[llvm] [RISCV][WIP] Correct getOperandLog2EEW for VCOMPRESS_VM. (PR #156719)

[Craig Topper via llvm-commits]

https://github.com/topperc updated https://github.com/llvm/llvm-project/pull/156719

>From 7fea6be51294f0f7b0f967d20c1eb32676f1c5b9 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper at sifive.com>
Date: Wed, 3 Sep 2025 11:13:21 -0700
Subject: [PATCH 1/2] Pre-commit tests

---
 llvm/test/CodeGen/RISCV/rvv/vl-opt.ll | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/llvm/test/CodeGen/RISCV/rvv/vl-opt.ll b/llvm/test/CodeGen/RISCV/rvv/vl-opt.ll
index cd282c265ae47..1b3219a5f1d74 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vl-opt.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vl-opt.ll
@@ -198,3 +198,28 @@ define void @fadd_fcmp_select_copy(<vscale x 4 x float> %v, <vscale x 4 x i1> %c
   call void @llvm.riscv.vsm(<vscale x 4 x i1> %select, ptr %p, iXLen %vl)
   ret void
 }
+
+define <vscale x 8 x i32> @vcompress_cmp(<vscale x 8 x i32> %a, <vscale x 8 x i32> %b, <vscale x 8 x i32> %c, iXLen %vl) {
+; CHECK-LABEL: vcompress_cmp:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli a1, zero, e32, m4, ta, ma
+; CHECK-NEXT:    vmseq.vv v20, v8, v12
+; CHECK-NEXT:    vsetvli zero, a0, e32, m4, ta, ma
+; CHECK-NEXT:    vcompress.vm v8, v16, v20
+; CHECK-NEXT:    ret
+  %cmp = icmp eq <vscale x 8 x i32> %a, %b
+  %compress = call <vscale x 8 x i32> @llvm.riscv.vcompress.nxv8i32(<vscale x 8 x i32> poison, <vscale x 8 x i32> %c, <vscale x 8 x i1> %cmp, iXLen %vl)
+  ret <vscale x 8 x i32> %compress
+}
+
+define <vscale x 8 x i32> @vcompress_add(<vscale x 8 x i32> %a, <vscale x 8 x i32> %b, <vscale x 8 x i1> %c, iXLen %vl) {
+; CHECK-LABEL: vcompress_add:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e32, m4, ta, ma
+; CHECK-NEXT:    vadd.vv v12, v8, v12
+; CHECK-NEXT:    vcompress.vm v8, v12, v0
+; CHECK-NEXT:    ret
+  %add = add <vscale x 8 x i32> %a, %b
+  %compress = call <vscale x 8 x i32> @llvm.riscv.vcompress.nxv8i32(<vscale x 8 x i32> poison, <vscale x 8 x i32> %add, <vscale x 8 x i1> %c, iXLen %vl)
+  ret <vscale x 8 x i32> %compress
+}

>From 1d9bea3feb0b2ea111e17316d1a41d140105250a Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper at sifive.com>
Date: Wed, 3 Sep 2025 10:14:30 -0700
Subject: [PATCH 2/2] [RISCV] Correct getOperandLog2EEW for VCOMPRESS_VM.

---
 llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp | 9 ++++++---
 llvm/test/CodeGen/RISCV/rvv/vl-opt.ll      | 3 +--
 2 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
index 9b70eb6c25b12..4dcc6c181abb9 100644
--- a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
+++ b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
@@ -422,9 +422,6 @@ getOperandLog2EEW(const MachineOperand &MO, const MachineRegisterInfo *MRI) {
   case RISCV::VRGATHER_VI:
   case RISCV::VRGATHER_VV:
   case RISCV::VRGATHER_VX:
-  // Vector Compress Instruction
-  // EEW=SEW.
-  case RISCV::VCOMPRESS_VM:
   // Vector Element Index Instruction
   case RISCV::VID_V:
   // Vector Single-Width Floating-Point Add/Subtract Instructions
@@ -674,6 +671,12 @@ getOperandLog2EEW(const MachineOperand &MO, const MachineRegisterInfo *MRI) {
     return MILog2SEW;
   }
 
+  // Vector Compress Instruction
+  // EEW=SEW except the mask operand has EEW=1. Mask operand is not handled
+  // before this switch.
+  case RISCV::VCOMPRESS_VM:
+    return MO.getOperandNo() == 3 ? 0 : MILog2SEW;
+
   // Vector Iota Instruction
   // EEW=SEW, except the mask operand has EEW=1. Mask operand is not handled
   // before this switch.
diff --git a/llvm/test/CodeGen/RISCV/rvv/vl-opt.ll b/llvm/test/CodeGen/RISCV/rvv/vl-opt.ll
index 1b3219a5f1d74..6b5b984a48789 100644
--- a/llvm/test/CodeGen/RISCV/rvv/vl-opt.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/vl-opt.ll
@@ -202,9 +202,8 @@ define void @fadd_fcmp_select_copy(<vscale x 4 x float> %v, <vscale x 4 x i1> %c
 define <vscale x 8 x i32> @vcompress_cmp(<vscale x 8 x i32> %a, <vscale x 8 x i32> %b, <vscale x 8 x i32> %c, iXLen %vl) {
 ; CHECK-LABEL: vcompress_cmp:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vsetvli a1, zero, e32, m4, ta, ma
-; CHECK-NEXT:    vmseq.vv v20, v8, v12
 ; CHECK-NEXT:    vsetvli zero, a0, e32, m4, ta, ma
+; CHECK-NEXT:    vmseq.vv v20, v8, v12
 ; CHECK-NEXT:    vcompress.vm v8, v16, v20
 ; CHECK-NEXT:    ret
   %cmp = icmp eq <vscale x 8 x i32> %a, %b