gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[TritonIR] make other optional and remove isOtherUnspecified (#67)

Browse Source 
[Triton] make other optional and remove isOtherUnspecified
This commit is contained in:
Shintaro Iwasaki
2022-08-18 18:19:55 -07:00
committed by GitHub
parent 192be76b3c
commit 9aa00249a6
14 changed files with 69 additions and 58 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
test/Analysis/test-alignment.mlir
View File

@@ -46,7 +46,7 @@ func @permute_2d(%arg0: !tt.ptr<f32> {tt.divisibility = 16 : i32}, %arg1: i32 {t
// CHECK-NEXT: Contiguity: [128, 1] ; Divisibility: [16, 1] ; Constancy: [1, 1]
%19 = tt.getelementptr %17, %18 : tensor<128x128x!tt.ptr<f32>>
// CHECK-NEXT: Contiguity: [1, 1] ; Divisibility: [1, 1] ; Constancy: [1, 1]
%20 = tt.load %10, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<128x128xf32>
%20 = tt.load %10, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<128x128xf32>
tt.store %19, %20, %cst, : tensor<128x128xf32>
return
}

16
test/Analysis/test-allocation.mlir
View File

@@ -21,10 +21,10 @@ func @matmul_loop(%lb : index, %ub : index, %step : index, %A : !tt.ptr<f16>, %B
%b_off = arith.constant dense<4> : tensor<32x128xi32, #BL>
scf.for %iv = %lb to %ub step %step iter_args(%a_ptr = %a_ptr_init, %b_ptr = %b_ptr_init, %prev_c = %c_init) -> (tensor<128x32x!tt.ptr<f16>, #AL>, tensor<32x128x!tt.ptr<f16>, #BL>, tensor<128x128xf32, #C>) {
%a_ = tt.load %a_ptr, %a_mask, %a_other {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<128x32xf16, #AL>
%a_ = tt.load %a_ptr, %a_mask, %a_other {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<128x32xf16, #AL>
// CHECK: offset = 0, size = 8192
%a = triton_gpu.convert_layout %a_ : (tensor<128x32xf16, #AL>) -> tensor<128x32xf16, #A>
%b_ = tt.load %b_ptr, %b_mask, %b_other {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<32x128xf16, #BL>
%b_ = tt.load %b_ptr, %b_mask, %b_other {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<32x128xf16, #BL>
// CHECK-NEXT: offset = 8192, size = 8192
%b = triton_gpu.convert_layout %b_ : (tensor<32x128xf16, #BL>) -> tensor<32x128xf16, #B>
@@ -49,17 +49,17 @@ func @reusable(%A : !tt.ptr<f16>) {
%a_ptr = tt.broadcast %A : (!tt.ptr<f16>) -> tensor<128x32x!tt.ptr<f16>, #AL>
%b_ptr = tt.broadcast %A : (!tt.ptr<f16>) -> tensor<32x128x!tt.ptr<f16>, #AL>
%a1_ = tt.load %a_ptr, %cst1, %cst2 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<128x32xf16, #AL>
// CHECK: offset = 0, size = 8192
%a1_ = tt.load %a_ptr, %cst1, %cst2 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<128x32xf16, #AL>
// CHECK: offset = 0, size = 8192
%a1 = triton_gpu.convert_layout %a1_ : (tensor<128x32xf16, #AL>) -> tensor<128x32xf16, #A>
%a2_ = tt.load %b_ptr, %cst3, %cst4 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<32x128xf16, #AL>
%a2_ = tt.load %b_ptr, %cst3, %cst4 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<32x128xf16, #AL>
// CHECK-NEXT: offset = 8192, size = 8192
%a2 = triton_gpu.convert_layout %a2_ : (tensor<32x128xf16, #AL>) -> tensor<32x128xf16, #A>
%a3_ = tt.load %a_ptr, %cst1, %cst2 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<128x32xf16, #AL>
%a3_ = tt.load %a_ptr, %cst1, %cst2 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<128x32xf16, #AL>
// CHECK-NEXT: offset = 16384, size = 8192
%a3 = triton_gpu.convert_layout %a3_ : (tensor<128x32xf16, #AL>) -> tensor<128x32xf16, #A>
%c = tt.dot %a1, %a2, %c_init {allowTF32 = true} : tensor<128x32xf16, #A> * tensor<32x128xf16, #B> -> tensor<128x128xf32, #C>
%a4_ = tt.load %b_ptr, %cst3, %cst4 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<32x128xf16, #AL>
%a4_ = tt.load %b_ptr, %cst3, %cst4 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<32x128xf16, #AL>
// CHECK-NEXT: offset = 0, size = 8192
%a4 = triton_gpu.convert_layout %a4_ : (tensor<32x128xf16, #AL>) -> tensor<32x128xf16, #A>
%c1 = tt.dot %a3, %a4, %c {allowTF32 = true} : tensor<128x32xf16, #A> * tensor<32x128xf16, #B> -> tensor<128x128xf32, #C>
@@ -210,5 +210,5 @@ func @multi_blocks_noreuse(%i1 : i1) {
// CHECK-NEXT: offset = 1024, size = 1024
%a = tt.cat %cst0, %cst1 {axis = 0} : (tensor<16x16xf16, #A>, tensor<16x16xf16, #A>) -> tensor<32x16xf16, #A>
return
// CHECK-NEXT: size = 3072
// CHECK-NEXT: size = 3072
}