ROCM IFU: Fix reduce_slice lit test

Skip tritongpu_to_llvm_hopper test as it is nvidia specific
This commit is contained in:
Ognjen
2023-11-16 14:51:53 +00:00
committed by Jason Furmanek
parent 38fbb7e472
commit 9f3d6656a7
2 changed files with 4 additions and 2 deletions

View File

@@ -2062,8 +2062,8 @@ module attributes {"triton_gpu.compute-capability" = 80 : i32, "triton_gpu.num-c
// -----
// CHECK-LABEL: reduce_slice
// GCN: llvm.store
// GCN: llvm.load
// GCN-NOT: llvm.store
// GCN-NOT: llvm.load
// PTX-NOT: st.shared
// PTX-NOT: ld.shared
#blocked = #triton_gpu.blocked<{sizePerThread = [1, 1, 1], threadsPerWarp = [4, 4, 2], warpsPerCTA = [2, 4, 2], order = [2, 0, 1], CTAsPerCGA = [1, 1, 1], CTASplitNum = [1, 1, 1], CTAOrder = [0, 1, 2]}>

View File

@@ -1,4 +1,6 @@
// RUN: triton-opt %s -split-input-file --convert-triton-gpu-to-llvm=compute-capability=90 2>&1 | FileCheck %s
// This is nvidia specific test, so we skip it in amd case
// XFAIL: *
#blocked = #triton_gpu.blocked<{sizePerThread = [1, 4], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0], CTAsPerCGA = [1, 4], CTASplitNum = [1, 4], CTAOrder = [0, 1]}>
#shared = #triton_gpu.shared<{vec = 8, perPhase = 1, maxPhase = 8, order = [1, 0], CTAsPerCGA = [1, 4], CTASplitNum = [1, 4], CTAOrder = [0, 1], hasLeadingOffset = true}>