Skip to content

[AArch64] Correct SCVTF/UCVTF instructions for vector input #152974

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
merged 3 commits into from
Sep 12, 2025
Merged

[AArch64] Correct SCVTF/UCVTF instructions for vector input #152974

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e82a221
[AArch64] Correct SCVTF/UCVTF instructions for vector inputs
Amichaxx Aug 11, 2025
e3cced3
Add isel patterns for SCVTF/UCVTF
Amichaxx Aug 20, 2025
675d79d
Revision
Amichaxx Sep 9, 2025
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
1 change: 0 additions & 1 deletion llvm/lib/Target/AArch64/AArch64InstrFormats.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5520,7 +5520,6 @@ multiclass IntegerToFPSIMDScalar<bits<2> rmode, bits<3> opcode, string asm, SDPa
let Inst{31} = 1; // 64-bit FPR flag
let Inst{23-22} = 0b00; // 32-bit FPR flag
}

def : Pat<(f16 (node (i32 (extractelt (v4i32 V128:$Rn), (i64 0))))),
(!cast<Instruction>(NAME # HSr) (EXTRACT_SUBREG V128:$Rn, ssub))>;
def : Pat<(f64 (node (i32 (extractelt (v4i32 V128:$Rn), (i64 0))))),
Expand Down
5 changes: 5 additions & 0 deletions llvm/lib/Target/AArch64/AArch64InstrInfo.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5287,6 +5287,11 @@ defm UCVTF : IntegerToFP<0b00, 0b011, "ucvtf", any_uint_to_fp>;
let Predicates = [HasNEON, HasFPRCVT] in {
defm SCVTF : IntegerToFPSIMDScalar<0b11, 0b100, "scvtf", any_sint_to_fp>;
defm UCVTF : IntegerToFPSIMDScalar<0b11, 0b101, "ucvtf", any_uint_to_fp>;

def : Pat<(v1f64 (extract_subvector (v2f64 (sint_to_fp (v2i64 (sext (v2i32 V64:$Rn))))), (i64 0))),
(SCVTFDSr (EXTRACT_SUBREG V64:$Rn, ssub))>;
def : Pat<(v1f64 (extract_subvector (v2f64 (uint_to_fp (v2i64 (zext (v2i32 V64:$Rn))))), (i64 0))),
(UCVTFDSr (EXTRACT_SUBREG V64:$Rn, ssub))>;
}

def : Pat<(f16 (fdiv (f16 (any_sint_to_fp (i32 GPR32:$Rn))), fixedpoint_f16_i32:$scale)),
Expand Down
45 changes: 27 additions & 18 deletions llvm/test/CodeGen/AArch64/fprcvt-cvtf.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,16 +94,10 @@ define double @scvtf_f64i32_neg(<4 x i32> %x) {
ret double %conv
}

; This test does not give the indended result of scvtf d0, s0
; This is due to the input being loaded as a 2 item vector and
; therefore using vector inputs that do not match the pattern
; This test will be fixed in a future revision
define <1 x double> @scvtf_f64i32_simple(<1 x i32> %x) {
; CHECK-LABEL: scvtf_f64i32_simple:
; CHECK: // %bb.0:
; CHECK-NEXT: sshll v0.2d, v0.2s, #0
; CHECK-NEXT: scvtf v0.2d, v0.2d
; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-NEXT: scvtf d0, s0
; CHECK-NEXT: ret
;
; CHECK-NO-FPRCVT-LABEL: scvtf_f64i32_simple:
Expand Down Expand Up @@ -202,10 +196,8 @@ define float @scvtf_f32i64_neg(<2 x i64> %x) {
ret float %conv
}

; This test does not give the indended result of scvtf s0, d0
; This is due to the input being loaded as a 2 item vector and
; therefore using vector inputs that do not match the pattern
; This test will be fixed in a future revision
; <1 x float> is illegal on AArch64 and is widened to <2 x float>.
; This widening introduces the extra insert/extract/zeroing instructions.
define <1 x float> @scvtf_f32i64_simple(<1 x i64> %x) {
; CHECK-LABEL: scvtf_f32i64_simple:
; CHECK: // %bb.0:
Expand Down Expand Up @@ -315,16 +307,10 @@ define double @ucvtf_f64i32_neg(<4 x i32> %x) {
ret double %conv
}

; This test does not give the indended result of ucvtf d0, s0
; This is due to the input being loaded as a 2 item vector and
; therefore using vector inputs that do not match the pattern
; This test will be fixed in a future revision
define <1 x double> @ucvtf_f64i32_simple(<1 x i32> %x) {
; CHECK-LABEL: ucvtf_f64i32_simple:
; CHECK: // %bb.0:
; CHECK-NEXT: ushll v0.2d, v0.2s, #0
; CHECK-NEXT: ucvtf v0.2d, v0.2d
; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-NEXT: ucvtf d0, s0
; CHECK-NEXT: ret
;
; CHECK-NO-FPRCVT-LABEL: ucvtf_f64i32_simple:
Expand Down Expand Up @@ -449,3 +435,26 @@ define <1 x float> @ucvtf_f32i64_simple(<1 x i64> %x) {
%conv = uitofp <1 x i64> %x to <1 x float>
ret <1 x float> %conv
}

define <1 x double> @uitofp_sext_v2i32_extract_lane0(<2 x i32> %x) {
; CHECK-LABEL: uitofp_sext_v2i32_extract_lane0:
; CHECK: // %bb.0:
; CHECK-NEXT: sshll v0.2d, v0.2s, #0
; CHECK-NEXT: ucvtf v0.2d, v0.2d
; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-NOT: ucvtf d0, s0
; CHECK-NOT: scvtf d0, s0
; CHECK-NEXT: ret
;
; CHECK-NO-FPRCVT-LABEL: uitofp_sext_v2i32_extract_lane0:
; CHECK-NO-FPRCVT: // %bb.0:
; CHECK-NO-FPRCVT-NEXT: sshll v0.2d, v0.2s, #0
; CHECK-NO-FPRCVT-NEXT: ucvtf v0.2d, v0.2d
; CHECK-NO-FPRCVT-NEXT: // kill: def $d0 killed $d0 killed $q0
; CHECK-NO-FPRCVT-NEXT: ret
%wide = sext <2 x i32> %x to <2 x i64>
%fpv2 = uitofp <2 x i64> %wide to <2 x double>
%lane0 = shufflevector <2 x double> %fpv2, <2 x double> poison, <1 x i32> zeroinitializer
ret <1 x double> %lane0
}