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

[BACKEND] Added support for 1D conversion blocked -> slice (#831)

Browse Source
This commit is contained in:
Philippe Tillet
2022-11-01 13:19:58 -07:00
committed by GitHub
parent c9d84237e8
commit 12d60cb4a3
5 changed files with 103 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
test/Conversion/tritongpu_to_llvm.mlir
View File

@@ -715,6 +715,32 @@ module attributes {"triton_gpu.num-warps" = 1 : i32} {
}
}
// -----
#blocked0 = #triton_gpu.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}>
#blocked1 = #triton_gpu.blocked<{sizePerThread = [1, 4], threadsPerWarp = [4, 8], warpsPerCTA = [1, 1], order = [1, 0]}>
module attributes {"triton_gpu.num-warps" = 1 : i32} {
// CHECK-LABEL: convert_blocked1d_to_slice0
func @convert_blocked1d_to_slice0(%src:tensor<32xi32, #blocked0>) {
// CHECK-COUNT-4: llvm.load {{.*}} : !llvm.ptr<vector<1xi32>, 3>
%cvt = triton_gpu.convert_layout %src : (tensor<32xi32, #blocked0>) -> tensor<32xi32, #triton_gpu.slice<{dim = 0, parent = #blocked1}>>
return
}
}
// -----
#blocked0 = #triton_gpu.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}>
#blocked1 = #triton_gpu.blocked<{sizePerThread = [1, 4], threadsPerWarp = [4, 8], warpsPerCTA = [1, 1], order = [1, 0]}>
module attributes {"triton_gpu.num-warps" = 1 : i32} {
// CHECK-LABEL: convert_blocked1d_to_slice1
func @convert_blocked1d_to_slice1(%src:tensor<32xi32, #blocked0>) {
// CHECK-COUNT-32: llvm.load {{.*}} : !llvm.ptr<vector<1xi32>, 3>
%cvt = triton_gpu.convert_layout %src : (tensor<32xi32, #blocked0>) -> tensor<32xi32, #triton_gpu.slice<{dim = 1, parent = #blocked1}>>
return
}
}
// -----
#blocked = #triton_gpu.blocked<{sizePerThread = [1, 4], threadsPerWarp = [2, 16], warpsPerCTA = [1, 4], order = [1, 0]}>
#shared = #triton_gpu.shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [1, 0]}>