Skip to content

Commit bf06295

Browse files
againullagainull
authored
[SYCL] Support both opaque and no-opaque modes for matrix tests (#7227)
This PR is intended to fix matrix test failures caused by upcoming pulldown: #7186
1 parent 19e9941 commit bf06295

8 files changed

+214
-8
lines changed

‎sycl/test/check_device_code/matrix/matrix-nvptx-bf16-test.cpp

Lines changed: 32 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,7 @@
11
// REQUIRES: cuda
22

3-
// RUN: %clangxx -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX_VERSION=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
3+
// RUN: %clangxx -Xclang -no-opaque-pointers -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX_VERSION=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s
4+
// RUN: %clangxx -Xclang -opaque-pointers -fsycl-device-only -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX_VERSION=3 -S -Xclang -emit-llvm %s -o -| FileCheck %s --check-prefixes=CHECK-OPAQUE
45

56
#include <sycl/sycl.hpp>
67

@@ -42,14 +43,19 @@ int main() {
4243
sub_b;
4344

4445
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m16n16k16.load.c.row.stride.f32.p1f32(float addrspace(1)* %_arg_accC, i32 16)
46+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m16n16k16.load.c.row.stride.f32.p1(ptr addrspace(1) %_arg_accC, i32 16)
4547
joint_matrix_load(sg, sub_c, accC.get_pointer(), stride);
4648
// CHECK: tail call { i32, i32, i32, i32 } @llvm.nvvm.wmma.m16n16k16.load.a.row.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
49+
// CHECK-OPAQUE: tail call { i32, i32, i32, i32 } @llvm.nvvm.wmma.m16n16k16.load.a.row.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
4750
joint_matrix_load(sg, sub_a, accA.get_pointer(), stride);
4851
// CHECK: tail call { i32, i32, i32, i32 } @llvm.nvvm.wmma.m16n16k16.load.b.row.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
52+
// CHECK-OPAQUE: tail call { i32, i32, i32, i32 } @llvm.nvvm.wmma.m16n16k16.load.b.row.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
4953
joint_matrix_load(sg, sub_b, accB.get_pointer(), stride);
5054
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m16n16k16.mma.row.row.bf16(i32 %11, i32 %12, i32 %13, i32 %14, i32 %17, i32 %18, i32 %19, i32 %20, float %1, float %2, float %3, float %4, float %5, float %6, float %7, float %8)
55+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m16n16k16.mma.row.row.bf16(i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}})
5156
sub_c = joint_matrix_mad(sg, sub_a, sub_b, sub_c);
5257
// CHECK: tail call void @llvm.nvvm.wmma.m16n16k16.store.d.row.stride.f32.p1f32(float addrspace(1)* %_arg_accD, float %22, float %23, float %24, float %25, float %26, float %27, float %28, float %29, i32 16)
58+
// CHECK-OPAQUE: tail call void @llvm.nvvm.wmma.m16n16k16.store.d.row.stride.f32.p1(ptr addrspace(1) %_arg_accD, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, i32 16)
5359
joint_matrix_store(sg, sub_c, accD.get_pointer(), stride);
5460
});
5561

@@ -71,14 +77,19 @@ int main() {
7177
sub_b;
7278

7379
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m16n16k16.load.c.col.stride.f32.p1f32(float addrspace(1)* %_arg_accC, i32 16)
80+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m16n16k16.load.c.col.stride.f32.p1(ptr addrspace(1) %_arg_accC, i32 16)
7481
joint_matrix_load(sg, sub_c, accC.get_pointer(), stride);
7582
// CHECK: tail call { i32, i32, i32, i32 } @llvm.nvvm.wmma.m16n16k16.load.a.col.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
83+
// CHECK-OPAQUE: tail call { i32, i32, i32, i32 } @llvm.nvvm.wmma.m16n16k16.load.a.col.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
7684
joint_matrix_load(sg, sub_a, accA.get_pointer(), stride);
7785
// CHECK: tail call { i32, i32, i32, i32 } @llvm.nvvm.wmma.m16n16k16.load.b.col.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
86+
// CHECK-OPAQUE: tail call { i32, i32, i32, i32 } @llvm.nvvm.wmma.m16n16k16.load.b.col.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
7887
joint_matrix_load(sg, sub_b, accB.get_pointer(), stride);
7988
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m16n16k16.mma.col.col.bf16(i32 %11, i32 %12, i32 %13, i32 %14, i32 %17, i32 %18, i32 %19, i32 %20, float %1, float %2, float %3, float %4, float %5, float %6, float %7, float %8)
89+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m16n16k16.mma.col.col.bf16(i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}})
8090
sub_c = joint_matrix_mad(sg, sub_a, sub_b, sub_c);
8191
// CHECK: tail call void @llvm.nvvm.wmma.m16n16k16.store.d.col.stride.f32.p1f32(float addrspace(1)* %_arg_accD, float %22, float %23, float %24, float %25, float %26, float %27, float %28, float %29, i32 16)
92+
// CHECK-OPAQUE: tail call void @llvm.nvvm.wmma.m16n16k16.store.d.col.stride.f32.p1(ptr addrspace(1) %_arg_accD, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, i32 16)
8293
joint_matrix_store(sg, sub_c, accD.get_pointer(), stride);
8394
});
8495

