Skip to content

[RISCV] Allow constants in tryFoldSelectIntoOp #157376

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
preames merged 1 commit into from
Sep 8, 2025
Merged

[RISCV] Allow constants in tryFoldSelectIntoOp #157376

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion llvm/lib/Target/RISCV/RISCVISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18838,7 +18838,7 @@ static SDValue tryFoldSelectIntoOp(SDNode *N, SelectionDAG &DAG,
break;
}

if (!TrueVal.hasOneUse() || isa<ConstantSDNode>(FalseVal))
if (!TrueVal.hasOneUse())
Copy link
Collaborator Author

@preames preames Sep 8, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@topperc Is there a non-obvious reason for this restriction? It looks like you added this in the original commit that added this transform.

Copy link
Collaborator

@topperc topperc Sep 8, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't recall. I might have done it because InstCombine doesn't do it for constants. I think InstCombine's motivation for the transform is to reduce the number of uses of Y, but that doesn't matter for constants.

return SDValue();

unsigned OpToFold;
Expand Down
175 changes: 63 additions & 112 deletions llvm/test/CodeGen/RISCV/select.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -290,38 +290,28 @@ entry:
define i32 @select_xor_5(i1 zeroext %cond, i32 %x) {
; RV32IM-LABEL: select_xor_5:
; RV32IM: # %bb.0:
; RV32IM-NEXT: bnez a0, .LBB8_2
; RV32IM-NEXT: # %bb.1:
; RV32IM-NEXT: xori a0, a1, 128
; RV32IM-NEXT: ret
; RV32IM-NEXT: .LBB8_2:
; RV32IM-NEXT: li a0, 128
; RV32IM-NEXT: addi a0, a0, -1
; RV32IM-NEXT: and a0, a0, a1
; RV32IM-NEXT: xori a0, a0, 128
; RV32IM-NEXT: ret
;
; RV64IM-LABEL: select_xor_5:
; RV64IM: # %bb.0:
; RV64IM-NEXT: bnez a0, .LBB8_2
; RV64IM-NEXT: # %bb.1:
; RV64IM-NEXT: xori a0, a1, 128
; RV64IM-NEXT: ret
; RV64IM-NEXT: .LBB8_2:
; RV64IM-NEXT: li a0, 128
; RV64IM-NEXT: addi a0, a0, -1
; RV64IM-NEXT: and a0, a0, a1
; RV64IM-NEXT: xori a0, a0, 128
; RV64IM-NEXT: ret
;
; RV64IMXVTCONDOPS-LABEL: select_xor_5:
; RV64IMXVTCONDOPS: # %bb.0:
; RV64IMXVTCONDOPS-NEXT: xori a1, a1, 128
; RV64IMXVTCONDOPS-NEXT: addi a1, a1, -128
; RV64IMXVTCONDOPS-NEXT: vt.maskcn a0, a1, a0
; RV64IMXVTCONDOPS-NEXT: addi a0, a0, 128
; RV64IMXVTCONDOPS-NEXT: xori a0, a0, 128
; RV64IMXVTCONDOPS-NEXT: ret
;
; CHECKZICOND-LABEL: select_xor_5:
; CHECKZICOND: # %bb.0:
; CHECKZICOND-NEXT: xori a1, a1, 128
