1 // RUN: mlir-opt -split-input-file -verify-diagnostics %s
3 func.func @ldmatrix_address_space_f16_x4(%arg0: memref<128x128xf16, 2>) -> vector<4x1xf16> {
4 %c0 = arith.constant 0 : index
5 // expected-error @below {{expected nvgpu.ldmatrix srcMemref must have a memory space attribute of IntegerAttr(3) or gpu::AddressSpaceAttr(Workgroup)}}
6 %a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf16, 2> -> vector<4x1xf16>
7 return %a : vector<4x1xf16>
11 func.func @ldmatrix_num_elements_f16_x4(%arg0: memref<128x128xf16, 3>) -> vector<4x1xf16> {
12 %c0 = arith.constant 0 : index
13 // expected-error @+1 {{expected vector register shape[1] = 2}}
14 %a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf16, 3> -> vector<4x1xf16>
15 return %a : vector<4x1xf16>
19 func.func @ldmatrix_num_tiles_f16_x4(%arg0: memref<128x128xf16, 3>) -> vector<2x2xf16> {
20 %c0 = arith.constant 0 : index
21 // expected-error @+1 {{expected vector register shape[0] and numTiles to match}}
22 %a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf16, 3> -> vector<2x2xf16>
23 return %a : vector<2x2xf16>
27 func.func @ldmatrix_num_tiles_f32_x4(%arg0: memref<128x128xf32, 3>) -> vector<4x2xf32> {
28 %c0 = arith.constant 0 : index
29 // expected-error @+1 {{expected vector register shape[1] = 1}}
30 %a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf32, 3> -> vector<4x2xf32>
31 return %a : vector<4x2xf32>
35 func.func @ldmatrix_trans_f32_x4(%arg0: memref<128x128xf32, 3>) -> vector<4x1xf32> {
36 %c0 = arith.constant 0 : index
37 // expected-error @+1 {{nvgpu.ldmatrix transpose works only at 16b granularity}}
38 %a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = true, numTiles = 4 : i32} : memref<128x128xf32, 3> -> vector<4x1xf32>
39 return %a : vector<4x1xf32>
43 func.func @ldmatrix_trans_f32_x4(%arg0: memref<128x128xf32, 3>) -> vector<4x1xf32> {
44 %c0 = arith.constant 0 : index
45 // expected-error @+1 {{results must be 2 dimensional vector}}
46 %a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf32, 3> -> vector<4xf32>
47 return %a : vector<4xf32>
51 func.func @ldmatrix_type_x4(%arg0: memref<128x128xf32, 3>) -> vector<4x2xf16> {
52 %c0 = arith.constant 0 : index
53 // expected-error @+1 {{'nvgpu.ldmatrix' op failed to verify that srcMemref and res have same element type}}
54 %a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf32, 3> -> vector<4x2xf16>
55 return %a : vector<4x2xf16>
59 func.func @m16n8k16_fp16_vector_shape_a(%arg0: vector<4x4xf16>, %arg1: vector<2x2xf16>, %arg2: vector<2x2xf16>) -> vector<2x2xf16> {
60 // expected-error @+1 {{expected 256 warp-wide matrix A elements}}
61 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 16]} : (vector<4x4xf16>, vector<2x2xf16>, vector<2x2xf16>) -> vector<2x2xf16>
62 return %d : vector<2x2xf16>
66 func.func @m16n8k16_fp16_vector_shape_b(%arg0: vector<4x2xf16>, %arg1: vector<2x4xf16>, %arg2: vector<2x2xf16>) -> vector<2x2xf16> {
67 // expected-error @+1 {{expected 128 warp-wide matrix B elements}}
68 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 16]} : (vector<4x2xf16>, vector<2x4xf16>, vector<2x2xf16>) -> vector<2x2xf16>
69 return %d : vector<2x2xf16>
73 func.func @m16n8k16_fp16_vector_shape_c(%arg0: vector<4x2xf16>, %arg1: vector<2x2xf16>, %arg2: vector<2x4xf16>) -> vector<2x4xf16> {
74 // expected-error @+1 {{expected 128 warp-wide matrix C elements}}
