This is an archive of the discontinued LLVM Phabricator instance.

Differential D118292

[mlir][amx] add a full tile matrix mult integral example to integration tests
ClosedPublic

Authored by aartbik on Jan 26 2022, 4:08 PM.

Details

Reviewers
nicolasvasilache
dcaballe
Commits
rG11fb049954ba: [mlir][amx] add a full tile matrix mult integral example to integration tests

Diff Detail

Event Timeline

aartbik created this revision.Jan 26 2022, 4:08 PM
Herald added subscribers: sdasgup3, wenzhicui, wrengr and 20 others. · View Herald TranscriptJan 26 2022, 4:08 PM
aartbik requested review of this revision.Jan 26 2022, 4:08 PM
Herald added a reviewer: nicolasvasilache. · View Herald TranscriptJan 26 2022, 4:08 PM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: stephenneuendorffer, nicolasvasilache. · View Herald Transcript
aartbik updated this revision to Diff 403445.Jan 26 2022, 4:55 PM

fixed tensor issue in original test with better flags

aartbik added a reviewer: dcaballe.Jan 26 2022, 4:56 PM
dcaballe accepted this revision.Jan 26 2022, 5:00 PM

Thanks, Aart!

This revision is now accepted and ready to land.Jan 26 2022, 5:00 PM
This revision was landed with ongoing or failed builds.Jan 26 2022, 5:10 PM
Closed by commit rG11fb049954ba: [mlir][amx] add a full tile matrix mult integral example to integration tests (authored by aartbik). · Explain Why
This revision was automatically updated to reflect the committed changes.
aartbik added a commit: rG11fb049954ba: [mlir][amx] add a full tile matrix mult integral example to integration tests.
bondhugula added a subscriber: bondhugula.Jan 26 2022, 6:28 PM
bondhugula added inline comments.
mlir/test/Integration/Dialect/Vector/CPU/AMX/test-muli-full.mlir
101–103

Since this is an integration test, it's more compact and readable to just use a splat constant and lower it? (or just a alloc + fill). Also, it wouldn't be great to "temporary" bufferization ops in an integration test.

Harbormaster completed remote builds in B145879: Diff 403445.Jan 27 2022, 7:04 AM