; CHECKZICOND-NEXT: addi a1, a1, -128
; CHECKZICOND-NEXT: czero.nez a0, a1, a0
; CHECKZICOND-NEXT: addi a0, a0, 128
; CHECKZICOND-NEXT: xori a0, a0, 128
; CHECKZICOND-NEXT: ret
%add = xor i32 %x, 128
%sel = select i1 %cond, i32 128, i32 %add
Expand Down Expand Up @@ -643,38 +633,28 @@ entry:
define i32 @select_or_4(i1 zeroext %cond, i32 %x) {
; RV32IM-LABEL: select_or_4:
; RV32IM: # %bb.0:
; RV32IM-NEXT: bnez a0, .LBB17_2
; RV32IM-NEXT: # %bb.1:
; RV32IM-NEXT: ori a0, a1, 128
; RV32IM-NEXT: ret
; RV32IM-NEXT: .LBB17_2:
; RV32IM-NEXT: li a0, 128
; RV32IM-NEXT: addi a0, a0, -1
; RV32IM-NEXT: and a0, a0, a1
; RV32IM-NEXT: ori a0, a0, 128
; RV32IM-NEXT: ret
;
; RV64IM-LABEL: select_or_4:
; RV64IM: # %bb.0:
; RV64IM-NEXT: bnez a0, .LBB17_2
; RV64IM-NEXT: # %bb.1:
; RV64IM-NEXT: ori a0, a1, 128
; RV64IM-NEXT: ret
; RV64IM-NEXT: .LBB17_2:
; RV64IM-NEXT: li a0, 128
; RV64IM-NEXT: addi a0, a0, -1
; RV64IM-NEXT: and a0, a0, a1
; RV64IM-NEXT: ori a0, a0, 128
; RV64IM-NEXT: ret
;
; RV64IMXVTCONDOPS-LABEL: select_or_4:
; RV64IMXVTCONDOPS: # %bb.0:
; RV64IMXVTCONDOPS-NEXT: ori a1, a1, 128
; RV64IMXVTCONDOPS-NEXT: addi a1, a1, -128
; RV64IMXVTCONDOPS-NEXT: vt.maskcn a0, a1, a0
; RV64IMXVTCONDOPS-NEXT: addi a0, a0, 128
; RV64IMXVTCONDOPS-NEXT: ori a0, a0, 128
; RV64IMXVTCONDOPS-NEXT: ret
;
; CHECKZICOND-LABEL: select_or_4:
; CHECKZICOND: # %bb.0:
; CHECKZICOND-NEXT: ori a1, a1, 128
; CHECKZICOND-NEXT: addi a1, a1, -128
; CHECKZICOND-NEXT: czero.nez a0, a1, a0
; CHECKZICOND-NEXT: addi a0, a0, 128
; CHECKZICOND-NEXT: ori a0, a0, 128
; CHECKZICOND-NEXT: ret
%add = or i32 %x, 128
%sel = select i1 %cond, i32 128, i32 %add
Expand Down Expand Up @@ -801,30 +781,22 @@ entry:
define i32 @select_add_4(i1 zeroext %cond, i32 %x) {
; RV32IM-LABEL: select_add_4:
; RV32IM: # %bb.0:
; RV32IM-NEXT: bnez a0, .LBB21_2
; RV32IM-NEXT: # %bb.1:
; RV32IM-NEXT: addi a0, a1, 128
; RV32IM-NEXT: ret
; RV32IM-NEXT: .LBB21_2:
; RV32IM-NEXT: li a0, 128
; RV32IM-NEXT: addi a0, a0, -1
; RV32IM-NEXT: and a0, a0, a1
; RV32IM-NEXT: addi a0, a0, 128
; RV32IM-NEXT: ret
;
; RV64IM-LABEL: select_add_4:
; RV64IM: # %bb.0:
; RV64IM-NEXT: bnez a0, .LBB21_2
; RV64IM-NEXT: # %bb.1:
; RV64IM-NEXT: addiw a0, a1, 128
; RV64IM-NEXT: ret
; RV64IM-NEXT: .LBB21_2:
; RV64IM-NEXT: li a0, 128
; RV64IM-NEXT: addi a0, a0, -1
; RV64IM-NEXT: and a0, a0, a1
; RV64IM-NEXT: addiw a0, a0, 128
; RV64IM-NEXT: ret
;
; RV64IMXVTCONDOPS-LABEL: select_add_4:
