Skip to content

Commit 9fcf8c7

Browse files
simollsimoll
committed
Factored common code into function in check-shader-stages.hlsl
1 parent e15eacb commit 9fcf8c7

1 file changed

Lines changed: 20 additions & 187 deletions

File tree

‎tools/clang/test/CodeGenDXIL/hlsl/intrinsics/linalg_builtins/check-shader-stages.hlsl‎

Lines changed: 20 additions & 187 deletions
Original file line numberDiff line numberDiff line change
@@ -4,15 +4,7 @@ ByteAddressBuffer matrix_buffer;
44
ByteAddressBuffer bias_buffer;
55
RWByteAddressBuffer rw_matrix_buffer;
66

7-
// CHECK: define void @ps_main()
8-
// CHECK: call <4 x float> @dx.op.matVecMul
9-
// CHECK: call <4 x float> @dx.op.matVecMulAdd
10-
// CHECK: call void @dx.op.outerProductAccumulate
11-
// CHECK: call void @dx.op.vectorAccumulate
12-
13-
[Shader("pixel")]
14-
void ps_main()
15-
{
7+
void UseCoopVec() {
168
vector<float, 4> output_vector;
179
static const uint is_output_unsigned = 0;
1810

@@ -59,6 +51,18 @@ void ps_main()
5951
va_matrix_offset);
6052
}
6153

54+
// CHECK: define void @ps_main()
55+
// CHECK: call <4 x float> @dx.op.matVecMul
56+
// CHECK: call <4 x float> @dx.op.matVecMulAdd
57+
// CHECK: call void @dx.op.outerProductAccumulate
58+
// CHECK: call void @dx.op.vectorAccumulate
59+
60+
[Shader("pixel")]
61+
void ps_main()
62+
{
63+
UseCoopVec();
64+
}
65+
6266
// CHECK: define void @cs_main()
6367
// CHECK: call <4 x float> @dx.op.matVecMul
6468
// CHECK: call <4 x float> @dx.op.matVecMulAdd
@@ -69,50 +73,7 @@ void ps_main()
6973
[NumThreads(1,1,1)]
7074
void cs_main()
7175
{
72-
vector<float, 4> output_vector;
73-
static const uint is_output_unsigned = 0;
74-
75-
vector<float, 4> input_vector;
76-
const uint is_input_unsigned = 0;
77-
const uint input_interpretation = 9; /*F32*/
78-
79-
const uint matrix_offset = 0;
80-
const uint matrix_interpretation = 9; /*F32*/
81-
const uint matrix_dimM = 4;
82-
const uint matrix_dimK = 4;
83-
const uint matrix_layout = 0; /*RowMajor*/
84-
const bool matrix_is_transposed = false;
85-
const uint matrix_stride = 64;
86-
87-
__builtin_MatVecMul(output_vector, is_output_unsigned, input_vector,
88-
is_input_unsigned, input_interpretation, matrix_buffer, matrix_offset,
89-
matrix_interpretation, matrix_dimM, matrix_dimK, matrix_layout,
90-
matrix_is_transposed, matrix_stride);
91-
92-
const uint bias_offset = 0;
93-
const uint bias_interpretation = 9; /*F32*/
94-
95-
__builtin_MatVecMulAdd(output_vector, is_output_unsigned, input_vector,
96-
is_input_unsigned, input_interpretation, matrix_buffer, matrix_offset,
97-
matrix_interpretation, matrix_dimM, matrix_dimK, matrix_layout,
98-
matrix_is_transposed, matrix_stride, bias_buffer, bias_offset,
99-
bias_interpretation);
100-
101-
vector<uint, 8> input_vector1;
102-
vector<uint, 8> input_vector2;
103-
const uint opa_matrix_offset = 0;
104-
const uint opa_matrix_interpretation = 5; /*U32*/
105-
const uint opa_matrix_layout = 3; /*OuterProductOptimal*/
106-
const uint opa_matrix_stride = 64;
107-
108-
__builtin_OuterProductAccumulate(input_vector1, input_vector2,
109-
rw_matrix_buffer, opa_matrix_offset, opa_matrix_interpretation,
110-
opa_matrix_layout, opa_matrix_stride);
111-
112-
const uint va_matrix_offset = 0;
113-
114-
__builtin_VectorAccumulate(input_vector1, rw_matrix_buffer,
115-
va_matrix_offset);
76+
UseCoopVec();
11677
}
11778

