Skip to content

Commit 1d94ac0

Browse files
tgymnichtgymnich
committed
remove index from asm
1 parent 44a2b7e commit 1d94ac0

File tree

2 files changed

+19
-19
lines changed

2 files changed

+19
-19
lines changed

‎mlir/include/mlir/Dialect/AMDGPU/IR/AMDGPU.td‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -194,7 +194,7 @@ def AMDGPU_PackedScaledTruncOp
194194
packed vectors of float values.
195195
}];
196196
let assemblyFormat = [{
197-
attr-dict $source `into` ($existing^):(`undef`)? `[` `index` $index `]`
197+
attr-dict $source `into` ($existing^):(`undef`)? `[` $index `]`
198198
`,` $scale
199199
`:` type($source) `to` type($res) (`into` type($existing)^)?
200200
}];

‎mlir/test/Conversion/AMDGPUToROCDL/packed-trunc.mlir‎

Lines changed: 18 additions & 18 deletions
Original file line numberDiff line numberDiff line change
@@ -11,7 +11,7 @@
1111
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E4M3FN>
1212
// CHECK: return [[CAST]] : vector<4xf8E4M3FN>
1313
func.func @packed_scaled_trunc_f8e4m3_f32(%v: vector<2xf32>, %scale: f32) -> vector<4xf8E4M3FN> {
14-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xf32> to vector<4xf8E4M3FN>
14+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xf32> to vector<4xf8E4M3FN>
1515
func.return %ret : vector<4xf8E4M3FN>
1616
}
1717

@@ -27,7 +27,7 @@ func.func @packed_scaled_trunc_f8e4m3_f32(%v: vector<2xf32>, %scale: f32) -> vec
2727
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E4M3FN>
2828
// CHECK: return [[CAST]] : vector<4xf8E4M3FN>
2929
func.func @packed_scaled_trunc_into_f8e4m3_f32(%v: vector<2xf32>, %existing: vector<4xf8E4M3FN>, %scale: f32) -> vector<4xf8E4M3FN> {
30-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xf32> to vector<4xf8E4M3FN> into vector<4xf8E4M3FN>
30+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xf32> to vector<4xf8E4M3FN> into vector<4xf8E4M3FN>
3131
func.return %ret : vector<4xf8E4M3FN>
3232
}
3333

@@ -38,7 +38,7 @@ func.func @packed_scaled_trunc_into_f8e4m3_f32(%v: vector<2xf32>, %existing: vec
3838
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E4M3FN>
3939
// CHECK: return [[CAST]] : vector<4xf8E4M3FN>
4040
func.func @packed_scaled_trunc_f8e4m3_f16(%v: vector<2xf16>, %scale: f32) -> vector<4xf8E4M3FN> {
41-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xf16> to vector<4xf8E4M3FN>
41+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xf16> to vector<4xf8E4M3FN>
4242
func.return %ret : vector<4xf8E4M3FN>
4343
}
4444

@@ -50,7 +50,7 @@ func.func @packed_scaled_trunc_f8e4m3_f16(%v: vector<2xf16>, %scale: f32) -> vec
5050
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E4M3FN>
5151
// CHECK: return [[CAST]] : vector<4xf8E4M3FN>
5252
func.func @packed_scaled_trunc_into_f8e4m3_f16(%v: vector<2xf16>, %existing: vector<4xf8E4M3FN>, %scale: f32) -> vector<4xf8E4M3FN> {
53-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xf16> to vector<4xf8E4M3FN> into vector<4xf8E4M3FN>
53+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xf16> to vector<4xf8E4M3FN> into vector<4xf8E4M3FN>
5454
func.return %ret : vector<4xf8E4M3FN>
5555
}
5656

@@ -61,7 +61,7 @@ func.func @packed_scaled_trunc_into_f8e4m3_f16(%v: vector<2xf16>, %existing: vec
6161
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E4M3FN>
6262
// CHECK: return [[CAST]] : vector<4xf8E4M3FN>
6363
func.func @packed_scaled_trunc_f8e4m3_bf16(%v: vector<2xbf16>, %scale: f32) -> vector<4xf8E4M3FN> {
64-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xbf16> to vector<4xf8E4M3FN>
64+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xbf16> to vector<4xf8E4M3FN>
6565
func.return %ret : vector<4xf8E4M3FN>
6666
}
6767