; RV64IMXVTCONDOPS: # %bb.0:
; RV64IMXVTCONDOPS-NEXT: addiw a1, a1, 128
; RV64IMXVTCONDOPS-NEXT: addi a1, a1, -128
; RV64IMXVTCONDOPS-NEXT: vt.maskcn a0, a1, a0
; RV64IMXVTCONDOPS-NEXT: addi a0, a0, 128
; RV64IMXVTCONDOPS-NEXT: addiw a0, a0, 128
; RV64IMXVTCONDOPS-NEXT: ret
;
; RV32IMZICOND-LABEL: select_add_4:
Expand All @@ -835,10 +807,8 @@ define i32 @select_add_4(i1 zeroext %cond, i32 %x) {
;
; RV64IMZICOND-LABEL: select_add_4:
; RV64IMZICOND: # %bb.0:
; RV64IMZICOND-NEXT: addiw a1, a1, 128
; RV64IMZICOND-NEXT: addi a1, a1, -128
; RV64IMZICOND-NEXT: czero.nez a0, a1, a0
; RV64IMZICOND-NEXT: addi a0, a0, 128
; RV64IMZICOND-NEXT: addiw a0, a0, 128
; RV64IMZICOND-NEXT: ret
%add = add i32 %x, 128
%sel = select i1 %cond, i32 128, i32 %add
Expand All @@ -848,26 +818,19 @@ define i32 @select_add_4(i1 zeroext %cond, i32 %x) {
define i64 @select_add_5(i1 zeroext %cond, i64 %x) {
; RV32IM-LABEL: select_add_5:
; RV32IM: # %bb.0:
; RV32IM-NEXT: addi a3, a1, 128
; RV32IM-NEXT: sltu a1, a3, a1
; RV32IM-NEXT: add a2, a2, a1
; RV32IM-NEXT: beqz a0, .LBB22_2
; RV32IM-NEXT: # %bb.1:
; RV32IM-NEXT: li a3, 128
; RV32IM-NEXT: .LBB22_2:
; RV32IM-NEXT: addi a0, a0, -1
; RV32IM-NEXT: and a1, a0, a2
; RV32IM-NEXT: mv a0, a3
; RV32IM-NEXT: addi a3, a0, -1
; RV32IM-NEXT: and a1, a3, a1
; RV32IM-NEXT: addi a0, a1, 128
; RV32IM-NEXT: sltu a1, a0, a1
; RV32IM-NEXT: and a2, a3, a2
; RV32IM-NEXT: add a1, a2, a1
; RV32IM-NEXT: ret
;
; RV64IM-LABEL: select_add_5:
; RV64IM: # %bb.0:
; RV64IM-NEXT: bnez a0, .LBB22_2
; RV64IM-NEXT: # %bb.1:
; RV64IM-NEXT: addi a0, a1, 128
; RV64IM-NEXT: ret
; RV64IM-NEXT: .LBB22_2:
; RV64IM-NEXT: li a0, 128
; RV64IM-NEXT: addi a0, a0, -1
; RV64IM-NEXT: and a0, a0, a1
; RV64IM-NEXT: addi a0, a0, 128
; RV64IM-NEXT: ret
;
; RV64IMXVTCONDOPS-LABEL: select_add_5:
Expand All @@ -878,12 +841,12 @@ define i64 @select_add_5(i1 zeroext %cond, i64 %x) {
;
; RV32IMZICOND-LABEL: select_add_5:
; RV32IMZICOND: # %bb.0:
; RV32IMZICOND-NEXT: czero.nez a1, a1, a0
; RV32IMZICOND-NEXT: addi a3, a1, 128
; RV32IMZICOND-NEXT: sltu a3, a3, a1
; RV32IMZICOND-NEXT: czero.nez a4, a1, a0
; RV32IMZICOND-NEXT: add a2, a2, a3
; RV32IMZICOND-NEXT: czero.nez a1, a2, a0
; RV32IMZICOND-NEXT: addi a0, a4, 128
; RV32IMZICOND-NEXT: sltu a1, a3, a1
; RV32IMZICOND-NEXT: czero.nez a0, a2, a0
; RV32IMZICOND-NEXT: add a1, a0, a1
; RV32IMZICOND-NEXT: mv a0, a3
Copy link
Collaborator Author

