dpbf16_ps(__m512 D, __m512bh A, __m512bh B, __mmask16 U) {
76-
// CHECK-LABEL: @test_mm512_mask_dpbf16_ps
77-
// CHECK: @llvm.x86.avx512bf16.dpbf16ps.512
69+
// CHECK-LABEL: test_mm512_mask_dpbf16_ps
70+
// CHECK: call {{.*}}<16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> %{{.*}}, <32 x bfloat> %{{.*}}, <32 x bfloat> %{{.*}})
7871
// CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
79-
// CHECK: ret <16 x float> %{{.*}}
8072
return _mm512_mask_dpbf16_ps(D, U, A, B);
8173
}
8274

8375
__m512 test_mm512_cvtpbh_ps(__m256bh A) {
84-
// CHECK-LABEL: @test_mm512_cvtpbh_ps
76+
// CHECK-LABEL: test_mm512_cvtpbh_ps
8577
// CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
86-
// CHECK: @llvm.x86.avx512.pslli.d.512
87-
// CHECK: ret <16 x float> %{{.*}}
78+
// CHECK: call <16 x i32> @llvm.x86.avx512.pslli.d.512(<16 x i32> %{{.*}}, i32 %{{.*}})
8879
return _mm512_cvtpbh_ps(A);
8980
}
9081

9182
__m512 test_mm512_maskz_cvtpbh_ps(__mmask16 M, __m256bh A) {
92-
// CHECK-LABEL: @test_mm512_maskz_cvtpbh_ps
83+
// CHECK-LABEL: test_mm512_maskz_cvtpbh_ps
9384
// CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
9485
// CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
95-
// CHECK: @llvm.x86.avx512.pslli.d.512
96-
// CHECK: ret <16 x float> %{{.*}}
86+
// CHECK: call <16 x i32> @llvm.x86.avx512.pslli.d.512(<16 x i32> %{{.*}}, i32 %{{.*}})
9787
return _mm512_maskz_cvtpbh_ps(M, A);
9888
}
9989

10090
__m512 test_mm512_mask_cvtpbh_ps(__m512 S, __mmask16 M, __m256bh A) {
101-
// CHECK-LABEL: @test_mm512_mask_cvtpbh_ps
91+
// CHECK-LABEL: test_mm512_mask_cvtpbh_ps
10292
// CHECK: sext <16 x i16> %{{.*}} to <16 x i32>
103-
// CHECK: @llvm.x86.avx512.pslli.d.512
93+
// CHECK: call <16 x i32> @llvm.x86.avx512.pslli.d.512(<16 x i32> %{{.*}}, i32 %{{.*}})
10494
// CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
105-
// CHECK: ret <16 x float> %{{.*}}
10695
return _mm512_mask_cvtpbh_ps(S, M, A);
10796
}

0 commit comments

Comments
 (0)