Revision Contents

PathSize
mlir/
test/
Integration/
Dialect/
Vector/
CPU/
AMX/
23 lines
137 lines

Diff 403446

mlir/test/Integration/Dialect/Vector/CPU/AMX/test-mulf-full.mlir

// RUN: mlir-opt %s -convert-vector-to-scf -lower-affine -convert-scf-to-std -convert-vector-to-llvm="enable-amx" -convert-memref-to-llvm -convert-std-to-llvm -reconcile-unrealized-casts | \ // RUN: mlir-opt %s -convert-vector-to-scf -lower-affine -convert-scf-to-std \
// RUN: -tensor-constant-bufferize -convert-vector-to-llvm="enable-amx" \
// RUN: -convert-memref-to-llvm -convert-std-to-llvm -reconcile-unrealized-casts | \
// RUN: mlir-translate -mlir-to-llvmir | \ // RUN: mlir-translate -mlir-to-llvmir | \
// RUN: %lli --entry-function=entry --mattr="+amx-tile,+amx-int8,+amx-bf16" --dlopen=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \ // RUN: %lli --entry-function=entry --mattr="+amx-tile,+amx-int8,+amx-bf16" \
// RUN: --dlopen=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
// RUN: FileCheck %s // RUN: FileCheck %s
// Note: To run this test, your CPU must support AMX. // Note: To run this test, your CPU must support AMX.
// Multiply full size tiles into zero destination. // Multiply full size tiles into zero destination.
func @kernel(%arg0: memref<16x32xbf16>, func @kernel(%arg0: memref<16x32xbf16>,
%arg1: memref<16x32xbf16>, %arg1: memref<16x32xbf16>,
%arg2: memref<16x16xf32>) { %arg2: memref<16x16xf32>) {
%0 = arith.constant 0 : index %0 = arith.constant 0 : index
%1 = amx.tile_load %arg0[%0, %0] : memref<16x32xbf16> into vector<16x32xbf16> %1 = amx.tile_load %arg0[%0, %0] : memref<16x32xbf16> into vector<16x32xbf16>
%2 = amx.tile_load %arg1[%0, %0] : memref<16x32xbf16> into vector<16x32xbf16> %2 = amx.tile_load %arg1[%0, %0] : memref<16x32xbf16> into vector<16x32xbf16>
%3 = amx.tile_zero : vector<16x16xf32> %3 = amx.tile_zero : vector<16x16xf32>
%4 = amx.tile_mulf %1, %2, %3 : vector<16x32xbf16>, vector<16x32xbf16>, vector<16x16xf32> %4 = amx.tile_mulf %1, %2, %3 : vector<16x32xbf16>, vector<16x32xbf16>, vector<16x16xf32>
amx.tile_store %arg2[%0, %0], %4 : memref<16x16xf32>, vector<16x16xf32> amx.tile_store %arg2[%0, %0], %4 : memref<16x16xf32>, vector<16x16xf32>
return return
} }
func @entry() -> i32 { func @entry() -> i32 {
%fu = arith.constant -1.0: f32 %fu = arith.constant -1.0: f32
%c0 = arith.constant 0: index %c0 = arith.constant 0: index
%c1 = arith.constant 1: index %c1 = arith.constant 1: index
%c16 = arith.constant 16: index %c16 = arith.constant 16: index
%c32 = arith.constant 32: index %c32 = arith.constant 32: index
// Setup simple test data. Note that bf16 does not seem to work well with the // Setup simple test data.
// tensor type yet, which is why we use vectors and transfer the data into memref.
// TODO: use tensors and bufferization.to_memref instead
%0 = arith.constant dense<[ %0 = arith.constant dense<[
[ 1.1, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.1, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.2, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.2, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.3, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.3, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.4, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.4, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
Show All 17 Lines %0 = arith.constant dense<[
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.4, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.4, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.3, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.3, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.2, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.2, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.1, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.1,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ] 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ]
]> : vector<16x32xbf16> ]> : tensor<16x32xbf16>
%1 = arith.constant dense<[ %1 = arith.constant dense<[
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.1, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.1, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.2, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.2, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.3, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.3, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.4, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.4, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
Show All 17 Lines %1 = arith.constant dense<[
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.4, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.4, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.3, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.3, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.2, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.2, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ],
[ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.1, 1.0, 1.0, 1.0, 1.0, 1.0, [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.1, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ] 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ]
]> : vector<16x32xbf16> ]> : tensor<16x32xbf16>
// Set up memory. // Set up memory.
%a = memref.alloc() : memref<16x32xbf16> %a = bufferization.to_memref %0 : memref<16x32xbf16>
%b = memref.alloc() : memref<16x32xbf16> %b = bufferization.to_memref %1 : memref<16x32xbf16>
%c = memref.alloc() : memref<16x16xf32> %c = memref.alloc() : memref<16x16xf32>
vector.transfer_write %0, %a[%c0, %c0] : vector<16x32xbf16>, memref<16x32xbf16>
vector.transfer_write %1, %b[%c0, %c0] : vector<16x32xbf16>, memref<16x32xbf16>
// Call kernel. // Call kernel.
call @kernel(%a, %b, %c) : (memref<16x32xbf16>, memref<16x32xbf16>, memref<16x16xf32>) -> () call @kernel(%a, %b, %c) : (memref<16x32xbf16>, memref<16x32xbf16>, memref<16x16xf32>) -> ()
// //
// Print and verify the 16x16 result. // Print and verify the 16x16 result.
// //
// CHECK: ( 32.1016, 34.3984, 34.5078, 33.6953, 32.9062, 32.2031, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016 ) // CHECK: ( 32.1016, 34.3984, 34.5078, 33.6953, 32.9062, 32.2031, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016 )
Show All 14 Linesfunc @entry() -> i32 {
// CHECK-NEXT: ( 32.1016, 34.3984, 34.5078, 33.6953, 32.9062, 32.2031, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016 ) // CHECK-NEXT: ( 32.1016, 34.3984, 34.5078, 33.6953, 32.9062, 32.2031, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016, 32.1016 )
// //
scf.for %i = %c0 to %c16 step %c1 { scf.for %i = %c0 to %c16 step %c1 {
%v = vector.transfer_read %c[%i, %c0], %fu: memref<16x16xf32>, vector<16xf32> %v = vector.transfer_read %c[%i, %c0], %fu: memref<16x16xf32>, vector<16xf32>
vector.print %v : vector<16xf32> vector.print %v : vector<16xf32>
} }
// Release resources. // Release resources.
memref.dealloc %a : memref<16x32xbf16>
memref.dealloc %b : memref<16x32xbf16>
memref.dealloc %c : memref<16x16xf32> memref.dealloc %c : memref<16x16xf32>
%i0 = arith.constant 0 : i32 %i0 = arith.constant 0 : i32
return %i0 : i32 return %i0 : i32
} }