11879
// CHECK: define void @vs_main()
@@ -123,51 +84,9 @@ void cs_main()
12384

12485
[Shader("vertex")]
12586
void vs_main()
126-
{
127-
vector<float, 4> output_vector;
128-
static const uint is_output_unsigned = 0;
129-
130-
vector<float, 4> input_vector;
131-
const uint is_input_unsigned = 0;
132-
const uint input_interpretation = 9; /*F32*/
133-
134-
const uint matrix_offset = 0;
135-
const uint matrix_interpretation = 9; /*F32*/
136-
const uint matrix_dimM = 4;
137-
const uint matrix_dimK = 4;
138-
const uint matrix_layout = 0; /*RowMajor*/
139-
const bool matrix_is_transposed = false;
140-
const uint matrix_stride = 64;
141-
142-
__builtin_MatVecMul(output_vector, is_output_unsigned, input_vector,
143-
is_input_unsigned, input_interpretation, matrix_buffer, matrix_offset,
144-
matrix_interpretation, matrix_dimM, matrix_dimK, matrix_layout,
145-
matrix_is_transposed, matrix_stride);
146-
147-
const uint bias_offset = 0;
148-
const uint bias_interpretation = 9; /*F32*/
149-
150-
__builtin_MatVecMulAdd(output_vector, is_output_unsigned, input_vector,
151-
is_input_unsigned, input_interpretation, matrix_buffer, matrix_offset,
152-
matrix_interpretation, matrix_dimM, matrix_dimK, matrix_layout,
153-
matrix_is_transposed, matrix_stride, bias_buffer, bias_offset,
154-
bias_interpretation);
155-
156-
vector<uint, 8> input_vector1;
157-
vector<uint, 8> input_vector2;
158-
const uint opa_matrix_offset = 0;
159-
const uint opa_matrix_interpretation = 5; /*U32*/
160-
const uint opa_matrix_layout = 3; /*OuterProductOptimal*/
161-
const uint opa_matrix_stride = 64;
162-
163-
__builtin_OuterProductAccumulate(input_vector1, input_vector2,
164-
rw_matrix_buffer, opa_matrix_offset, opa_matrix_interpretation,
165-
opa_matrix_layout, opa_matrix_stride);
166-
167-
const uint va_matrix_offset = 0;
168-
169-
__builtin_VectorAccumulate(input_vector1, rw_matrix_buffer,
170-
va_matrix_offset); }
87+
{
88+
UseCoopVec();
89+
}
17190