75 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 16]} : (vector<4x2xf16>, vector<2x2xf16>, vector<2x4xf16>) -> vector<2x4xf16>
76 return %d : vector<2x4xf16>
80 func.func @m16n8k16_fp16_vector_shape_a_extended(%arg0: vector<2x4xf16>, %arg1: vector<2x2xf16>, %arg2: vector<2x2xf16>) -> vector<2x2xf16> {
81 // expected-error @+1 {{expected matrix A to be shaped (4 x 2)}}
82 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 16]} : (vector<2x4xf16>, vector<2x2xf16>, vector<2x2xf16>) -> vector<2x2xf16>
83 return %d : vector<2x2xf16>
87 func.func @m16n8k16_fp16_tf32Enabled(%arg0: vector<4x2xf16>, %arg1: vector<2x2xf16>, %arg2: vector<2x2xf16>) -> vector<2x2xf16> {
88 // expected-error @+1 {{expected tf32 tensor cores only for F32 operands}}
89 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 16], tf32Enabled} : (vector<4x2xf16>, vector<2x2xf16>, vector<2x2xf16>) -> vector<2x2xf16>
90 return %d : vector<2x2xf16>
94 func.func @m16n8k8_fp32_vector_shape_a(%arg0: vector<4x2xf32>, %arg1: vector<2x1xf32>, %arg2: vector<2x2xf32>) -> vector<2x2xf32> {
95 // expected-error @+1 {{expected 128 warp-wide matrix A elements}}
96 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 8]} : (vector<4x2xf32>, vector<2x1xf32>, vector<2x2xf32>) -> vector<2x2xf32>
97 return %d : vector<2x2xf32>
101 func.func @m16n8k8_fp32_vector_shape_a_extended(%arg0: vector<1x4xf32>, %arg1: vector<2x1xf32>, %arg2: vector<2x2xf32>) -> vector<2x2xf32> {
102 // expected-error @+1 {{expected matrix A to be shaped (4 x 1)}}
103 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 8]} : (vector<1x4xf32>, vector<2x1xf32>, vector<2x2xf32>) -> vector<2x2xf32>
104 return %d : vector<2x2xf32>
108 func.func @m8n8k4_fp64_vector_shape_a(%arg0: vector<1x2xf64>, %arg1: vector<1x1xf64>, %arg2: vector<1x2xf64>) -> vector<1x2xf64> {
109 // expected-error @+1 {{expected 32 warp-wide matrix A elements}}
110 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [8, 8, 4]} : (vector<1x2xf64>, vector<1x1xf64>, vector<1x2xf64>) -> vector<1x2xf64>
111 return %d : vector<1x2xf64>
115 func.func @m8n8k4_fp64_vector_shape_c_extended(%arg0: vector<1x1xf64>, %arg1: vector<1x1xf64>, %arg2: vector<2x1xf64>) -> vector<2x1xf64> {
116 // expected-error @+1 {{expected matrix C to be shaped (1 x 2)}}
117 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [8, 8, 4]} : (vector<1x1xf64>, vector<1x1xf64>, vector<2x1xf64>) -> vector<2x1xf64>
118 return %d : vector<2x1xf64>
122 func.func @m16n8k32_int8_vector_shape_b(%arg0: vector<4x4xi8>, %arg1: vector<4x4xi8>, %arg2: vector<2x2xi32>) -> vector<2x2xi32> {
123 // expected-error @+1 {{expected 256 warp-wide matrix B elements}}
124 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 32]} : (vector<4x4xi8>, vector<4x4xi8>, vector<2x2xi32>) -> vector<2x2xi32>
125 return %d : vector<2x2xi32>
129 func.func @m16n8k32_int32_datatype(%arg0: vector<4x4xi32>, %arg1: vector<2x4xi8>, %arg2: vector<2x2xi32>) -> vector<2x2xi32> {
130 // expected-error @+1 {{op failed to verify that matrixA and matrixB have same element type}}
131 %d = nvgpu.mma.sync (%arg0, %arg1, %arg2) {mmaShape = [16, 8, 32]} : (vector<4x4xi32>, vector<2x4xi8>, vector<2x2xi32>) -> vector<2x2xi32>
132 return %d : vector<2x2xi32>
136 func.func @async_cp_memory_space(%dst : memref<16xf32>, %src : memref<16xf32>, %i : index) -> () {
137 // expected-error @below {{destination memref must have a memory space attribute of IntegerAttr(3) or gpu::AddressSpaceAttr(Workgroup)}}
