[llvm] 9b7bf1f - [RISCV] Set the SEW field in the VPseudoTiedBinaryRoundingMode.

[Craig Topper via llvm-commits]

Author: Craig Topper
Date: 2025-02-22T14:36:50-08:00
New Revision: 9b7bf1fdd3e7727cdde7ca5bb2f95a8f4f26471c

URL: https://github.com/llvm/llvm-project/commit/9b7bf1fdd3e7727cdde7ca5bb2f95a8f4f26471c
DIFF: https://github.com/llvm/llvm-project/commit/9b7bf1fdd3e7727cdde7ca5bb2f95a8f4f26471c.diff

LOG: [RISCV] Set the SEW field in the VPseudoTiedBinaryRoundingMode.

Noticed some oddities in the InversePseudo table. TIED pseudos are
in there but sorted differently than their non-TIED version. I think
we always check non-zero SEW first which should hit the non-tied version.

I'm working on a larger patch to remove _MASK and _TIED from the
InversePseudo table.

Added: 
    

Modified: 
    llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
index 7407d64aa92a9..1795bf73d0a22 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
@@ -2168,7 +2168,7 @@ multiclass VPseudoTiedBinaryRoundingMode<VReg RetClass,
                                          int sew = 0,
                                          bits<2> TargetConstraintType = 1> {
     defvar suffix = !if(sew, "_" # MInfo.MX # "_E" # sew, "_" # MInfo.MX);
-    let VLMul = MInfo.value in {
+    let VLMul = MInfo.value, SEW=sew in {
     def suffix # "_TIED":
       VPseudoTiedBinaryNoMaskRoundingMode<RetClass, Op2Class, Constraint, TargetConstraintType>;
     def suffix # "_MASK_TIED" :