[llvm] [AMDGPU] Validate WMMA scale/format combination (PR #203074)

[Shilei Tian via llvm-commits]

================
@@ -0,0 +1,20 @@
+; RUN: not llc -mtriple=amdgcn -mcpu=gfx1250 < %s 2>&1 | FileCheck -check-prefix=ERR %s
+
+; ERR: error: invalid matrix and scale format combination in wmma call
----------------
shiltian wrote:

This might be a debatable topic. From my perspective, if an intrinsic argument is an immediate value, it's our responsibility to verify that it is valid and semantically meaningful.

https://github.com/llvm/llvm-project/pull/203074