bbf362b784
The entire compiler now looks at structural properties of the function (e.g. does it have one block, does it contain an if/for stmt, etc) so the only thing holding up this difference is round tripping through the parser/printer syntax. Removing this shrinks the compile by ~140LOC. This is step 31/n towards merging instructions and statements. The last step is updating the docs, which I will do as a separate patch in order to split it from this mostly mechanical patch. PiperOrigin-RevId: 227540453
29 lines
1.1 KiB
MLIR
29 lines
1.1 KiB
MLIR
// RUN: mlir-opt %s -vectorize -virtual-vector-size 32 -virtual-vector-size 64 -virtual-vector-size 256 --test-fastest-varying=2 --test-fastest-varying=1 --test-fastest-varying=0 | FileCheck %s
|
|
|
|
// Permutation maps used in vectorization.
|
|
// CHECK: #[[map_proj_d0d1d2_d0d1d2:map[0-9]+]] = (d0, d1, d2) -> (d0, d1, d2)
|
|
|
|
func @vec3d(%A : memref<?x?x?xf32>) {
|
|
%0 = dim %A, 0 : memref<?x?x?xf32>
|
|
%1 = dim %A, 1 : memref<?x?x?xf32>
|
|
%2 = dim %A, 2 : memref<?x?x?xf32>
|
|
// CHECK: for %i0 = 0 to %0 {
|
|
// CHECK: for %i1 = 0 to %0 {
|
|
// CHECK: for %i2 = 0 to %0 step 32 {
|
|
// CHECK: for %i3 = 0 to %1 step 64 {
|
|
// CHECK: for %i4 = 0 to %2 step 256 {
|
|
// CHECK: %3 = vector_transfer_read %arg0, %i2, %i3, %i4 {permutation_map: #[[map_proj_d0d1d2_d0d1d2]]} : (memref<?x?x?xf32>, index, index, index) -> vector<32x64x256xf32>
|
|
for %t0 = 0 to %0 {
|
|
for %t1 = 0 to %0 {
|
|
for %i0 = 0 to %0 {
|
|
for %i1 = 0 to %1 {
|
|
for %i2 = 0 to %2 {
|
|
%a2 = load %A[%i0, %i1, %i2] : memref<?x?x?xf32>
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return
|
|
}
|