@@ -99,14 +110,19 @@ int main() {
99110
sub_b;
100111

101112
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m32n8k16.load.c.row.stride.f32.p1f32(float addrspace(1)* %_arg_accC, i32 16)
113+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m32n8k16.load.c.row.stride.f32.p1(ptr addrspace(1) %_arg_accC, i32 16)
102114
joint_matrix_load(sg, sub_c, accC.get_pointer(), stride);
103115
// CHECK: tail call { i32, i32, i32, i32, i32, i32, i32, i32 } @llvm.nvvm.wmma.m32n8k16.load.a.row.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
116+
// CHECK-OPAQUE: tail call { i32, i32, i32, i32, i32, i32, i32, i32 } @llvm.nvvm.wmma.m32n8k16.load.a.row.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
104117
joint_matrix_load(sg, sub_a, accA.get_pointer(), stride);
105118
// CHECK: tail call { i32, i32 } @llvm.nvvm.wmma.m32n8k16.load.b.row.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
119+
// CHECK-OPAQUE: tail call { i32, i32 } @llvm.nvvm.wmma.m32n8k16.load.b.row.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
106120
joint_matrix_load(sg, sub_b, accB.get_pointer(), stride);
107121
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m32n8k16.mma.row.row.bf16(i32 %11, i32 %12, i32 %13, i32 %14, i32 %15, i32 %16, i32 %17, i32 %18, i32 %21, i32 %22, float %1, float %2, float %3, float %4, float %5, float %6, float %7, float %8)
122+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m32n8k16.mma.row.row.bf16(i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}})
108123
sub_c = joint_matrix_mad(sg, sub_a, sub_b, sub_c);
109124
// CHECK: tail call void @llvm.nvvm.wmma.m32n8k16.store.d.row.stride.f32.p1f32(float addrspace(1)* %_arg_accD, float %24, float %25, float %26, float %27, float %28, float %29, float %30, float %31, i32 16)
125+
// CHECK-OPAQUE: tail call void @llvm.nvvm.wmma.m32n8k16.store.d.row.stride.f32.p1(ptr addrspace(1) %_arg_accD, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, i32 16)
110126
joint_matrix_store(sg, sub_c, accD.get_pointer(), stride);
111127
});
112128

@@ -127,14 +143,19 @@ int main() {
127143
sub_b;
128144

129145
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m32n8k16.load.c.col.stride.f32.p1f32(float addrspace(1)* %_arg_accC, i32 16)
146+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m32n8k16.load.c.col.stride.f32.p1(ptr addrspace(1) %_arg_accC, i32 16)
130147
joint_matrix_load(sg, sub_c, accC.get_pointer(), stride);
131148
// CHECK: tail call { i32, i32, i32, i32, i32, i32, i32, i32 } @llvm.nvvm.wmma.m32n8k16.load.a.col.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
149+
// CHECK-OPAQUE: tail call { i32, i32, i32, i32, i32, i32, i32, i32 } @llvm.nvvm.wmma.m32n8k16.load.a.col.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
132150
joint_matrix_load(sg, sub_a, accA.get_pointer(), stride);
133151
// CHECK: tail call { i32, i32 } @llvm.nvvm.wmma.m32n8k16.load.b.col.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
152+
// CHECK-OPAQUE: tail call { i32, i32 } @llvm.nvvm.wmma.m32n8k16.load.b.col.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
134153
joint_matrix_load(sg, sub_b, accB.get_pointer(), stride);
135154
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m32n8k16.mma.col.col.bf16(i32 %11, i32 %12, i32 %13, i32 %14, i32 %15, i32 %16, i32 %17, i32 %18, i32 %21, i32 %22, float %1, float %2, float %3, float %4, float %5, float %6, float %7, float %8)
155+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m32n8k16.mma.col.col.bf16(i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}})
136156
sub_c = joint_matrix_mad(sg, sub_a, sub_b, sub_c);
137157
// CHECK: tail call void @llvm.nvvm.wmma.m32n8k16.store.d.col.stride.f32.p1f32(float addrspace(1)* %_arg_accD, float %24, float %25, float %26, float %27, float %28, float %29, float %30, float %31, i32 16)
158+
// CHECK-OPAQUE: tail call void @llvm.nvvm.wmma.m32n8k16.store.d.col.stride.f32.p1(ptr addrspace(1) %_arg_accD, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, i32 16)
138159
joint_matrix_store(sg, sub_c, accD.get_pointer(), stride);
139160
});
140161

