[llvm-branch-commits] [clang] [llvm] [mlir] [AMDGPU] Canonicalize num_records to its actual width in InstCombine (PR #217068)

[Krzysztof Drewniak via llvm-branch-commits]

================
@@ -353,6 +353,16 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
   bool hasRelaxedBufferOOBMode() const { return BufferOOBRelaxed; }
   bool hasRelaxedTBufferOOBMode() const { return TBufferOOBRelaxed; }
 
+  // Return the width of the num_records field in buffer resources if it
+  // has a known value.
+  std::optional<unsigned> getBufferResourceNumRecordsWidth() const {
+    if (AMDGPU::isPseudoTarget(getCPU()))
+      return std::nullopt;
+    if (has45BitNumRecordsBufferResource())
----------------
krzysz00 wrote:

Makes sense, will change that out

https://github.com/llvm/llvm-project/pull/217068