@@ -73,7 +73,7 @@ func.func @packed_scaled_trunc_f8e4m3_bf16(%v: vector<2xbf16>, %scale: f32) -> v
7373
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E4M3FN>
7474
// CHECK: return [[CAST]] : vector<4xf8E4M3FN>
7575
func.func @packed_scaled_trunc_into_f8e4m3_bf16(%v: vector<2xbf16>, %existing: vector<4xf8E4M3FN>, %scale: f32) -> vector<4xf8E4M3FN> {
76-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xbf16> to vector<4xf8E4M3FN> into vector<4xf8E4M3FN>
76+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xbf16> to vector<4xf8E4M3FN> into vector<4xf8E4M3FN>
7777
func.return %ret : vector<4xf8E4M3FN>
7878
}
7979

@@ -88,7 +88,7 @@ func.func @packed_scaled_trunc_into_f8e4m3_bf16(%v: vector<2xbf16>, %existing: v
8888
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E5M2>
8989
// CHECK: return [[CAST]] : vector<4xf8E5M2>
9090
func.func @packed_scaled_trunc_f8e5m2_f32(%v: vector<2xf32>, %scale: f32) -> vector<4xf8E5M2> {
91-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xf32> to vector<4xf8E5M2>
91+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xf32> to vector<4xf8E5M2>
9292
func.return %ret : vector<4xf8E5M2>
9393
}
9494

@@ -104,7 +104,7 @@ func.func @packed_scaled_trunc_f8e5m2_f32(%v: vector<2xf32>, %scale: f32) -> vec
104104
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E5M2>
105105
// CHECK: return [[CAST]] : vector<4xf8E5M2>
106106
func.func @packed_scaled_trunc_into_f8e5m2_f32(%v: vector<2xf32>, %existing: vector<4xf8E5M2>, %scale: f32) -> vector<4xf8E5M2> {
107-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xf32> to vector<4xf8E5M2> into vector<4xf8E5M2>
107+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xf32> to vector<4xf8E5M2> into vector<4xf8E5M2>
108108
func.return %ret : vector<4xf8E5M2>
109109
}
110110

@@ -115,7 +115,7 @@ func.func @packed_scaled_trunc_into_f8e5m2_f32(%v: vector<2xf32>, %existing: vec
115115
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E5M2>
116116
// CHECK: return [[CAST]] : vector<4xf8E5M2>
117117
func.func @packed_scaled_trunc_f8e5m2_f16(%v: vector<2xf16>, %scale: f32) -> vector<4xf8E5M2> {
118-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xf16> to vector<4xf8E5M2>
118+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xf16> to vector<4xf8E5M2>
119119
func.return %ret : vector<4xf8E5M2>
120120
}
121121

@@ -127,7 +127,7 @@ func.func @packed_scaled_trunc_f8e5m2_f16(%v: vector<2xf16>, %scale: f32) -> vec
127127
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E5M2>
128128
// CHECK: return [[CAST]] : vector<4xf8E5M2>
129129
func.func @packed_scaled_trunc_into_f8e5m2_f16(%v: vector<2xf16>, %existing: vector<4xf8E5M2>, %scale: f32) -> vector<4xf8E5M2> {
130-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xf16> to vector<4xf8E5M2> into vector<4xf8E5M2>
130+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xf16> to vector<4xf8E5M2> into vector<4xf8E5M2>
131131
func.return %ret : vector<4xf8E5M2>
132132
}
133133

@@ -138,7 +138,7 @@ func.func @packed_scaled_trunc_into_f8e5m2_f16(%v: vector<2xf16>, %existing: vec
138138
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E5M2>
139139
// CHECK: return [[CAST]] : vector<4xf8E5M2>
140140
func.func @packed_scaled_trunc_f8e5m2_bf16(%v: vector<2xbf16>, %scale: f32) -> vector<4xf8E5M2> {
141-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xbf16> to vector<4xf8E5M2>
141+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xbf16> to vector<4xf8E5M2>
142142
func.return %ret : vector<4xf8E5M2>
143143
}
144144

@@ -150,7 +150,7 @@ func.func @packed_scaled_trunc_f8e5m2_bf16(%v: vector<2xbf16>, %scale: f32) -> v
150150
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<4xi8> to vector<4xf8E5M2>
151151
// CHECK: return [[CAST]] : vector<4xf8E5M2>
152152
func.func @packed_scaled_trunc_into_f8e5m2_bf16(%v: vector<2xbf16>, %existing: vector<4xf8E5M2>, %scale: f32) -> vector<4xf8E5M2> {
153-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xbf16> to vector<4xf8E5M2> into vector<4xf8E5M2>
153+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xbf16> to vector<4xf8E5M2> into vector<4xf8E5M2>
154154
func.return %ret : vector<4xf8E5M2>
155155
}
156156