@@ -155,14 +176,19 @@ int main() {
155176
sub_b;
156177

157178
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m8n32k16.load.c.row.stride.f32.p1f32(float addrspace(1)* %_arg_accC, i32 16)
179+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m8n32k16.load.c.row.stride.f32.p1(ptr addrspace(1) %_arg_accC, i32 16)
158180
joint_matrix_load(sg, sub_c, accC.get_pointer(), stride);
159181
// CHECK: tail call { i32, i32 } @llvm.nvvm.wmma.m8n32k16.load.a.row.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
182+
// CHECK-OPAQUE: tail call { i32, i32 } @llvm.nvvm.wmma.m8n32k16.load.a.row.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
160183
joint_matrix_load(sg, sub_a, accA.get_pointer(), stride);
161184
// CHECK: tail call { i32, i32, i32, i32, i32, i32, i32, i32 } @llvm.nvvm.wmma.m8n32k16.load.b.row.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
185+
// CHECK-OPAQUE: tail call { i32, i32, i32, i32, i32, i32, i32, i32 } @llvm.nvvm.wmma.m8n32k16.load.b.row.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
162186
joint_matrix_load(sg, sub_b, accB.get_pointer(), stride);
163187
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m8n32k16.mma.row.row.bf16(i32 %11, i32 %12, i32 %15, i32 %16, i32 %17, i32 %18, i32 %19, i32 %20, i32 %21, i32 %22, float %1, float %2, float %3, float %4, float %5, float %6, float %7, float %8)
188+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m8n32k16.mma.row.row.bf16(i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}})
164189
sub_c = joint_matrix_mad(sg, sub_a, sub_b, sub_c);
165190
// CHECK: tail call void @llvm.nvvm.wmma.m8n32k16.store.d.row.stride.f32.p1f32(float addrspace(1)* %_arg_accD, float %24, float %25, float %26, float %27, float %28, float %29, float %30, float %31, i32 16)
191+
// CHECK-OPAQUE: tail call void @llvm.nvvm.wmma.m8n32k16.store.d.row.stride.f32.p1(ptr addrspace(1) %_arg_accD, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, i32 16)
166192
joint_matrix_store(sg, sub_c, accD.get_pointer(), stride);
167193
});
168194

@@ -183,14 +209,19 @@ int main() {
183209
sub_b;
184210

185211
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m8n32k16.load.c.col.stride.f32.p1f32(float addrspace(1)* %_arg_accC, i32 16)
212+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m8n32k16.load.c.col.stride.f32.p1(ptr addrspace(1) %_arg_accC, i32 16)
186213
joint_matrix_load(sg, sub_c, accC.get_pointer(), stride);
187214
// CHECK: tail call { i32, i32 } @llvm.nvvm.wmma.m8n32k16.load.a.col.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
215+
// CHECK-OPAQUE: tail call { i32, i32 } @llvm.nvvm.wmma.m8n32k16.load.a.col.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
188216
joint_matrix_load(sg, sub_a, accA.get_pointer(), stride);
189217
// CHECK: tail call { i32, i32, i32, i32, i32, i32, i32, i32 } @llvm.nvvm.wmma.m8n32k16.load.b.col.stride.bf16.p0i32(i32* %call.ascast.i.i{{.*}}.i, i32 16)
218+
// CHECK-OPAQUE: tail call { i32, i32, i32, i32, i32, i32, i32, i32 } @llvm.nvvm.wmma.m8n32k16.load.b.col.stride.bf16.p0(ptr %call.ascast.i.i{{.*}}.i, i32 16)
190219
joint_matrix_load(sg, sub_b, accB.get_pointer(), stride);
191220
// CHECK: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m8n32k16.mma.col.col.bf16(i32 %11, i32 %12, i32 %15, i32 %16, i32 %17, i32 %18, i32 %19, i32 %20, i32 %21, i32 %22, float %1, float %2, float %3, float %4, float %5, float %6, float %7, float %8)
221+
// CHECK-OPAQUE: tail call { float, float, float, float, float, float, float, float } @llvm.nvvm.wmma.m8n32k16.mma.col.col.bf16(i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, i32 {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}})
192222
sub_c = joint_matrix_mad(sg, sub_a, sub_b, sub_c);
193223
// CHECK: tail call void @llvm.nvvm.wmma.m8n32k16.store.d.col.stride.f32.p1f32(float addrspace(1)* %_arg_accD, float %24, float %25, float %26, float %27, float %28, float %29, float %30, float %31, i32 16)
224+
// CHECK-OPAQUE: tail call void @llvm.nvvm.wmma.m8n32k16.store.d.col.stride.f32.p1(ptr addrspace(1) %_arg_accD, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, float {{.*}}, i32 16)
194225
joint_matrix_store(sg, sub_c, accD.get_pointer(), stride);
195226
});
196227
});

0 commit comments

Comments
 (0)