[llvm] [RISCV][CostModel] Remove inst cost of cmp inst in cmp-select sequence. (PR #91158)
via llvm-commits
llvm-commits at lists.llvm.org
Sun May 5 18:38:54 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-risc-v
Author: Elvis Wang (ElvisWang123)
<details>
<summary>Changes</summary>
Cmp and select instructinos will lower to SELECT_CC and lower to PseudoCCMOVGPR which will generate a conditional branch inst and a move inst. Remove the instruction cost of the cmp instruction because of the generated branch instruction has 0 cost in the cost model.
---
Full diff: https://github.com/llvm/llvm-project/pull/91158.diff
3 Files Affected:
- (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp (+8)
- (added) llvm/test/Analysis/CostModel/RISCV/cmp-select.ll (+14)
- (modified) llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll (+1-1)
``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index ce26e61880fd05..885697e48b59ad 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -1436,6 +1436,14 @@ InstructionCost RISCVTTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
}
}
+ // The cmp + select instructions will lower to SELECT_CC and lower to
+ // PseudoCCMOVGPR which will generate a conditional branch + mv. Remove
+ // the cost of the cmp instruction because the estimated cost of
+ // branch instruction is 0.
+ if (I && I->hasOneUser() && isa<SelectInst>(I->user_back()) &&
+ I->user_back()->getOperand(0) == I)
+ return 0;
+
// TODO: Add cost for scalar type.
return BaseT::getCmpSelInstrCost(Opcode, ValTy, CondTy, VecPred, CostKind, I);
diff --git a/llvm/test/Analysis/CostModel/RISCV/cmp-select.ll b/llvm/test/Analysis/CostModel/RISCV/cmp-select.ll
new file mode 100644
index 00000000000000..6c88467afe6351
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/RISCV/cmp-select.ll
@@ -0,0 +1,14 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -mtriple=riscv64 -passes="print<cost-model>" -cost-kind=throughput 2>&1 -disable-output | FileCheck %s --check-prefixes=CHECK
+
+define void @cmp-select() {
+;
+; CHECK-LABEL: 'cmp-select'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %cmp1 = icmp slt i64 0, 1
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %select1 = select i1 %cmp1, i32 5, i32 4
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %cmp1 = icmp slt i64 0, 1
+ %select1 = select i1 %cmp1, i32 5, i32 4
+ ret void
+}
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll
index 12fdf2149daf47..af62aad84d6be3 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll
@@ -699,7 +699,7 @@ define void @double_stride_ptr_iv(ptr %p, ptr %p2, i64 %stride) {
; STRIDED-NEXT: entry:
; STRIDED-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; STRIDED-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 4
-; STRIDED-NEXT: [[TMP2:%.*]] = call i64 @llvm.umax.i64(i64 32, i64 [[TMP1]])
+; STRIDED-NEXT: [[TMP2:%.*]] = call i64 @llvm.umax.i64(i64 24, i64 [[TMP1]])
; STRIDED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1024, [[TMP2]]
; STRIDED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
; STRIDED: vector.memcheck:
``````````
</details>
https://github.com/llvm/llvm-project/pull/91158
More information about the llvm-commits
mailing list