AZero13 6a425f1e54
[ARM] Have custom lowering for ucmp and scmp (#149315)
Limited to non-thumb1 for scmp at the moment, since there is no good way
to do it.
2025-08-08 06:51:18 +01:00

377 lines
11 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=thumbv6m-eabi %s -o - | FileCheck %s -check-prefix=THUMB1
; RUN: llc -mtriple=thumbv7-apple-darwin %s -o - | FileCheck %s -check-prefix=THUMB2
; RUN: llc -mtriple thumbv8.1m.main-none-eabi -o - %s | FileCheck %s --check-prefix=V81M
define i8 @ucmp_8_8(i8 zeroext %x, i8 zeroext %y) nounwind {
; THUMB1-LABEL: ucmp_8_8:
; THUMB1: @ %bb.0:
; THUMB1-NEXT: subs r2, r0, r1
; THUMB1-NEXT: sbcs r2, r2
; THUMB1-NEXT: cmp r1, r0
; THUMB1-NEXT: sbcs r1, r1
; THUMB1-NEXT: subs r0, r2, r1
; THUMB1-NEXT: bx lr
;
; THUMB2-LABEL: ucmp_8_8:
; THUMB2: @ %bb.0:
; THUMB2-NEXT: subs r0, r0, r1
; THUMB2-NEXT: it hi
; THUMB2-NEXT: movhi r0, #1
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r0, #-1
; THUMB2-NEXT: bx lr
;
; V81M-LABEL: ucmp_8_8:
; V81M: @ %bb.0:
; V81M-NEXT: cmp r0, r1
; V81M-NEXT: cset r0, hi
; V81M-NEXT: it lo
; V81M-NEXT: movlo.w r0, #-1
; V81M-NEXT: bx lr
%1 = call i8 @llvm.ucmp(i8 %x, i8 %y)
ret i8 %1
}
define i8 @ucmp_8_16(i16 zeroext %x, i16 zeroext %y) nounwind {
; THUMB1-LABEL: ucmp_8_16:
; THUMB1: @ %bb.0:
; THUMB1-NEXT: subs r2, r0, r1
; THUMB1-NEXT: sbcs r2, r2
; THUMB1-NEXT: cmp r1, r0
; THUMB1-NEXT: sbcs r1, r1
; THUMB1-NEXT: subs r0, r2, r1
; THUMB1-NEXT: bx lr
;
; THUMB2-LABEL: ucmp_8_16:
; THUMB2: @ %bb.0:
; THUMB2-NEXT: subs r0, r0, r1
; THUMB2-NEXT: it hi
; THUMB2-NEXT: movhi r0, #1
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r0, #-1
; THUMB2-NEXT: bx lr
;
; V81M-LABEL: ucmp_8_16:
; V81M: @ %bb.0:
; V81M-NEXT: cmp r0, r1
; V81M-NEXT: cset r0, hi
; V81M-NEXT: it lo
; V81M-NEXT: movlo.w r0, #-1
; V81M-NEXT: bx lr
%1 = call i8 @llvm.ucmp(i16 %x, i16 %y)
ret i8 %1
}
define i8 @ucmp_8_32(i32 %x, i32 %y) nounwind {
; THUMB1-LABEL: ucmp_8_32:
; THUMB1: @ %bb.0:
; THUMB1-NEXT: subs r2, r0, r1
; THUMB1-NEXT: sbcs r2, r2
; THUMB1-NEXT: cmp r1, r0
; THUMB1-NEXT: sbcs r1, r1
; THUMB1-NEXT: subs r0, r2, r1
; THUMB1-NEXT: bx lr
;
; THUMB2-LABEL: ucmp_8_32:
; THUMB2: @ %bb.0:
; THUMB2-NEXT: subs r0, r0, r1
; THUMB2-NEXT: it hi
; THUMB2-NEXT: movhi r0, #1
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r0, #-1
; THUMB2-NEXT: bx lr
;
; V81M-LABEL: ucmp_8_32:
; V81M: @ %bb.0:
; V81M-NEXT: cmp r0, r1
; V81M-NEXT: cset r0, hi
; V81M-NEXT: it lo
; V81M-NEXT: movlo.w r0, #-1
; V81M-NEXT: bx lr
%1 = call i8 @llvm.ucmp(i32 %x, i32 %y)
ret i8 %1
}
define i8 @ucmp_8_64(i64 %x, i64 %y) nounwind {