138 nvgpu.device_async_copy %src[%i], %dst[%i], 16 : memref<16xf32> to memref<16xf32>
143 func.func @async_cp_memref_type(%dst : memref<16xi32, 3>, %src : memref<16xf32>, %i : index) -> () {
144 // expected-error @+1 {{source and destination must have the same element type}}
145 nvgpu.device_async_copy %src[%i], %dst[%i], 16 : memref<16xf32> to memref<16xi32, 3>
150 func.func @async_cp_num_src_indices(%dst : memref<16xf32, 3>, %src : memref<16x16xf32>, %i : index) -> () {
151 // expected-error @+1 {{expected 2 source indices, got 1}}
152 nvgpu.device_async_copy %src[%i], %dst[%i], 16 : memref<16x16xf32> to memref<16xf32, 3>
157 func.func @async_cp_num_dst_indices(%dst : memref<16x16xf32, 3>, %src : memref<16xf32>, %i : index) -> () {
158 // expected-error @+1 {{expected 2 destination indices, got 1}}
159 nvgpu.device_async_copy %src[%i], %dst[%i], 16 : memref<16xf32> to memref<16x16xf32, 3>
164 func.func @async_cp_num_src_stride(
165 %dst : memref<200x100xf32, 3>,
166 %src : memref<200x100xf32, affine_map<(d0, d1) -> (200*d0 + 2*d1)>>,
168 // expected-error @+1 {{source memref most minor dim must have unit stride}}
169 nvgpu.device_async_copy %src[%i, %i], %dst[%i, %i], 16 :
170 memref<200x100xf32, affine_map<(d0, d1) -> (200*d0 + 2*d1)>> to memref<200x100xf32, 3>
175 func.func @async_cp_num_dst_stride(
176 %dst : memref<200x100xf32, affine_map<(d0, d1) -> (200*d0 + 2*d1)>, 3>,
177 %src : memref<200x100xf32>,
179 // expected-error @+1 {{destination memref most minor dim must have unit stride}}
180 nvgpu.device_async_copy %src[%i, %i], %dst[%i, %i], 16 :
181 memref<200x100xf32> to memref<200x100xf32, affine_map<(d0, d1) -> (200*d0 + 2*d1)>, 3>
186 // 42 is never the answer!
187 func.func @mma_sp_sync_f16_16816(%arg0: vector<2x2xf16>,
188 %arg1: vector<2x2xf16>,
189 %arg2: vector<2x2xf16>,
190 %arg3: vector<2xi16>) -> vector<2x2xf16> {
191 // expected-error @+1 {{'nvgpu.mma.sp.sync' op sparsity selector should be 0 or 1}}
192 %d = nvgpu.mma.sp.sync(%arg0, %arg1, %arg2) metadata(%arg3) {mmaShape = [16, 8, 16], sparsitySelector = 42 : i32} :
193 (vector<2x2xf16>, vector<2x2xf16>, vector<2x2xf16>) -> vector<2x2xf16>
194 return %d : vector<2x2xf16>
199 func.func @async_cp_zfill_f32_align1(
200 %src: memref<128x128xf32>, %dst: memref<3x16x128xf32, 3>, %i : index, %srcElements : index) {
201 // expected-error @+1 {{'nvgpu.device_async_copy' op bypassL1 does not satify alignment for 'memref<3x16x128xf32, 3>' with destination element 1. Unset bypassL1, or set destination element to 4}}
202 %0 = nvgpu.device_async_copy %src[%i, %i], %dst[%i, %i, %i], 1, %srcElements {bypassL1} : memref<128x128xf32> to memref<3x16x128xf32, 3>
208 func.func @async_cp_size_invalid_f32(
209 %src: memref<128x128xf32>, %dst: memref<3x16x128xf32, 3>, %i : index) {
210 // expected-error @+1 {{Requested copy elements is 3 with width 32. But copy elements could be one of 1, 2, 4.}}
211 %0 = nvgpu.device_async_copy %src[%i, %i], %dst[%i, %i, %i], 3: memref<128x128xf32> to memref<3x16x128xf32, 3>
217 func.func @async_cp_size_invalid_f16(
218 %src: memref<128x128xf16>, %dst: memref<3x16x128xf16, 3>, %i : index) {
219 // expected-error @+1 {{Requested copy elements is 3 with width 16. But copy elements could be one of 2, 4, 8.}}
220 %0 = nvgpu.device_async_copy %src[%i, %i], %dst[%i, %i, %i], 3: memref<128x128xf16> to memref<3x16x128xf16, 3>
