[llvm] [AMDGPU] Add RewriteMFMAForm scheduling stage tests (PR #196008)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,512 @@
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a \
+; RUN:     -amdgpu-disable-rewrite-mfma-form-sched-stage=false \
+; RUN:     < %s | FileCheck %s
+;
+; Verify RewriteMFMAFormStage output:
+;   Case 3: v_accvgpr_write inserted in entry block — zeroinitializer acc
+;           converted VGPR->AGPR before first loop iteration.
+;   Loop:   all 9 MFMAs rewritten to AGPR form (a[...] C/D operands).
+;   Case 2: v_accvgpr_read inserted at exit block entry (ReachingUseTracker)
+;           to convert AGPR MFMA results -> VGPR before fptrunc (V_CVT_F16_F32).
+;
----------------
arsenm wrote:

Same, replace AI description 

https://github.com/llvm/llvm-project/pull/196008