; THUMB1-LABEL: ucmp_8_64:
; THUMB1: @ %bb.0:
; THUMB1-NEXT: .save {r4, r5, r6, lr}
; THUMB1-NEXT: push {r4, r5, r6, lr}
; THUMB1-NEXT: movs r4, #1
; THUMB1-NEXT: movs r5, #0
; THUMB1-NEXT: subs r6, r0, r2
; THUMB1-NEXT: mov r6, r1
; THUMB1-NEXT: sbcs r6, r3
; THUMB1-NEXT: mov r6, r4
; THUMB1-NEXT: blo .LBB3_2
; THUMB1-NEXT: @ %bb.1:
; THUMB1-NEXT: mov r6, r5
; THUMB1-NEXT: .LBB3_2:
; THUMB1-NEXT: subs r0, r2, r0
; THUMB1-NEXT: sbcs r3, r1
; THUMB1-NEXT: blo .LBB3_4
; THUMB1-NEXT: @ %bb.3:
; THUMB1-NEXT: mov r4, r5
; THUMB1-NEXT: .LBB3_4:
; THUMB1-NEXT: subs r0, r4, r6
; THUMB1-NEXT: pop {r4, r5, r6, pc}
;
; THUMB2-LABEL: ucmp_8_64:
; THUMB2: @ %bb.0:
; THUMB2-NEXT: subs.w r12, r0, r2
; THUMB2-NEXT: mov.w r9, #0
; THUMB2-NEXT: sbcs.w r12, r1, r3
; THUMB2-NEXT: mov.w r12, #0
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r12, #1
; THUMB2-NEXT: subs r0, r2, r0
; THUMB2-NEXT: sbcs.w r0, r3, r1
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r9, #1
; THUMB2-NEXT: sub.w r0, r9, r12
; THUMB2-NEXT: bx lr
;
; V81M-LABEL: ucmp_8_64:
; V81M: @ %bb.0:
; V81M-NEXT: subs.w r12, r0, r2
; V81M-NEXT: sbcs.w r12, r1, r3
; V81M-NEXT: cset r12, lo
; V81M-NEXT: subs r0, r2, r0
; V81M-NEXT: sbcs.w r0, r3, r1
; V81M-NEXT: cset r0, lo
; V81M-NEXT: sub.w r0, r0, r12
; V81M-NEXT: bx lr
%1 = call i8 @llvm.ucmp(i64 %x, i64 %y)
ret i8 %1
}
define i8 @ucmp_8_128(i128 %x, i128 %y) nounwind {
; THUMB1-LABEL: ucmp_8_128:
; THUMB1: @ %bb.0:
; THUMB1-NEXT: .save {r4, r5, r6, r7, lr}
; THUMB1-NEXT: push {r4, r5, r6, r7, lr}
; THUMB1-NEXT: .pad #20
; THUMB1-NEXT: sub sp, #20
; THUMB1-NEXT: str r3, [sp, #16] @ 4-byte Spill
; THUMB1-NEXT: movs r3, #1
; THUMB1-NEXT: str r3, [sp] @ 4-byte Spill
; THUMB1-NEXT: movs r3, #0
; THUMB1-NEXT: str r3, [sp, #12] @ 4-byte Spill
; THUMB1-NEXT: ldr r6, [sp, #52]
; THUMB1-NEXT: add r7, sp, #40
; THUMB1-NEXT: ldm r7, {r3, r5, r7}
; THUMB1-NEXT: subs r4, r0, r3
; THUMB1-NEXT: str r1, [sp, #4] @ 4-byte Spill
; THUMB1-NEXT: mov r4, r1
; THUMB1-NEXT: ldr r1, [sp] @ 4-byte Reload
; THUMB1-NEXT: sbcs r4, r5
; THUMB1-NEXT: str r2, [sp, #8] @ 4-byte Spill
; THUMB1-NEXT: mov r4, r2
; THUMB1-NEXT: sbcs r4, r7
; THUMB1-NEXT: ldr r4, [sp, #16] @ 4-byte Reload
; THUMB1-NEXT: sbcs r4, r6
; THUMB1-NEXT: mov r2, r1
; THUMB1-NEXT: blo .LBB4_2
; THUMB1-NEXT: @ %bb.1:
; THUMB1-NEXT: ldr r2, [sp, #12] @ 4-byte Reload
; THUMB1-NEXT: .LBB4_2:
; THUMB1-NEXT: subs r0, r3, r0
; THUMB1-NEXT: ldr r0, [sp, #4] @ 4-byte Reload
; THUMB1-NEXT: sbcs r5, r0
; THUMB1-NEXT: ldr r0, [sp, #8] @ 4-byte Reload
; THUMB1-NEXT: sbcs r7, r0
