[llvm] 1687aa2 - [RISCV][VLOPT] Don't reduce the VL is the same as CommonVL (#123878)

[via llvm-commits]

Author: Michael Maitland
Date: 2025-01-22T13:49:54-05:00
New Revision: 1687aa2a996f4059f275c83d5db635d43165d36c

URL: https://github.com/llvm/llvm-project/commit/1687aa2a996f4059f275c83d5db635d43165d36c
DIFF: https://github.com/llvm/llvm-project/commit/1687aa2a996f4059f275c83d5db635d43165d36c.diff

LOG: [RISCV][VLOPT] Don't reduce the VL is the same as CommonVL (#123878)

This fixes the slowdown in #123862.

Added: 
    llvm/test/CodeGen/RISCV/rvv/vlopt-same-vl.ll

Modified: 
    llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
index 66d26bf5b11e2d..fc3300247b1909 100644
--- a/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
+++ b/llvm/lib/Target/RISCV/RISCVVLOptimizer.cpp
@@ -1313,6 +1313,12 @@ bool RISCVVLOptimizer::tryReduceVL(MachineInstr &MI) {
     return false;
   }
 
+  if (CommonVL->isIdenticalTo(VLOp)) {
+    LLVM_DEBUG(
+        dbgs() << "    Abort due to CommonVL == VLOp, no point in reducing.\n");
+    return false;
+  }
+
   if (CommonVL->isImm()) {
     LLVM_DEBUG(dbgs() << "  Reduce VL from " << VLOp << " to "
                       << CommonVL->getImm() << " for " << MI << "\n");

diff  --git a/llvm/test/CodeGen/RISCV/rvv/vlopt-same-vl.ll b/llvm/test/CodeGen/RISCV/rvv/vlopt-same-vl.ll
new file mode 100644
index 00000000000000..65e6eddfb3cd60
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/vlopt-same-vl.ll
@@ -0,0 +1,27 @@
+; RUN: llc -mtriple=riscv64 -mattr=+v -riscv-enable-vl-optimizer \
+; RUN:   -verify-machineinstrs -debug-only=riscv-vl-optimizer -o - 2>&1 %s | FileCheck %s 
+
+; REQUIRES: asserts
+
+; GitHub Issue #123862 provided a case where the riscv-vl-optimizer pass was
+; very slow. It was found that that case benefited greatly from aborting due
+; to CommonVL == VLOp. Adding the case provided in the issue would show up
+; as a long running test instead of a test failure. We would likley have a hard
+; time figuring if that case had a regression. So instead, we check this output
+; which was responsible for speeding it up. 
+
+define <vscale x 4 x i32> @same_vl_imm(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+  ; CHECK: User VL is: 4
+  ; CHECK-NEXT: Abort due to CommonVL == VLOp, no point in reducing.
+  %v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, i64 4)
+  %w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, i64 4)
+  ret <vscale x 4 x i32> %w
+}
+
+define <vscale x 4 x i32> @same_vl_reg(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, i64 %vl) {
+  ; CHECK: User VL is: %3:gprnox0
+  ; CHECK-NEXT: Abort due to CommonVL == VLOp, no point in reducing.
+  %v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, i64 %vl)
+  %w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, i64 %vl)
+  ret <vscale x 4 x i32> %w
+}