226 func.func @async_cp_size_invalid_f64(
227 %src: memref<128x128xf64>, %dst: memref<3x16x128xf64, 3>, %i : index) {
228 // expected-error @+1 {{Requested copy elements is 3 with width 64. But copy elements could be one of 1, 2.}}
229 %0 = nvgpu.device_async_copy %src[%i, %i], %dst[%i, %i, %i], 3: memref<128x128xf64> to memref<3x16x128xf64, 3>
235 !tResult = !nvgpu.warpgroup.accumulator<fragmented = vector<128x128xf32>>
236 !tDescA = !nvgpu.warpgroup.descriptor<tensor = memref<128x64xf16, 3>>
237 !tDescB = !nvgpu.warpgroup.descriptor<tensor = memref<64x121xf16, 3>>
239 func.func @warpgroup_mma_wrong_input(%descA: !tDescA, %descB: !tDescB, %acc: !tResult) {
240 // expected-error @+1 {{'nvgpu.warpgroup.mma' op 2nd dim matrix-B ( 121 ) != 2nd dim matrix-C ( 128 )}}
241 %0 = nvgpu.warpgroup.mma %descA, %descB, %acc: !tDescA, !tDescB, !tResult -> !tResult
247 !tResult = !nvgpu.warpgroup.accumulator<fragmented = vector<128xf32>>
248 !tDescA = !nvgpu.warpgroup.descriptor<tensor = memref<128x64xf16, 3>>
249 !tDescB = !nvgpu.warpgroup.descriptor<tensor = memref<64x128xf16, 3>>
250 func.func @warpgroup_mma_wrong_input(%descA: !tDescA, %descB: !tDescB, %acc: !tResult) {
251 // expected-error @+1 {{'nvgpu.warpgroup.mma' op has matrices A, B, C and D, they must be 2 dimensional}}
252 %0 = nvgpu.warpgroup.mma %descA, %descB, %acc: !tDescA, !tDescB, !tResult -> !tResult
257 !tResult = !nvgpu.warpgroup.accumulator<fragmented = vector<128x128xf32>>
258 !tDescA = !nvgpu.warpgroup.descriptor<tensor = memref<128x64xf16, 3>>
259 !tDescB = !nvgpu.warpgroup.descriptor<tensor = memref<64x128xf32, 3>>
260 func.func @warpgroup_mma_wrong_input(%descA: !tDescA, %descB: !tDescB, %acc: !tResult) {
261 // expected-error @+1 {{'nvgpu.warpgroup.mma' op 'f32' += 'f16' * 'f32', it is not supported.}}
262 %0 = nvgpu.warpgroup.mma %descA, %descB, %acc: !tDescA, !tDescB, !tResult -> !tResult
268 !tResult = !nvgpu.warpgroup.accumulator<fragmented = vector<128x128xf32>>
269 !tDescA = !nvgpu.warpgroup.descriptor<tensor = memref<128x64xf16, 3>>
270 !tDescB = !nvgpu.warpgroup.descriptor<tensor = memref<64x512xf16, 3>>
271 func.func @warpgroup_mma_wrong_input(%descA: !tDescA, %descB: !tDescB, %acc: !tResult) {
272 // expected-error @+1 {{'nvgpu.warpgroup.mma' op 2nd dim matrix-B ( 512 ) != 2nd dim matrix-C ( 128 )}}
273 %0 = nvgpu.warpgroup.mma %descA, %descB, %acc: !tDescA, !tDescB, !tResult -> !tResult
279 !desc = !nvgpu.tensormap.descriptor<tensor = memref<32x32xf32,3>, swizzle=swizzle_32b, l2promo = none, oob = zero, interleave = none>
280 !mbarrier = !nvgpu.mbarrier.group<memorySpace = #gpu.address_space<workgroup>>
281 func.func @tma_load_1(%desc: !desc, %buffer1: memref<128xf32,3>, %buffer2: memref<32x32xf32,3>, %buffer3: memref<32x32xf32>, %mbarrier: !mbarrier) {
282 %c0 = arith.constant 0 : index
284 nvgpu.tma.async.load %desc[%c0, %c0], %mbarrier[%c0] to %buffer2 : !desc, !mbarrier -> memref<32x32xf32,3>
285 // expected-error @+1 {{Maximum 5 coordinates are supported.}}
286 nvgpu.tma.async.load %desc[%c0, %c0, %c0, %c0, %c0, %c0], %mbarrier[%c0] to %buffer2 : !desc, !mbarrier -> memref<32x32xf32,3>
291 !desc = !nvgpu.tensormap.descriptor<tensor = memref<32x32xf32>, swizzle=swizzle_32b, l2promo = none, oob = zero, interleave = none>
292 !mbarrier = !nvgpu.mbarrier.group<memorySpace = #gpu.address_space<workgroup>>
293 func.func @tma_load_2(%desc: !desc, %buffer1: memref<128xf32,3>, %buffer2: memref<32x32xf32,3>, %buffer3: memref<32x32xf32>, %mbarrier: !mbarrier) {