mlir/test/Integration/Dialect/Vector/CPU/AMX/test-muli-full.mlir

  • This file was added.
// RUN: mlir-opt %s -convert-vector-to-scf -lower-affine -convert-scf-to-std \
// RUN: -tensor-constant-bufferize -convert-vector-to-llvm="enable-amx" \
// RUN: -convert-memref-to-llvm -convert-std-to-llvm -reconcile-unrealized-casts | \
// RUN: mlir-translate -mlir-to-llvmir | \
// RUN: %lli --entry-function=entry --mattr="+amx-tile,+amx-int8,+amx-bf16" \
// RUN: --dlopen=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \
// RUN: FileCheck %s
// Note: To run this test, your CPU must support AMX.
// Multiply full size tiles into zero destination.
func @kernel(%arg0: memref<16x64xi8>,
%arg1: memref<16x64xi8>,
%arg2: memref<16x16xi32>) {
%0 = arith.constant 0 : index
%1 = amx.tile_load %arg0[%0, %0] : memref<16x64xi8> into vector<16x64xi8>
%2 = amx.tile_load %arg1[%0, %0] : memref<16x64xi8> into vector<16x64xi8>
%3 = amx.tile_zero : vector<16x16xi32>
%4 = amx.tile_muli %1 zext, %2 zext, %3 : vector<16x64xi8>, vector<16x64xi8>, vector<16x16xi32>
amx.tile_store %arg2[%0, %0], %4 : memref<16x16xi32>, vector<16x16xi32>
return
}
func @entry() -> i32 {
%iu = arith.constant -1: i32
%c0 = arith.constant 0: index
%c1 = arith.constant 1: index
%c16 = arith.constant 16: index
// Setup simple test data.
%0 = arith.constant dense<[
[ 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ]
]> : tensor<16x64xi8>
%1 = arith.constant dense<[
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ],
[ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ]
]> : tensor<16x64xi8>
// Set up memory.
%a = bufferization.to_memref %0 : memref<16x64xi8>
%b = bufferization.to_memref %1 : memref<16x64xi8>
%c = memref.alloc() : memref<16x16xi32>
bondhugulaUnsubmitted
Not Done
ReplyInline Actions

Since this is an integration test, it's more compact and readable to just use a splat constant and lower it? (or just a alloc + fill). Also, it wouldn't be great to "temporary" bufferization ops in an integration test.

bondhugula: Since this is an integration test, it's more compact and readable to just use a splat constant…
// Call kernel.
call @kernel(%a, %b, %c) : (memref<16x64xi8>, memref<16x64xi8>, memref<16x16xi32>) -> ()
//
// Print and verify the 16x16 result.
//
// CHECK: ( 69, 69, 69, 69, 65, 65, 65, 65, 89, 89, 89, 89, 89, 89, 89, 75 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 72, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 72, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 72, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 72, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 75, 75, 75, 75, 71, 71, 71, 71, 95, 95, 95, 95, 95, 95, 95, 81 )
// CHECK-NEXT: ( 69, 69, 69, 69, 65, 65, 65, 65, 89, 89, 89, 89, 89, 89, 89, 75 )
//
scf.for %i = %c0 to %c16 step %c1 {
%v = vector.transfer_read %c[%i, %c0], %iu: memref<16x16xi32>, vector<16xi32>
vector.print %v : vector<16xi32>
}
// Release resources.
memref.dealloc %c : memref<16x16xi32>
%i0 = arith.constant 0 : i32
return %i0 : i32
}