Fix AxisInfo handling of PoisonOp producing MemDesc (#8489) (#5501)

The PoisonOpAxisInfoVisitor incorrectly returns rank=1 for ub.poison
operations producing pointer-to-tensor types like
!tt.ptr<tensor<128x64xf16>>. This wrong rank propagates through
unrealized_conversion_cast operations created during lowering, causing
assertion failures in AxisInfo::join().

Fixes #5464

Author: whitneywhtsang

lib/Analysis/AxisInfo.cpp

@@ -276,8 +276,12 @@ class PoisonOpAxisInfoVisitor final : public AxisInfoVisitorImpl<ub::PoisonOp> {
   getAxisInfo(ub::PoisonOp op,
               ArrayRef<const dataflow::Lattice<AxisInfo> *> operands) override {
     unsigned rank = 1;
-    if (auto shape = dyn_cast<mlir::ShapedType>(op.getType()))
+    if (auto shape = dyn_cast<RankedTensorType>(op.getType())) {
       rank = shape.getRank();
+    } else if (auto ptrTy = dyn_cast<PointerType>(op.getType())) {
+      if (auto tensorType = dyn_cast<RankedTensorType>(ptrTy.getPointeeType()))
+        rank = tensorType.getRank();
+    }
 
     // Poison values are never accessed, thus assume optimistic values.
     return AxisInfo(AxisInfo::DimVectorT(rank, kMaxDivisor),
@@ -1229,6 +1233,7 @@ void AxisInfo::initDimVectorFromHint(Attribute attr, DimVectorT *vec) {
     return rhs;
   if (rhs.getRank() == 0)
     return lhs;
+  assert(lhs.getRank() == rhs.getRank() && "Mismatched ranks");
   DimVectorT contiguity;
   DimVectorT divisibility;
   DimVectorT constancy;

test/TritonGPU/pipeline-assign-latencies.mlir

@@ -1149,7 +1149,23 @@ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.targ
 }
 
 // -----
+// Test that ub.poison producing a memdesc does not get treated like a tensor
+// value in AxisInfo analysis.
+#shared = #ttg.nvmma_shared<{swizzlingByteWidth = 128, transposed = false, elementBitWidth = 16}>
+#smem = #ttg.shared_memory
+module attributes {"ttg.num-warps" = 4 : i32} {
+  tt.func public @minimal_crash(%lb: i32, %ub: i32) -> !ttg.memdesc<2x2xf16, #shared, #smem, mutable> {
+    %c1 = arith.constant 1 : i32
+    %poison = ub.poison : !ttg.memdesc<2x2xf16, #shared, #smem, mutable>
+    %normal = ttg.local_alloc : () -> !ttg.memdesc<2x2xf16, #shared, #smem, mutable>
+    %result = scf.for %i = %lb to %ub step %c1 iter_args(%current = %poison) -> !ttg.memdesc<2x2xf16, #shared, #smem, mutable> : i32 {
+      scf.yield %normal : !ttg.memdesc<2x2xf16, #shared, #smem, mutable>
+    }
+    tt.return %result : !ttg.memdesc<2x2xf16, #shared, #smem, mutable>
+  }
+}
 
+// -----
 #blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [4, 1], order = [1, 0]}>
 #blocked1 = #ttg.blocked<{sizePerThread = [1, 128], threadsPerWarp = [32, 1], warpsPerCTA = [4, 1], order = [1, 0]}>
 #shared = #ttg.nvmma_shared<{swizzlingByteWidth = 128, transposed = false, elementBitWidth = 16}>

test/TritonIntelGPU/pipeline-assign-latencies.mlir

@@ -0,0 +1,29 @@
+// RUN: triton-opt %s -split-input-file -allow-unregistered-dialect -tritongpu-assign-latencies=num-stages=3 -canonicalize | FileCheck %s
+
+// Test that ub.poison producing a ptr<tensor> gets correct rank in AxisInfo
+// analysis (rank=2 for tensor<128x64>, not rank=1).
+#mma = #ttig.dpas<{repeatCount = 8, systolicDepth = 8, executionSize = 16, opsPerChan = 2, threadsPerWarp = 16, warpsPerCTA = [2, 2], repCluster = [4, 2], A = [32, 16], B = [16, 32], C = [32, 32]}>
+// CHECK-LABEL: @test_poison_rank
+module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, "ttg.threads-per-warp" = 16 : i32} {
+  tt.func public @test_poison_rank(%arg0: !tt.ptr<f16>, %lb: i32, %ub: i32) {
+    %c0_i32 = arith.constant 0 : i32
+    %c1_i32 = arith.constant 1 : i32
+    %c1_i64 = arith.constant 1 : i64
+    %c128_i64 = arith.constant 128 : i64
+    %c64_i64 = arith.constant 64 : i64
+
+    %0 = ub.poison : !tt.ptr<tensor<128x64xf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 1}>>>
+
+    %1 = tt.make_tensor_ptr %arg0, [%c128_i64, %c64_i64], [%c64_i64, %c1_i64], [%c0_i32, %c0_i32] {order = array<i32: 1, 0>} : <tensor<128x64xf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 1}>>>
+
+    %result = scf.for %i = %lb to %ub step %c1_i32
+      iter_args(%ptr = %0) -> !tt.ptr<tensor<128x64xf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 1}>>> : i32 {
+
+      %advanced = tt.advance %ptr, [%c0_i32, %c0_i32] : <tensor<128x64xf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 1}>>>
+
+      scf.yield %advanced : !tt.ptr<tensor<128x64xf16, #ttg.dot_op<{opIdx = 0, parent = #mma, kWidth = 1}>>>
+    }
+
+    tt.return
+  }
+}

third_party/intel/lib/Analysis/AxisInfo.cpp

@@ -295,7 +295,12 @@ class PoisonOpAxisInfoVisitor final : public AxisInfoVisitorImpl<ub::PoisonOp> {
               ArrayRef<const dataflow::Lattice<AxisInfo> *> operands) override {
     constexpr int64_t largePowerOf2 = int64_t(1) << 32;
     // Poison values are never accessed, thus assume optimistic values.
-    if (auto shape = dyn_cast<mlir::ShapedType>(op.getType())) {
+    Type type = op.getType();
+    if (auto ptrTy = dyn_cast<triton::PointerType>(type)) {
+      type = ptrTy.getPointeeType();
+    }
+
+    if (auto shape = dyn_cast<mlir::ShapedType>(type)) {
       unsigned rank = shape.getRank();
       return AxisInfo(
           /*contiguity=*/AxisInfo::DimVectorT(rank, largePowerOf2),