@preames preames Sep 8, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A quick comment on the RV64IMZICOND case just below. This one isn't changing because we have a case for lowering constant + nonconstant in zicond which ends lowering as (select c, (nonconstant-constant), 0) + constant. When nonconstant = ADD X, constant, this simplifies away to exactly the form tryFoldSelectIntoOp produces. Note that this only occurs when the type is legal, otherwise you end up with an extend blocking the special case.

(Let's not talk about how long it took me to figure this out, and that the extend was really a red-herring.)

; RV32IMZICOND-NEXT: ret
;
; RV64IMZICOND-LABEL: select_add_5:
Expand All @@ -899,63 +862,51 @@ define i64 @select_add_5(i1 zeroext %cond, i64 %x) {
define i64 @select_add_6(i1 zeroext %cond, i64 %x) {
; RV32IM-LABEL: select_add_6:
; RV32IM: # %bb.0:
; RV32IM-NEXT: lui a3, 14
; RV32IM-NEXT: addi a3, a3, 1005
; RV32IM-NEXT: add a4, a1, a3
; RV32IM-NEXT: sltu a1, a4, a1
; RV32IM-NEXT: add a2, a2, a1
; RV32IM-NEXT: bnez a0, .LBB23_2
; RV32IM-NEXT: # %bb.1:
; RV32IM-NEXT: mv a3, a4
; RV32IM-NEXT: .LBB23_2:
; RV32IM-NEXT: addi a0, a0, -1
; RV32IM-NEXT: and a1, a0, a2
; RV32IM-NEXT: mv a0, a3
; RV32IM-NEXT: addi a3, a0, -1
; RV32IM-NEXT: lui a0, 14
; RV32IM-NEXT: and a1, a3, a1
; RV32IM-NEXT: addi a0, a0, 1005
; RV32IM-NEXT: add a0, a1, a0
; RV32IM-NEXT: sltu a1, a0, a1
; RV32IM-NEXT: and a2, a3, a2
; RV32IM-NEXT: add a1, a2, a1
; RV32IM-NEXT: ret
;
; RV64IM-LABEL: select_add_6:
; RV64IM: # %bb.0:
; RV64IM-NEXT: mv a2, a0
; RV64IM-NEXT: lui a0, 14
; RV64IM-NEXT: addi a0, a0, 1005
; RV64IM-NEXT: bnez a2, .LBB23_2
; RV64IM-NEXT: # %bb.1:
; RV64IM-NEXT: add a0, a1, a0
; RV64IM-NEXT: .LBB23_2:
; RV64IM-NEXT: addi a0, a0, -1
; RV64IM-NEXT: and a0, a0, a1
; RV64IM-NEXT: lui a1, 14
; RV64IM-NEXT: addi a1, a1, 1005
; RV64IM-NEXT: add a0, a0, a1
; RV64IM-NEXT: ret
;
; RV64IMXVTCONDOPS-LABEL: select_add_6:
; RV64IMXVTCONDOPS: # %bb.0:
; RV64IMXVTCONDOPS-NEXT: lui a2, 14
; RV64IMXVTCONDOPS-NEXT: addi a2, a2, 1005
; RV64IMXVTCONDOPS-NEXT: add a1, a1, a2
; RV64IMXVTCONDOPS-NEXT: vt.maskc a2, a2, a0
; RV64IMXVTCONDOPS-NEXT: vt.maskcn a0, a1, a0
; RV64IMXVTCONDOPS-NEXT: or a0, a2, a0
; RV64IMXVTCONDOPS-NEXT: lui a1, 14
; RV64IMXVTCONDOPS-NEXT: addi a1, a1, 1005
; RV64IMXVTCONDOPS-NEXT: add a0, a0, a1
; RV64IMXVTCONDOPS-NEXT: ret
;
; RV32IMZICOND-LABEL: select_add_6:
; RV32IMZICOND: # %bb.0:
; RV32IMZICOND-NEXT: czero.nez a1, a1, a0
