[RISCV] Efficiently lower (select %cond, andn (f, x), f) using zicond #147369
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
|
@llvm/pr-subscribers-backend-risc-v Author: Ryan Buchner (bababuck) ChangesThe following case is now optimized: Full diff: https://github.com/llvm/llvm-project/pull/147369.diff 2 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index dfb003a9cb1c1..80b5db51f20a1 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -9064,18 +9064,51 @@ SDValue RISCVTargetLowering::lowerSELECT(SDValue Op, SelectionDAG &DAG) const {
if (isNullConstant(TrueV))
return DAG.getNode(RISCVISD::CZERO_NEZ, DL, VT, FalseV, CondV);
+ // Check to see if a given operation is a 'NOT', if so return the negated
+ // operand
+ auto getNotOperand = [](const SDValue &Op) -> std::optional<const SDValue> {
+ using namespace llvm::SDPatternMatch;
+ SDValue Xor;
+ if (sd_match(Op, m_OneUse(m_Not(m_Value(Xor))))) {
+ return std::optional<const SDValue>{Xor};
+ }
+ return std::nullopt;
+ };
// (select c, (and f, x), f) -> (or (and f, x), (czero_nez f, c))
+ // (select c, (and f, ~x), f) -> (andn f, (czero_eqz x, c))
if (TrueV.getOpcode() == ISD::AND &&
- (TrueV.getOperand(0) == FalseV || TrueV.getOperand(1) == FalseV))
+ (TrueV.getOperand(0) == FalseV || TrueV.getOperand(1) == FalseV)) {
+ auto NotOperand = (TrueV.getOperand(0) == FalseV)
+ ? getNotOperand(TrueV.getOperand(1))
+ : getNotOperand(TrueV.getOperand(0));
+ if (NotOperand) {
+ SDValue CMOV =
+ DAG.getNode(RISCVISD::CZERO_EQZ, DL, VT, *NotOperand, CondV);
+ SDValue NOT = DAG.getNOT(DL, CMOV, VT);
+ return DAG.getNode(ISD::AND, DL, VT, FalseV, NOT);
+ }
return DAG.getNode(
ISD::OR, DL, VT, TrueV,
DAG.getNode(RISCVISD::CZERO_NEZ, DL, VT, FalseV, CondV));
+ }
+
// (select c, t, (and t, x)) -> (or (czero_eqz t, c), (and t, x))
+ // (select c, t, (and t, ~x)) -> (andn t, (czero_nez x, c))
if (FalseV.getOpcode() == ISD::AND &&
- (FalseV.getOperand(0) == TrueV || FalseV.getOperand(1) == TrueV))
+ (FalseV.getOperand(0) == TrueV || FalseV.getOperand(1) == TrueV)) {
+ auto NotOperand = (FalseV.getOperand(0) == TrueV)
+ ? getNotOperand(FalseV.getOperand(1))
+ : getNotOperand(FalseV.getOperand(0));
+ if (NotOperand) {
+ SDValue CMOV =
+ DAG.getNode(RISCVISD::CZERO_NEZ, DL, VT, *NotOperand, CondV);
+ SDValue NOT = DAG.getNOT(DL, CMOV, VT);
+ return DAG.getNode(ISD::AND, DL, VT, TrueV, NOT);
+ }
return DAG.getNode(
ISD::OR, DL, VT, FalseV,
DAG.getNode(RISCVISD::CZERO_EQZ, DL, VT, TrueV, CondV));
+ }
// Try some other optimizations before falling back to generic lowering.
if (SDValue V = combineSelectToBinOp(Op.getNode(), DAG, Subtarget))
diff --git a/llvm/test/CodeGen/RISCV/zicond-opts.ll b/llvm/test/CodeGen/RISCV/zicond-opts.ll
index 35b06c4f4fb41..a16145d15db81 100644
--- a/llvm/test/CodeGen/RISCV/zicond-opts.ll
+++ b/llvm/test/CodeGen/RISCV/zicond-opts.ll
@@ -233,9 +233,8 @@ define i64 @test_inv_and_nez(i64 %f, i64 %x, i1 %cond) {
; RV64ZICOND-LABEL: test_inv_and_nez:
; RV64ZICOND: # %bb.0:
; RV64ZICOND-NEXT: andi a2, a2, 1
-; RV64ZICOND-NEXT: andn a1, a0, a1
-; RV64ZICOND-NEXT: czero.nez a0, a0, a2
-; RV64ZICOND-NEXT: or a0, a1, a0
+; RV64ZICOND-NEXT: czero.eqz a1, a1, a2
+; RV64ZICOND-NEXT: andn a0, a0, a1
; RV64ZICOND-NEXT: ret
%5 = xor i64 %x, -1
%6 = select i1 %cond, i64 %5, i64 -1
@@ -258,9 +257,8 @@ define i64 @test_inv_and_eqz(i64 %f, i64 %x, i1 %cond) {
; RV64ZICOND-LABEL: test_inv_and_eqz:
; RV64ZICOND: # %bb.0:
; RV64ZICOND-NEXT: andi a2, a2, 1
-; RV64ZICOND-NEXT: andn a1, a0, a1
-; RV64ZICOND-NEXT: czero.eqz a0, a0, a2
-; RV64ZICOND-NEXT: or a0, a1, a0
+; RV64ZICOND-NEXT: czero.nez a1, a1, a2
+; RV64ZICOND-NEXT: andn a0, a0, a1
; RV64ZICOND-NEXT: ret
%5 = xor i64 %x, -1
%6 = select i1 %cond, i64 -1, i64 %5
|
topperc
reviewed
Jul 7, 2025
… extension The following case is now optimized: (select c, (and f, ~x), f) -> (andn f, (czero_eqz x, c))
bababuck
force-pushed
the
bababuck/ZicondOptAndn
branch
from
a9d6c50 to
f5031ab
Compare
Collaborator
|
Don't force push unless necessary. Use fixup commits. The final push will be done with squash so any messy history in the PR won't matter. |
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/55/builds/13976 Here is the relevant piece of the build log for the reference |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The following case is now optimized:
(select c, (and f, ~x), f) -> (andn f, (czero_eqz x, c))