294 %c0 = arith.constant 0 : index
295 // expected-error @+1 {{the tensor map descriptor has incorrect address space, it must be shared memory address space.}}
296 nvgpu.tma.async.load %desc[%c0, %c0], %mbarrier[%c0] to %buffer2 : !desc, !mbarrier -> memref<32x32xf32,3>
301 !desc = !nvgpu.tensormap.descriptor<tensor = memref<32x32xf32,3>, swizzle=swizzle_32b, l2promo = none, oob = zero, interleave = none>
302 !mbarrier = !nvgpu.mbarrier.group<memorySpace = #gpu.address_space<workgroup>>
303 func.func @tma_load_3(%desc: !desc, %buffer1: memref<128xf32,3>, %buffer2: memref<32x32xf32,3>, %buffer3: memref<32x32xf32>, %mbarrier: !mbarrier) {
304 %c0 = arith.constant 0 : index
305 // expected-error @+1 {{the destination memref has incorrect address space, it must be shared memory address space}}
306 nvgpu.tma.async.load %desc[%c0, %c0], %mbarrier[%c0] to %buffer3 : !desc, !mbarrier -> memref<32x32xf32>
311 !desc = !nvgpu.tensormap.descriptor<tensor = memref<32x32xf32,3>, swizzle=swizzle_32b, l2promo = none, oob = zero, interleave = none>
312 !mbarrier = !nvgpu.mbarrier.group<memorySpace = #gpu.address_space<workgroup>>
313 func.func @tma_load_4(%desc: !desc, %buffer1: memref<128xf32,3>, %buffer2: memref<32x32xf32,3>, %buffer3: memref<32x32xf32>, %mbarrier: !mbarrier) {
314 %c0 = arith.constant 0 : index
315 // expected-error @+1 {{the shape of tensor map descriptor and memref must have same rank}}
316 nvgpu.tma.async.load %desc[%c0, %c0], %mbarrier[%c0] to %buffer1 : !desc, !mbarrier -> memref<128xf32,3>
322 !desc = !nvgpu.tensormap.descriptor<tensor = memref<64x128xf16,3>, swizzle=swizzle_32b, l2promo = none, oob = zero, interleave = none>
323 func.func @tma_generate_descriptor_incorrect_last_dim(%b0 : index, %b1 : index, %mem : memref<*xf16>) {
324 // expected-error @+1 {{the tensormap descriptor must have last dimension of 128 bytes but it is 256 bytes}}
325 %descA = nvgpu.tma.create.descriptor %mem box[%b0, %b1] : memref<*xf16> -> !desc
331 !desc = !nvgpu.tensormap.descriptor<tensor = memref<64x128xf32,3>, swizzle=swizzle_32b, l2promo = none, oob = zero, interleave = none>
332 !mbarrier = !nvgpu.mbarrier.group<memorySpace = #gpu.address_space<workgroup>>
333 func.func @tma_generate_descriptor_incorrect_last_dim(%desc: !desc, %buffer2: memref<64x128xf32,3>, %mbarrier: !mbarrier) {
334 %c0 = arith.constant 0 : index
335 // expected-error @+1 {{the tensormap descriptor must have last dimension of 128 bytes but it is 512 bytes}}
336 nvgpu.tma.async.load %desc[%c0, %c0], %mbarrier[%c0] to %buffer2 : !desc, !mbarrier -> memref<64x128xf32,3>
341 func.func @rcp_unsupported_rounding_0(%in : vector<16xf32>) {
342 // expected-error @+1 {{'nvgpu.rcp' op has a limitation. #nvgpu<rcp_rounding_mode rn> or non-ftz is not supported yet.}}
343 %out = nvgpu.rcp %in {rounding = rn, ftz} : vector<16xf32>
347 func.func @rcp_unsupported_rounding_1(%in : vector<16xf32>) {
348 // expected-error @+1 {{'nvgpu.rcp' op has a limitation. #nvgpu<rcp_rounding_mode rz> or non-ftz is not supported yet.}}
349 %out = nvgpu.rcp %in {rounding = rz} : vector<16xf32>
353 func.func @rcp_unsupported_ftz(%in : vector<16xf32>) {
354 // expected-error @+1 {{'nvgpu.rcp' op has a limitation. #nvgpu<rcp_rounding_mode approx> or non-ftz is not supported yet.}}
355 %out = nvgpu.rcp %in {rounding = approx} : vector<16xf32>