Revert "Workaround distribute-to-warps.mlir failure"

This reverts commit 5529b87.

Author: Dewei-Wang-sh

lib/Dialect/TritonGPU/IR/Dialect.cpp

@@ -1046,10 +1046,6 @@ SmallVector<unsigned> DotOperandEncodingAttr::getCTASplitNum() const {
 SmallVector<unsigned> DotOperandEncodingAttr::getWarpsPerCTA() const {
   auto distributedLayout = mlir::cast<DistributedEncodingTrait>(getParent());
   auto warps = distributedLayout.getWarpsPerCTA();
-  // FIXME: This is a temporary solution to avoid distribute-to-warps.mlir
-  // failure.
-  if (mlir::triton::tools::getBoolEnv("TRITON_INTEL_ADVANCED_PATH"))
-    return warps;
   auto rank = warps.size();
   auto kDim = getOpIdx() == 0 ? rank - 1 : rank - 2;
   warps[kDim] = 1;

test/TritonIntelGPU/distribute-to-warps.mlir

@@ -1,5 +1,4 @@
-// FIXME: Remove the env. variable once issue #2529 is fixed.
-// RUN: env TRITON_INTEL_ADVANCED_PATH=1 triton-opt %s -split-input-file -tritonintelgpu-distribute-to-warps | FileCheck %s
+// RUN: triton-opt %s -split-input-file -tritonintelgpu-distribute-to-warps | FileCheck %s
 
 #blocked1 = #triton_gpu.blocked<{sizePerThread = [32, 32], threadsPerWarp = [1, 1], warpsPerCTA = [4, 1], order = [1, 0], CTAsPerCGA = [1, 1], CTASplitNum = [1, 1], CTAOrder = [0, 1]}>
 #blocked2 = #triton_gpu.blocked<{sizePerThread = [32, 32], threadsPerWarp = [1, 1], warpsPerCTA = [1, 4], order = [1, 0], CTAsPerCGA = [1, 1], CTASplitNum = [1, 1], CTAOrder = [0, 1]}>