@@ -165,7 +165,7 @@ func.func @packed_scaled_trunc_into_f8e5m2_bf16(%v: vector<2xbf16>, %existing: v
165165
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<8xi4> to vector<8xf4E2M1FN>
166166
// CHECK: return [[CAST]] : vector<8xf4E2M1FN>
167167
func.func @packed_scaled_trunc_f4e2m1_f32(%v: vector<2xf32>, %scale: f32) -> vector<8xf4E2M1FN> {
168-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xf32> to vector<8xf4E2M1FN>
168+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xf32> to vector<8xf4E2M1FN>
169169
func.return %ret : vector<8xf4E2M1FN>
170170
}
171171

@@ -181,7 +181,7 @@ func.func @packed_scaled_trunc_f4e2m1_f32(%v: vector<2xf32>, %scale: f32) -> vec
181181
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<8xi4> to vector<8xf4E2M1FN>
182182
// CHECK: return [[CAST]] : vector<8xf4E2M1FN>
183183
func.func @packed_scaled_trunc_into_f4e2m1_f32(%v: vector<2xf32>, %existing: vector<8xf4E2M1FN>, %scale: f32) -> vector<8xf4E2M1FN> {
184-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xf32> to vector<8xf4E2M1FN> into vector<8xf4E2M1FN>
184+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xf32> to vector<8xf4E2M1FN> into vector<8xf4E2M1FN>
185185
func.return %ret : vector<8xf4E2M1FN>
186186
}
187187

@@ -192,7 +192,7 @@ func.func @packed_scaled_trunc_into_f4e2m1_f32(%v: vector<2xf32>, %existing: vec
192192
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<8xi4> to vector<8xf4E2M1FN>
193193
// CHECK: return [[CAST]] : vector<8xf4E2M1FN>
194194
func.func @packed_scaled_trunc_f4e2m1_f16(%v: vector<2xf16>, %scale: f32) -> vector<8xf4E2M1FN> {
195-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xf16> to vector<8xf4E2M1FN>
195+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xf16> to vector<8xf4E2M1FN>
196196
func.return %ret : vector<8xf4E2M1FN>
197197
}
198198

@@ -204,7 +204,7 @@ func.func @packed_scaled_trunc_f4e2m1_f16(%v: vector<2xf16>, %scale: f32) -> vec
204204
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<8xi4> to vector<8xf4E2M1FN>
205205
// CHECK: return [[CAST]] : vector<8xf4E2M1FN>
206206
func.func @packed_scaled_trunc_into_f4e2m1_f16(%v: vector<2xf16>, %existing: vector<8xf4E2M1FN>, %scale: f32) -> vector<8xf4E2M1FN> {
207-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xf16> to vector<8xf4E2M1FN> into vector<8xf4E2M1FN>
207+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xf16> to vector<8xf4E2M1FN> into vector<8xf4E2M1FN>
208208
func.return %ret : vector<8xf4E2M1FN>
209209
}
210210

@@ -215,7 +215,7 @@ func.func @packed_scaled_trunc_into_f4e2m1_f16(%v: vector<2xf16>, %existing: vec
215215
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<8xi4> to vector<8xf4E2M1FN>
216216
// CHECK: return [[CAST]] : vector<8xf4E2M1FN>
217217
func.func @packed_scaled_trunc_f4e2m1_bf16(%v: vector<2xbf16>, %scale: f32) -> vector<8xf4E2M1FN> {
218-
%ret = amdgpu.packed_scaled_trunc %v into undef[index 0], %scale : vector<2xbf16> to vector<8xf4E2M1FN>
218+
%ret = amdgpu.packed_scaled_trunc %v into undef[0], %scale : vector<2xbf16> to vector<8xf4E2M1FN>
219219
func.return %ret : vector<8xf4E2M1FN>
220220
}
221221

@@ -227,6 +227,6 @@ func.func @packed_scaled_trunc_f4e2m1_bf16(%v: vector<2xbf16>, %scale: f32) -> v
227227
// CHECK: [[CAST:%.+]] = builtin.unrealized_conversion_cast [[BITCAST]] : vector<8xi4> to vector<8xf4E2M1FN>
228228
// CHECK: return [[CAST]] : vector<8xf4E2M1FN>
229229
func.func @packed_scaled_trunc_into_f4e2m1_bf16(%v: vector<2xbf16>, %existing: vector<8xf4E2M1FN>, %scale: f32) -> vector<8xf4E2M1FN> {
230-
%ret = amdgpu.packed_scaled_trunc %v into %existing[index 0], %scale : vector<2xbf16> to vector<8xf4E2M1FN> into vector<8xf4E2M1FN>
230+
%ret = amdgpu.packed_scaled_trunc %v into %existing[0], %scale : vector<2xbf16> to vector<8xf4E2M1FN> into vector<8xf4E2M1FN>
231231
func.return %ret : vector<8xf4E2M1FN>
232232
}

0 commit comments

Comments
 (0)