[llvm] 2183846 - [AMDGPU] Fix AGPR_32 reg assign for mfma scale ops (#168964)

[via llvm-commits]

Author: hjagasiaAMD
Date: 2025-12-02T13:41:16-05:00
New Revision: 2183846a15a04791cf7d85ca5d61d4c89505d3ab

URL: https://github.com/llvm/llvm-project/commit/2183846a15a04791cf7d85ca5d61d4c89505d3ab
DIFF: https://github.com/llvm/llvm-project/commit/2183846a15a04791cf7d85ca5d61d4c89505d3ab.diff

LOG: [AMDGPU] Fix AGPR_32 reg assign for mfma scale ops (#168964)

In MFMA rewrite pass, prevent AGPR_32 reg class assignment for scale
operands, not permitted by instruction format.

---------

Co-authored-by: Matt Arsenault <arsenm2 at gmail.com>

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/AMDGPURewriteAGPRCopyMFMA.cpp
    llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-scale-to-agpr.mir

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/AMDGPURewriteAGPRCopyMFMA.cpp b/llvm/lib/Target/AMDGPU/AMDGPURewriteAGPRCopyMFMA.cpp
index 095f23ad22d97..ffbb1c183ca9e 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURewriteAGPRCopyMFMA.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURewriteAGPRCopyMFMA.cpp
@@ -97,8 +97,8 @@ class AMDGPURewriteAGPRCopyMFMAImpl {
 
   /// Compute the register class constraints based on the uses of \p Reg,
   /// excluding MFMA uses from which can be rewritten to change the register
-  /// class constraint. This should be nearly identical to
-  /// MachineRegisterInfo::recomputeRegClass.
+  /// class constraint. MFMA scale operands need to be constraint checked.
+  /// This should be nearly identical to MachineRegisterInfo::recomputeRegClass.
 
   /// \p RewriteCandidates will collect the set of MFMA instructions that need
   /// to have the opcode mutated to perform the replacement.
@@ -152,9 +152,16 @@ bool AMDGPURewriteAGPRCopyMFMAImpl::recomputeRegClassExceptRewritable(
 
       // We can swap the classes of dst + src2 as a pair to AGPR, so ignore the
       // effects of rewrite candidates. It just so happens that we can use
-      // either AGPR or VGPR in src0/src1, so don't bother checking the
-      // constraint effects of the individual operands.
+      // either AGPR or VGPR in src0/src1. We still need to check constraint
+      // effects for scale variant, which does not allow AGPR.
       if (isRewriteCandidate(*MI)) {
+        int AGPROp = AMDGPU::getMFMASrcCVDstAGPROp(MI->getOpcode());
+        const MCInstrDesc &AGPRDesc = TII.get(AGPROp);
+        const TargetRegisterClass *NewRC =
+            TII.getRegClass(AGPRDesc, MO.getOperandNo());
+        if (!TRI.hasAGPRs(NewRC))
+          return false;
+
         const MachineOperand *VDst =
             TII.getNamedOperand(*MI, AMDGPU::OpName::vdst);
         const MachineOperand *Src2 =

diff  --git a/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-scale-to-agpr.mir b/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-scale-to-agpr.mir
index 999ea42910d92..e35927e8bf00d 100644
--- a/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-scale-to-agpr.mir
+++ b/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-scale-to-agpr.mir
@@ -1,7 +1,9 @@
-# RUN: not --crash llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx950 -run-pass=greedy,amdgpu-rewrite-agpr-copy-mfma -verify-machineinstrs -o - %s 2>&1 | FileCheck %s
-# CHECK: Illegal virtual register for instruction
-# CHECK: Expected a VGPR_32 register, but got a AGPR_32 register
- 
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx950 -run-pass=greedy,amdgpu-rewrite-agpr-copy-mfma -verify-machineinstrs -o - %s | FileCheck %s
+# CHECK: bb.1:
+# CHECK: dead %{{[0-9]+}}:vreg_128_align2 = V_MFMA_SCALE_F32_16X16X128_F8F6F4_f4_f4_vgprcd_e64 %{{[0-9]+}}, %{{[0-9]+}}, %{{[0-9]+}}, 4, 4, %{{[0-9]+}}, %[[REG:[0-9]+]], 4, 0, implicit $mode, implicit $exec
+# CHECK: %{{[0-9]+}}:agpr_32 = IMPLICIT_DEF
+# CHECK: %[[REG]]:vgpr_32 = COPY %{{[0-9]+}}
+
 # Test for issue in amdgpu-rewrite-agpr-copy-mfma, which reassigns scale operand
 # in vgpr_32 register to agpr_32, not permitted by instruction format.
 ---