cb4ccd38fa
Since the recent MemRef refactoring that centralizes the lowering of complex MemRef operations outside of the conversion framework, the MemRefToLLVM pass doesn't directly convert these complex operations. Instead, to fully convert the whole MemRef dialect space, MemRefToLLVM needs to run after `expand-strided-metadata`. Make this more obvious by changing the name of the pass and the option associated with it from `convert-memref-to-llvm` to `finalize-memref-to-llvm`. The word "finalize" conveys that this pass needs to run after something else and that something else is documented in its tablegen description. This is a follow-up patch related to the conversation at: https://discourse.llvm.org/t/psa-you-need-to-run-expand-strided-metadata-before-memref-to-llvm-now/66956/14 Differential Revision: https://reviews.llvm.org/D142463
97 lines
4.7 KiB
MLIR
97 lines
4.7 KiB
MLIR
// RUN: mlir-opt %s -convert-vector-to-scf -lower-affine -convert-scf-to-cf -convert-vector-to-llvm="enable-amx" -finalize-memref-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts | \
|
|
// RUN: mlir-translate -mlir-to-llvmir | \
|
|
// RUN: %lli --entry-function=entry --mattr="+amx-tile,+amx-int8,+amx-bf16" --dlopen=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
|
|
// RUN: FileCheck %s
|
|
|
|
// Note: To run this test, your CPU must support AMX.
|
|
|
|
func.func @tilezero(%arg0: memref<?x?xi32>, %i: index, %j: index) {
|
|
%1 = amx.tile_zero : vector<16x16xi32>
|
|
amx.tile_store %arg0[%i, %j], %1 : memref<?x?xi32>, vector<16x16xi32>
|
|
return
|
|
}
|
|
|
|
func.func @entry() -> i32 {
|
|
%i0 = arith.constant 0: i32
|
|
%i1 = arith.constant 1: i32
|
|
%c0 = arith.constant 0: index
|
|
%c1 = arith.constant 1: index
|
|
%c3 = arith.constant 3: index
|
|
%c19 = arith.constant 19: index
|
|
|
|
// Set up memory.
|
|
%a = memref.alloc(%c19, %c19) : memref<?x?xi32>
|
|
scf.for %i = %c0 to %c19 step %c1 {
|
|
scf.for %j = %c0 to %c19 step %c1 {
|
|
memref.store %i1, %a[%i, %j] : memref<?x?xi32>
|
|
}
|
|
}
|
|
|
|
// Call kernel.
|
|
call @tilezero(%a, %c1, %c1) : (memref<?x?xi32>, index, index) -> ()
|
|
|
|
// Print and verify that the tilezero is correctly strided within
|
|
// the enveloping 19x19 buffer.
|
|
//
|
|
// CHECK: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
|
|
//
|
|
scf.for %i = %c0 to %c19 step %c1 {
|
|
%av = vector.transfer_read %a[%i, %c0], %i0: memref<?x?xi32>, vector<19xi32>
|
|
vector.print %av : vector<19xi32>
|
|
}
|
|
|
|
// Call kernel with different indices.
|
|
call @tilezero(%a, %c0, %c3) : (memref<?x?xi32>, index, index) -> ()
|
|
|
|
// Print and verify that the tilezero is again correctly strided
|
|
// within the enveloping 19x19 buffer.
|
|
//
|
|
// CHECK-NEXT: ( 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 )
|
|
// CHECK-NEXT: ( 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
|
|
// CHECK-NEXT: ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 )
|
|
//
|
|
scf.for %i = %c0 to %c19 step %c1 {
|
|
%av = vector.transfer_read %a[%i, %c0], %i0: memref<?x?xi32>, vector<19xi32>
|
|
vector.print %av : vector<19xi32>
|
|
}
|
|
|
|
// Release resources.
|
|
memref.dealloc %a : memref<?x?xi32>
|
|
|
|
return %i0 : i32
|
|
}
|