; THUMB1-NEXT: ldr r0, [sp, #16] @ 4-byte Reload
; THUMB1-NEXT: sbcs r6, r0
; THUMB1-NEXT: blo .LBB4_4
; THUMB1-NEXT: @ %bb.3:
; THUMB1-NEXT: ldr r1, [sp, #12] @ 4-byte Reload
; THUMB1-NEXT: .LBB4_4:
; THUMB1-NEXT: subs r0, r1, r2
; THUMB1-NEXT: add sp, #20
; THUMB1-NEXT: pop {r4, r5, r6, r7, pc}
;
; THUMB2-LABEL: ucmp_8_128:
; THUMB2: @ %bb.0:
; THUMB2-NEXT: push {r4, r5, r6, lr}
; THUMB2-NEXT: add.w lr, sp, #16
; THUMB2-NEXT: ldr r4, [sp, #28]
; THUMB2-NEXT: movs r5, #0
; THUMB2-NEXT: ldm.w lr, {r9, r12, lr}
; THUMB2-NEXT: subs.w r6, r0, r9
; THUMB2-NEXT: sbcs.w r6, r1, r12
; THUMB2-NEXT: sbcs.w r6, r2, lr
; THUMB2-NEXT: sbcs.w r6, r3, r4
; THUMB2-NEXT: mov.w r6, #0
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo r6, #1
; THUMB2-NEXT: subs.w r0, r9, r0
; THUMB2-NEXT: sbcs.w r0, r12, r1
; THUMB2-NEXT: sbcs.w r0, lr, r2
; THUMB2-NEXT: sbcs.w r0, r4, r3
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo r5, #1
; THUMB2-NEXT: subs r0, r5, r6
; THUMB2-NEXT: pop {r4, r5, r6, pc}
;
; V81M-LABEL: ucmp_8_128:
; V81M: @ %bb.0:
; V81M-NEXT: .save {r4, r5, r6, lr}
; V81M-NEXT: push {r4, r5, r6, lr}
; V81M-NEXT: ldrd r5, r4, [sp, #16]
; V81M-NEXT: ldrd lr, r12, [sp, #24]
; V81M-NEXT: subs r6, r0, r5
; V81M-NEXT: sbcs.w r6, r1, r4
; V81M-NEXT: sbcs.w r6, r2, lr
; V81M-NEXT: sbcs.w r6, r3, r12
; V81M-NEXT: cset r6, lo
; V81M-NEXT: subs r0, r5, r0
; V81M-NEXT: sbcs.w r0, r4, r1
; V81M-NEXT: sbcs.w r0, lr, r2
; V81M-NEXT: sbcs.w r0, r12, r3
; V81M-NEXT: cset r0, lo
; V81M-NEXT: subs r0, r0, r6
; V81M-NEXT: pop {r4, r5, r6, pc}
%1 = call i8 @llvm.ucmp(i128 %x, i128 %y)
ret i8 %1
}
define i32 @ucmp_32_32(i32 %x, i32 %y) nounwind {
; THUMB1-LABEL: ucmp_32_32:
; THUMB1: @ %bb.0:
; THUMB1-NEXT: subs r2, r0, r1
; THUMB1-NEXT: sbcs r2, r2
; THUMB1-NEXT: cmp r1, r0
; THUMB1-NEXT: sbcs r1, r1
; THUMB1-NEXT: subs r0, r2, r1
; THUMB1-NEXT: bx lr
;
; THUMB2-LABEL: ucmp_32_32:
; THUMB2: @ %bb.0:
; THUMB2-NEXT: subs r0, r0, r1
; THUMB2-NEXT: it hi
; THUMB2-NEXT: movhi r0, #1
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r0, #-1
; THUMB2-NEXT: bx lr
;
; V81M-LABEL: ucmp_32_32:
; V81M: @ %bb.0:
; V81M-NEXT: cmp r0, r1
; V81M-NEXT: cset r0, hi
; V81M-NEXT: it lo
; V81M-NEXT: movlo.w r0, #-1
; V81M-NEXT: bx lr
%1 = call i32 @llvm.ucmp(i32 %x, i32 %y)
ret i32 %1
}
define i32 @ucmp_32_64(i64 %x, i64 %y) nounwind {
; THUMB1-LABEL: ucmp_32_64:
; THUMB1: @ %bb.0:
; THUMB1-NEXT: .save {r4, r5, r6, lr}
; THUMB1-NEXT: push {r4, r5, r6, lr}
; THUMB1-NEXT: movs r4, #1
; THUMB1-NEXT: movs r5, #0
; THUMB1-NEXT: subs r6, r0, r2
; THUMB1-NEXT: mov r6, r1
; THUMB1-NEXT: sbcs r6, r3
