[Mlir-commits] [mlir] [mlir][gpu] Support arith.truncf in subgroup MMA elementwise ops (PR #182499)

[Simone Pellegrini]

================
@@ -270,6 +271,8 @@ convertElementwiseOpToMMA(Operation *op) {
     return gpu::MMAElementwiseOp::NEGATEF;
   if (isa<arith::ExtFOp>(op))
     return gpu::MMAElementwiseOp::EXTF;
+  if (isa<arith::TruncFOp>(op))
+    return gpu::MMAElementwiseOp::TRUNCF;
   return std::nullopt;
----------------
simpel01 wrote:

I will never shy away from such challenge!

https://github.com/llvm/llvm-project/pull/182499