17291
struct MyRecord{
17392
uint a;
@@ -183,50 +102,7 @@ struct MyRecord{
183102
[NodeLaunch("thread")]
184103
void ns_main(ThreadNodeInputRecord<MyRecord> input)
185104
{
186-
vector<float, 4> output_vector;
187-
static const uint is_output_unsigned = 0;
188-
189-
vector<float, 4> input_vector;
190-
const uint is_input_unsigned = 0;
191-
const uint input_interpretation = 9; /*F32*/
192-
193-
const uint matrix_offset = 0;
194-
const uint matrix_interpretation = 9; /*F32*/
195-
const uint matrix_dimM = 4;
196-
const uint matrix_dimK = 4;
197-
const uint matrix_layout = 0; /*RowMajor*/
198-
const bool matrix_is_transposed = false;
199-
const uint matrix_stride = 64;
200-
201-
__builtin_MatVecMul(output_vector, is_output_unsigned, input_vector,
202-
is_input_unsigned, input_interpretation, matrix_buffer, matrix_offset,
203-
matrix_interpretation, matrix_dimM, matrix_dimK, matrix_layout,
204-
matrix_is_transposed, matrix_stride);
205-
206-
const uint bias_offset = 0;
207-
const uint bias_interpretation = 9; /*F32*/
208-
209-
__builtin_MatVecMulAdd(output_vector, is_output_unsigned, input_vector,
210-
is_input_unsigned, input_interpretation, matrix_buffer, matrix_offset,
211-
matrix_interpretation, matrix_dimM, matrix_dimK, matrix_layout,
212-
matrix_is_transposed, matrix_stride, bias_buffer, bias_offset,
213-
bias_interpretation);
214-
215-
vector<uint, 8> input_vector1;
216-
vector<uint, 8> input_vector2;
217-
const uint opa_matrix_offset = 0;
218-
const uint opa_matrix_interpretation = 5; /*U32*/
219-
const uint opa_matrix_layout = 3; /*OuterProductOptimal*/
220-
const uint opa_matrix_stride = 64;
221-
222-
__builtin_OuterProductAccumulate(input_vector1, input_vector2,
223-
rw_matrix_buffer, opa_matrix_offset, opa_matrix_interpretation,
224-
opa_matrix_layout, opa_matrix_stride);
225-
226-
const uint va_matrix_offset = 0;
227-
228-
__builtin_VectorAccumulate(input_vector1, rw_matrix_buffer,
229-
va_matrix_offset);
105+
UseCoopVec();
230106
}
231107

232108
// Vertex shader output structure
@@ -250,49 +126,6 @@ struct GS_OUT {
250126
[maxvertexcount(3)]
251127
void gs_main(point VS_OUT input[1],
252128
inout TriangleStream<GS_OUT> OutputStream)
253-
{
254-
vector<float, 4> output_vector;
255-
static const uint is_output_unsigned = 0;
256-
257-
vector<float, 4> input_vector;
258-
const uint is_input_unsigned = 0;
259-
const uint input_interpretation = 9; /*F32*/
260-
261-
const uint matrix_offset = 0;
262-
const uint matrix_interpretation = 9; /*F32*/
263-
const uint matrix_dimM = 4;
264-
const uint matrix_dimK = 4;
265-
const uint matrix_layout = 0; /*RowMajor*/
266-
const bool matrix_is_transposed = false;
267-
const uint matrix_stride = 64;
268-
269-
__builtin_MatVecMul(output_vector, is_output_unsigned, input_vector,
270-
is_input_unsigned, input_interpretation, matrix_buffer, matrix_offset,
271-
matrix_interpretation, matrix_dimM, matrix_dimK, matrix_layout,
272-
matrix_is_transposed, matrix_stride);
273-
274-
const uint bias_offset = 0;
275-
const uint bias_interpretation = 9; /*F32*/
276-
277-
__builtin_MatVecMulAdd(output_vector, is_output_unsigned, input_vector,
278-
is_input_unsigned, input_interpretation, matrix_buffer, matrix_offset,
279-
matrix_interpretation, matrix_dimM, matrix_dimK, matrix_layout,
280-
matrix_is_transposed, matrix_stride, bias_buffer, bias_offset,
281-
bias_interpretation);
282-
283-
vector<uint, 8> input_vector1;
284-
vector<uint, 8> input_vector2;
285-
const uint opa_matrix_offset = 0;
286-
const uint opa_matrix_interpretation = 5; /*U32*/
287-
const uint opa_matrix_layout = 3; /*OuterProductOptimal*/
288-
const uint opa_matrix_stride = 64;
289-
290-
__builtin_OuterProductAccumulate(input_vector1, input_vector2,
291-
rw_matrix_buffer, opa_matrix_offset, opa_matrix_interpretation,
292-
opa_matrix_layout, opa_matrix_stride);
293-
294-
const uint va_matrix_offset = 0;
295-
296-
__builtin_VectorAccumulate(input_vector1, rw_matrix_buffer,
297-
va_matrix_offset);
129+
{
130+
UseCoopVec();
298131
}

0 commit comments

Comments
 (0)