; THUMB1-NEXT: mov r6, r4
; THUMB1-NEXT: blo .LBB6_2
; THUMB1-NEXT: @ %bb.1:
; THUMB1-NEXT: mov r6, r5
; THUMB1-NEXT: .LBB6_2:
; THUMB1-NEXT: subs r0, r2, r0
; THUMB1-NEXT: sbcs r3, r1
; THUMB1-NEXT: blo .LBB6_4
; THUMB1-NEXT: @ %bb.3:
; THUMB1-NEXT: mov r4, r5
; THUMB1-NEXT: .LBB6_4:
; THUMB1-NEXT: subs r0, r4, r6
; THUMB1-NEXT: pop {r4, r5, r6, pc}
;
; THUMB2-LABEL: ucmp_32_64:
; THUMB2: @ %bb.0:
; THUMB2-NEXT: subs.w r12, r0, r2
; THUMB2-NEXT: mov.w r9, #0
; THUMB2-NEXT: sbcs.w r12, r1, r3
; THUMB2-NEXT: mov.w r12, #0
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r12, #1
; THUMB2-NEXT: subs r0, r2, r0
; THUMB2-NEXT: sbcs.w r0, r3, r1
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r9, #1
; THUMB2-NEXT: sub.w r0, r9, r12
; THUMB2-NEXT: bx lr
;
; V81M-LABEL: ucmp_32_64:
; V81M: @ %bb.0:
; V81M-NEXT: subs.w r12, r0, r2
; V81M-NEXT: sbcs.w r12, r1, r3
; V81M-NEXT: cset r12, lo
; V81M-NEXT: subs r0, r2, r0
; V81M-NEXT: sbcs.w r0, r3, r1
; V81M-NEXT: cset r0, lo
; V81M-NEXT: sub.w r0, r0, r12
; V81M-NEXT: bx lr
%1 = call i32 @llvm.ucmp(i64 %x, i64 %y)
ret i32 %1
}
define i64 @ucmp_64_64(i64 %x, i64 %y) nounwind {
; THUMB1-LABEL: ucmp_64_64:
; THUMB1: @ %bb.0:
; THUMB1-NEXT: .save {r4, r5, r6, lr}
; THUMB1-NEXT: push {r4, r5, r6, lr}
; THUMB1-NEXT: movs r4, #1
; THUMB1-NEXT: movs r5, #0
; THUMB1-NEXT: subs r6, r0, r2
; THUMB1-NEXT: mov r6, r1
; THUMB1-NEXT: sbcs r6, r3
; THUMB1-NEXT: mov r6, r4
; THUMB1-NEXT: blo .LBB7_2
; THUMB1-NEXT: @ %bb.1:
; THUMB1-NEXT: mov r6, r5
; THUMB1-NEXT: .LBB7_2:
; THUMB1-NEXT: subs r0, r2, r0
; THUMB1-NEXT: sbcs r3, r1
; THUMB1-NEXT: blo .LBB7_4
; THUMB1-NEXT: @ %bb.3:
; THUMB1-NEXT: mov r4, r5
; THUMB1-NEXT: .LBB7_4:
; THUMB1-NEXT: subs r0, r4, r6
; THUMB1-NEXT: asrs r1, r0, #31
; THUMB1-NEXT: pop {r4, r5, r6, pc}
;
; THUMB2-LABEL: ucmp_64_64:
; THUMB2: @ %bb.0:
; THUMB2-NEXT: subs.w r12, r0, r2
; THUMB2-NEXT: mov.w r9, #0
; THUMB2-NEXT: sbcs.w r12, r1, r3
; THUMB2-NEXT: mov.w r12, #0
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r12, #1
; THUMB2-NEXT: subs r0, r2, r0
; THUMB2-NEXT: sbcs.w r0, r3, r1
; THUMB2-NEXT: it lo
; THUMB2-NEXT: movlo.w r9, #1
; THUMB2-NEXT: sub.w r0, r9, r12
; THUMB2-NEXT: asrs r1, r0, #31
; THUMB2-NEXT: bx lr
;
; V81M-LABEL: ucmp_64_64:
; V81M: @ %bb.0:
; V81M-NEXT: subs.w r12, r0, r2
; V81M-NEXT: sbcs.w r12, r1, r3
; V81M-NEXT: cset r12, lo
; V81M-NEXT: subs r0, r2, r0
; V81M-NEXT: sbcs.w r0, r3, r1
; V81M-NEXT: cset r0, lo
; V81M-NEXT: sub.w r0, r0, r12
; V81M-NEXT: asrs r1, r0, #31
; V81M-NEXT: bx lr
%1 = call i64 @llvm.ucmp(i64 %x, i64 %y)
ret i64 %1
}