1 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
2 // REQUIRES: amdgpu-registered-target
3 // REQUIRES: x86-registered-target
5 // RUN: %clang_cc1 "-aux-triple" "x86_64-unknown-linux-gnu" "-triple" "amdgcn-amd-amdhsa" \
6 // RUN: -fcuda-is-device "-aux-target-cpu" "x86-64" -emit-llvm -o - %s | FileCheck %s
8 #include "Inputs/cuda.h"
10 // CHECK-LABEL: @_Z8test_argPDF16bDF16b(
12 // CHECK-NEXT: [[OUT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
13 // CHECK-NEXT: [[IN_ADDR:%.*]] = alloca bfloat, align 2, addrspace(5)
14 // CHECK-NEXT: [[BF16:%.*]] = alloca bfloat, align 2, addrspace(5)
15 // CHECK-NEXT: [[OUT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[OUT_ADDR]] to ptr
16 // CHECK-NEXT: [[IN_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[IN_ADDR]] to ptr
17 // CHECK-NEXT: [[BF16_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[BF16]] to ptr
18 // CHECK-NEXT: store ptr [[OUT:%.*]], ptr [[OUT_ADDR_ASCAST]], align 8
19 // CHECK-NEXT: store bfloat [[IN:%.*]], ptr [[IN_ADDR_ASCAST]], align 2
20 // CHECK-NEXT: [[TMP0:%.*]] = load bfloat, ptr [[IN_ADDR_ASCAST]], align 2
21 // CHECK-NEXT: store bfloat [[TMP0]], ptr [[BF16_ASCAST]], align 2
22 // CHECK-NEXT: [[TMP1:%.*]] = load bfloat, ptr [[BF16_ASCAST]], align 2
23 // CHECK-NEXT: [[TMP2:%.*]] = load ptr, ptr [[OUT_ADDR_ASCAST]], align 8
24 // CHECK-NEXT: store bfloat [[TMP1]], ptr [[TMP2]], align 2
25 // CHECK-NEXT: ret void
27 __device__ void test_arg(__bf16 *out, __bf16 in) {
32 // CHECK-LABEL: @_Z9test_loadPDF16bS_(
34 // CHECK-NEXT: [[OUT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
35 // CHECK-NEXT: [[IN_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
36 // CHECK-NEXT: [[BF16:%.*]] = alloca bfloat, align 2, addrspace(5)
37 // CHECK-NEXT: [[OUT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[OUT_ADDR]] to ptr
38 // CHECK-NEXT: [[IN_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[IN_ADDR]] to ptr
39 // CHECK-NEXT: [[BF16_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[BF16]] to ptr
40 // CHECK-NEXT: store ptr [[OUT:%.*]], ptr [[OUT_ADDR_ASCAST]], align 8
41 // CHECK-NEXT: store ptr [[IN:%.*]], ptr [[IN_ADDR_ASCAST]], align 8
42 // CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[IN_ADDR_ASCAST]], align 8
43 // CHECK-NEXT: [[TMP1:%.*]] = load bfloat, ptr [[TMP0]], align 2
44 // CHECK-NEXT: store bfloat [[TMP1]], ptr [[BF16_ASCAST]], align 2
45 // CHECK-NEXT: [[TMP2:%.*]] = load bfloat, ptr [[BF16_ASCAST]], align 2
46 // CHECK-NEXT: [[TMP3:%.*]] = load ptr, ptr [[OUT_ADDR_ASCAST]], align 8
47 // CHECK-NEXT: store bfloat [[TMP2]], ptr [[TMP3]], align 2
48 // CHECK-NEXT: ret void
50 __device__ void test_load(__bf16 *out, __bf16 *in) {
55 // CHECK-LABEL: @_Z8test_retDF16b(
57 // CHECK-NEXT: [[RETVAL:%.*]] = alloca bfloat, align 2, addrspace(5)
58 // CHECK-NEXT: [[IN_ADDR:%.*]] = alloca bfloat, align 2, addrspace(5)
59 // CHECK-NEXT: [[RETVAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL]] to ptr
60 // CHECK-NEXT: [[IN_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[IN_ADDR]] to ptr
61 // CHECK-NEXT: store bfloat [[IN:%.*]], ptr [[IN_ADDR_ASCAST]], align 2
62 // CHECK-NEXT: [[TMP0:%.*]] = load bfloat, ptr [[IN_ADDR_ASCAST]], align 2
63 // CHECK-NEXT: ret bfloat [[TMP0]]
65 __device__ __bf16 test_ret( __bf16 in) {
