69 lines
4.1 KiB
MLIR
69 lines
4.1 KiB
MLIR
// RUN: scalehls-opt -scalehls-loop-pipelining="pipeline-level=3 target-ii=2" %s | FileCheck %s
|
|
|
|
// CHECK-NOT: #map0 = affine_map<(d0) -> (d0)>
|
|
// CHECK-NOT: #map1 = affine_map<(d0) -> (d0 + 2)>
|
|
#map0 = affine_map<(d0) -> (d0)>
|
|
#map1 = affine_map<(d0) -> (d0 + 2)>
|
|
#map2 = affine_map<(d0) -> (d0 + 1)>
|
|
#set0 = affine_set<(d0, d1) : (d0 - d1 >= 0)>
|
|
#set1 = affine_set<(d0) : (d0 == 0)>
|
|
module {
|
|
func @test_syrk(%arg0: f32, %arg1: f32, %arg2: memref<16x16xf32>, %arg3: memref<16x16xf32>) attributes {func_directive = #hls.fd<pipeline=false, targetInterval=1, dataflow=false>, top_func} {
|
|
affine.for %arg4 = 0 to 16 step 2 {
|
|
affine.for %arg5 = 0 to 16 {
|
|
affine.for %arg6 = 0 to 16 {
|
|
|
|
// CHECK-NOT: affine.for %arg7 = #map0(%arg4) to #map1(%arg4) {
|
|
// CHECK-NOT: affine.for %arg8 = #map0(%arg5) to #map2(%arg5) {
|
|
// CHECK-NOT: affine.for %arg9 = #map0(%arg6) to #map2(%arg6) {
|
|
affine.for %arg7 = #map0(%arg4) to #map1(%arg4) {
|
|
affine.for %arg8 = #map0(%arg5) to #map2(%arg5) {
|
|
affine.for %arg9 = #map0(%arg6) to #map2(%arg6) {
|
|
affine.if #set0(%arg8, %arg9) {
|
|
%0 = affine.load %arg3[%arg8, %arg9] : memref<16x16xf32>
|
|
%1 = arith.mulf %arg1, %0 : f32
|
|
affine.if #set1(%arg7) {
|
|
affine.store %1, %arg3[%arg8, %arg9] : memref<16x16xf32>
|
|
}
|
|
%2 = affine.load %arg2[%arg8, %arg7] : memref<16x16xf32>
|
|
%3 = affine.load %arg2[%arg9, %arg7] : memref<16x16xf32>
|
|
%4 = affine.load %arg3[%arg8, %arg9] : memref<16x16xf32>
|
|
%5 = arith.mulf %arg0, %2 : f32
|
|
%6 = arith.mulf %5, %3 : f32
|
|
%7 = arith.addf %6, %4 : f32
|
|
affine.store %7, %arg3[%arg8, %arg9] : memref<16x16xf32>
|
|
}
|
|
|
|
// CHECK: %0 = affine.apply #map(%arg4)
|
|
// CHECK: affine.if #set0(%arg5, %arg6) {
|
|
// CHECK: %1 = affine.load %arg3[%arg5, %arg6] : memref<16x16xf32>
|
|
// CHECK: %2 = arith.mulf %arg1, %1 : f32
|
|
// CHECK: affine.if #set1(%0) {
|
|
// CHECK: affine.store %2, %arg3[%arg5, %arg6] : memref<16x16xf32>
|
|
// CHECK: }
|
|
// CHECK: %3 = affine.load %arg2[%arg5, %0] : memref<16x16xf32>
|
|
// CHECK: %4 = affine.load %arg2[%arg6, %0] : memref<16x16xf32>
|
|
// CHECK: %5 = affine.load %arg3[%arg5, %arg6] : memref<16x16xf32>
|
|
// CHECK: %6 = arith.mulf %arg0, %3 : f32
|
|
// CHECK: %7 = arith.mulf %6, %4 : f32
|
|
// CHECK: %8 = arith.addf %7, %5 : f32
|
|
// CHECK: affine.store %8, %arg3[%arg5, %arg6] : memref<16x16xf32>
|
|
// CHECK: }
|
|
|
|
// CHECK-NOT: } {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>, parallel}
|
|
// CHECK-NOT: } {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>, parallel}
|
|
// CHECK-NOT: } {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>}
|
|
} {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>, parallel}
|
|
} {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>, parallel}
|
|
} {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>}
|
|
|
|
// CHECK: } {loop_directive = #hls.ld<pipeline=true, targetII=2, dataflow=false, flatten=false>, parallel}
|
|
// CHECK: } {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=true>, parallel}
|
|
// CHECK: } {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=true>}
|
|
} {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>, parallel}
|
|
} {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>, parallel}
|
|
} {loop_directive = #hls.ld<pipeline=false, targetII=1, dataflow=false, flatten=false>}
|
|
return
|
|
}
|
|
}
|