; RV32IMZICOND-NEXT: lui a3, 14
; RV32IMZICOND-NEXT: addi a3, a3, 1005
; RV32IMZICOND-NEXT: add a4, a1, a3
; RV32IMZICOND-NEXT: czero.eqz a3, a3, a0
; RV32IMZICOND-NEXT: sltu a1, a4, a1
; RV32IMZICOND-NEXT: czero.nez a4, a4, a0
; RV32IMZICOND-NEXT: add a1, a2, a1
; RV32IMZICOND-NEXT: or a2, a3, a4
; RV32IMZICOND-NEXT: czero.nez a1, a1, a0
; RV32IMZICOND-NEXT: mv a0, a2
; RV32IMZICOND-NEXT: add a3, a1, a3
; RV32IMZICOND-NEXT: sltu a1, a3, a1
; RV32IMZICOND-NEXT: czero.nez a0, a2, a0
; RV32IMZICOND-NEXT: add a1, a0, a1
; RV32IMZICOND-NEXT: mv a0, a3
; RV32IMZICOND-NEXT: ret
;
; RV64IMZICOND-LABEL: select_add_6:
; RV64IMZICOND: # %bb.0:
; RV64IMZICOND-NEXT: lui a2, 14
; RV64IMZICOND-NEXT: addi a2, a2, 1005
; RV64IMZICOND-NEXT: add a1, a1, a2
; RV64IMZICOND-NEXT: czero.eqz a2, a2, a0
; RV64IMZICOND-NEXT: czero.nez a0, a1, a0
; RV64IMZICOND-NEXT: or a0, a2, a0
; RV64IMZICOND-NEXT: lui a1, 14
; RV64IMZICOND-NEXT: addi a1, a1, 1005
; RV64IMZICOND-NEXT: add a0, a0, a1
; RV64IMZICOND-NEXT: ret
%add = add i64 %x, 58349
%sel = select i1 %cond, i64 58349, i64 %add
Expand Down
20 changes: 9 additions & 11 deletions llvm/test/CodeGen/RISCV/zicond-opts.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -223,11 +223,10 @@ define i64 @test_inv_and_nez(i64 %f, i64 %x, i1 %cond) {
; RV32ZICOND-LABEL: test_inv_and_nez:
; RV32ZICOND: # %bb.0:
; RV32ZICOND-NEXT: andi a4, a4, 1
; RV32ZICOND-NEXT: addi a4, a4, -1
; RV32ZICOND-NEXT: orn a3, a4, a3
; RV32ZICOND-NEXT: orn a2, a4, a2
; RV32ZICOND-NEXT: and a0, a2, a0
; RV32ZICOND-NEXT: and a1, a3, a1
; RV32ZICOND-NEXT: czero.eqz a3, a3, a4
; RV32ZICOND-NEXT: czero.eqz a2, a2, a4
; RV32ZICOND-NEXT: andn a0, a0, a2
; RV32ZICOND-NEXT: andn a1, a1, a3
; RV32ZICOND-NEXT: ret
;
; RV64ZICOND-LABEL: test_inv_and_nez:
Expand All @@ -246,12 +245,11 @@ define i64 @test_inv_and_nez(i64 %f, i64 %x, i1 %cond) {
define i64 @test_inv_and_eqz(i64 %f, i64 %x, i1 %cond) {
; RV32ZICOND-LABEL: test_inv_and_eqz:
; RV32ZICOND: # %bb.0:
; RV32ZICOND-NEXT: slli a4, a4, 31
; RV32ZICOND-NEXT: srai a4, a4, 31
; RV32ZICOND-NEXT: orn a3, a4, a3
; RV32ZICOND-NEXT: orn a2, a4, a2
; RV32ZICOND-NEXT: and a0, a2, a0
; RV32ZICOND-NEXT: and a1, a3, a1
; RV32ZICOND-NEXT: andi a4, a4, 1
; RV32ZICOND-NEXT: czero.nez a3, a3, a4
; RV32ZICOND-NEXT: czero.nez a2, a2, a4
; RV32ZICOND-NEXT: andn a0, a0, a2
; RV32ZICOND-NEXT: andn a1, a1, a3
; RV32ZICOND-NEXT: ret
;
; RV64ZICOND-LABEL: test_inv_and_eqz:
Expand Down