69 // CHECK-LABEL: @_Z9test_callDF16b(
71 // CHECK-NEXT: [[RETVAL:%.*]] = alloca bfloat, align 2, addrspace(5)
72 // CHECK-NEXT: [[IN_ADDR:%.*]] = alloca bfloat, align 2, addrspace(5)
73 // CHECK-NEXT: [[RETVAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL]] to ptr
74 // CHECK-NEXT: [[IN_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[IN_ADDR]] to ptr
75 // CHECK-NEXT: store bfloat [[IN:%.*]], ptr [[IN_ADDR_ASCAST]], align 2
76 // CHECK-NEXT: [[TMP0:%.*]] = load bfloat, ptr [[IN_ADDR_ASCAST]], align 2
77 // CHECK-NEXT: [[CALL:%.*]] = call contract noundef bfloat @_Z8test_retDF16b(bfloat noundef [[TMP0]]) #[[ATTR1:[0-9]+]]
78 // CHECK-NEXT: ret bfloat [[CALL]]
80 __device__ __bf16 test_call( __bf16 in) {
85 // CHECK-LABEL: @_Z15test_vec_assignv(
87 // CHECK-NEXT: [[VEC2_A:%.*]] = alloca <2 x bfloat>, align 4, addrspace(5)
88 // CHECK-NEXT: [[VEC2_B:%.*]] = alloca <2 x bfloat>, align 4, addrspace(5)
89 // CHECK-NEXT: [[VEC4_A:%.*]] = alloca <4 x bfloat>, align 8, addrspace(5)
90 // CHECK-NEXT: [[VEC4_B:%.*]] = alloca <4 x bfloat>, align 8, addrspace(5)
91 // CHECK-NEXT: [[VEC8_A:%.*]] = alloca <8 x bfloat>, align 16, addrspace(5)
92 // CHECK-NEXT: [[VEC8_B:%.*]] = alloca <8 x bfloat>, align 16, addrspace(5)
93 // CHECK-NEXT: [[VEC16_A:%.*]] = alloca <16 x bfloat>, align 32, addrspace(5)
94 // CHECK-NEXT: [[VEC16_B:%.*]] = alloca <16 x bfloat>, align 32, addrspace(5)
95 // CHECK-NEXT: [[VEC2_A_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VEC2_A]] to ptr
96 // CHECK-NEXT: [[VEC2_B_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VEC2_B]] to ptr
97 // CHECK-NEXT: [[VEC4_A_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VEC4_A]] to ptr
98 // CHECK-NEXT: [[VEC4_B_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VEC4_B]] to ptr
99 // CHECK-NEXT: [[VEC8_A_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VEC8_A]] to ptr
100 // CHECK-NEXT: [[VEC8_B_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VEC8_B]] to ptr
101 // CHECK-NEXT: [[VEC16_A_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VEC16_A]] to ptr
102 // CHECK-NEXT: [[VEC16_B_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[VEC16_B]] to ptr
103 // CHECK-NEXT: [[TMP0:%.*]] = load <2 x bfloat>, ptr [[VEC2_B_ASCAST]], align 4
104 // CHECK-NEXT: store <2 x bfloat> [[TMP0]], ptr [[VEC2_A_ASCAST]], align 4
105 // CHECK-NEXT: [[TMP1:%.*]] = load <4 x bfloat>, ptr [[VEC4_B_ASCAST]], align 8
106 // CHECK-NEXT: store <4 x bfloat> [[TMP1]], ptr [[VEC4_A_ASCAST]], align 8
107 // CHECK-NEXT: [[TMP2:%.*]] = load <8 x bfloat>, ptr [[VEC8_B_ASCAST]], align 16
108 // CHECK-NEXT: store <8 x bfloat> [[TMP2]], ptr [[VEC8_A_ASCAST]], align 16
109 // CHECK-NEXT: [[TMP3:%.*]] = load <16 x bfloat>, ptr [[VEC16_B_ASCAST]], align 32
110 // CHECK-NEXT: store <16 x bfloat> [[TMP3]], ptr [[VEC16_A_ASCAST]], align 32
111 // CHECK-NEXT: ret void
113 __device__ void test_vec_assign() {
114 typedef __attribute__((ext_vector_type(2))) __bf16 bf16_x2;
115 bf16_x2 vec2_a, vec2_b;
118 typedef __attribute__((ext_vector_type(4))) __bf16 bf16_x4;
119 bf16_x4 vec4_a, vec4_b;
122 typedef __attribute__((ext_vector_type(8))) __bf16 bf16_x8;
123 bf16_x8 vec8_a, vec8_b;
126 typedef __attribute__((ext_vector_type(16))) __bf16 bf16_x16;
127 bf16_x16 vec16_a, vec16_b;