; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=armv7a-none-eabihf -mattr=+neon | FileCheck %s --check-prefixes=CHECK,CHECK-NEON
; RUN: llc < %s -mtriple=armv8a-none-eabihf -mattr=+neon,+fullfp16 | FileCheck %s --check-prefixes=CHECK,CHECK-FP16
; i32 saturate
define <2 x i32> @stest_f64i32(<2 x double> %x) {
; CHECK-LABEL: stest_f64i32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r11, lr}
; CHECK-NEXT: push {r4, r5, r11, lr}
; CHECK-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-NEXT: vpush {d8, d9, d10, d11}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, r1, d8
; CHECK-NEXT: bl __aeabi_d2lz
; CHECK-NEXT: mov r4, r0
; CHECK-NEXT: mov r5, r1
; CHECK-NEXT: vmov r0, r1, d9
; CHECK-NEXT: adr r2, .LCPI0_0
; CHECK-NEXT: vld1.64 {d8, d9}, [r2:128]
; CHECK-NEXT: vmov.32 d10[0], r4
; CHECK-NEXT: bl __aeabi_d2lz
; CHECK-NEXT: mvn r3, #-2147483648
; CHECK-NEXT: subs r4, r4, r3
; CHECK-NEXT: sbcs r4, r5, #0
; CHECK-NEXT: vmov.32 d11[0], r0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: subs r0, r0, r3
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: vmov.32 d11[1], r1
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: vmov.i32 q10, #0x80000000
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: vmov.32 d10[1], r5
; CHECK-NEXT: mvnne r4, #0
; CHECK-NEXT: vdup.32 d17, r0
; CHECK-NEXT: vdup.32 d16, r4
; CHECK-NEXT: mvn r4, #0
; CHECK-NEXT: vbsl q8, q5, q4
; CHECK-NEXT: vmov r0, r1, d16
; CHECK-NEXT: vmov r3, r5, d17
; CHECK-NEXT: rsbs r0, r0, #-2147483648
; CHECK-NEXT: sbcs r0, r4, r1
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: rsbs r1, r3, #-2147483648
; CHECK-NEXT: sbcs r1, r4, r5
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: vdup.32 d19, r2
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: vdup.32 d18, r0
; CHECK-NEXT: vbif q8, q10, q9
; CHECK-NEXT: vmovn.i64 d0, q8
; CHECK-NEXT: vpop {d8, d9, d10, d11}
; CHECK-NEXT: pop {r4, r5, r11, pc}
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: @ %bb.1:
; CHECK-NEXT: .LCPI0_0:
; CHECK-NEXT: .long 2147483647 @ 0x7fffffff
; CHECK-NEXT: .long 0 @ 0x0
; CHECK-NEXT: .long 2147483647 @ 0x7fffffff
; CHECK-NEXT: .long 0 @ 0x0
entry:
%conv = fptosi <2 x double> %x to <2 x i64>
%0 = icmp slt <2 x i64> %conv, <i64 2147483647, i64 2147483647>
%spec.store.select = select <2 x i1> %0, <2 x i64> %conv, <2 x i64> <i64 2147483647, i64 2147483647>
%1 = icmp sgt <2 x i64> %spec.store.select, <i64 -2147483648, i64 -2147483648>
%spec.store.select7 = select <2 x i1> %1, <2 x i64> %spec.store.select, <2 x i64> <i64 -2147483648, i64 -2147483648>
%conv6 = trunc <2 x i64> %spec.store.select7 to <2 x i32>
ret <2 x i32> %conv6
}
define <2 x i32> @utest_f64i32(<2 x double> %x) {
; CHECK-LABEL: utest_f64i32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r11, lr}
; CHECK-NEXT: push {r4, r5, r11, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, r1, d9
; CHECK-NEXT: bl __aeabi_d2ulz
; CHECK-NEXT: mov r4, r0
; CHECK-NEXT: mov r5, r1
; CHECK-NEXT: vmov r0, r1, d8
; CHECK-NEXT: vmov.32 d9[0], r4
; CHECK-NEXT: bl __aeabi_d2ulz
; CHECK-NEXT: mvn r3, #0
; CHECK-NEXT: vmov.32 d8[0], r0
; CHECK-NEXT: subs r0, r0, r3
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: movwlo r0, #1
; CHECK-NEXT: subs r1, r4, r3
; CHECK-NEXT: sbcs r1, r5, #0
; CHECK-NEXT: movwlo r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: vdup.32 d17, r2
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: vdup.32 d16, r0
; CHECK-NEXT: vand q9, q4, q8
; CHECK-NEXT: vorn q8, q9, q8
; CHECK-NEXT: vmovn.i64 d0, q8
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r11, pc}
entry:
%conv = fptoui <2 x double> %x to <2 x i64>
%0 = icmp ult <2 x i64> %conv, <i64 4294967295, i64 4294967295>
%spec.store.select = select <2 x i1> %0, <2 x i64> %conv, <2 x i64> <i64 4294967295, i64 4294967295>
%conv6 = trunc <2 x i64> %spec.store.select to <2 x i32>
ret <2 x i32> %conv6
}
define <2 x i32> @ustest_f64i32(<2 x double> %x) {
; CHECK-LABEL: ustest_f64i32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r11, lr}
; CHECK-NEXT: push {r4, r5, r11, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, r1, d8
; CHECK-NEXT: bl __aeabi_d2lz
; CHECK-NEXT: mov r4, r0
; CHECK-NEXT: mov r5, r1
; CHECK-NEXT: vmov r0, r1, d9
; CHECK-NEXT: vmov.32 d8[0], r4
; CHECK-NEXT: bl __aeabi_d2lz
; CHECK-NEXT: mvn r3, #0
; CHECK-NEXT: subs r4, r4, r3
; CHECK-NEXT: sbcs r4, r5, #0
; CHECK-NEXT: vmov.32 d9[0], r0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: vmov.i64 q9, #0xffffffff
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: subs r0, r0, r3
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: vmov.32 d9[1], r1
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: vmov.32 d8[1], r5
; CHECK-NEXT: mvnne r4, #0
; CHECK-NEXT: vdup.32 d17, r0
; CHECK-NEXT: vdup.32 d16, r4
; CHECK-NEXT: vbsl q8, q4, q9
; CHECK-NEXT: vmov r0, r1, d16
; CHECK-NEXT: vmov r3, r5, d17
; CHECK-NEXT: rsbs r0, r0, #0
; CHECK-NEXT: rscs r0, r1, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: rsbs r1, r3, #0
; CHECK-NEXT: rscs r1, r5, #0
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: vmov.32 d19[0], r2
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: vmov.32 d18[0], r0
; CHECK-NEXT: vand q8, q8, q9
; CHECK-NEXT: vmovn.i64 d0, q8
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r11, pc}
entry:
%conv = fptosi <2 x double> %x to <2 x i64>
%0 = icmp slt <2 x i64> %conv, <i64 4294967295, i64 4294967295>
%spec.store.select = select <2 x i1> %0, <2 x i64> %conv, <2 x i64> <i64 4294967295, i64 4294967295>
%1 = icmp sgt <2 x i64> %spec.store.select, zeroinitializer
%spec.store.select7 = select <2 x i1> %1, <2 x i64> %spec.store.select, <2 x i64> zeroinitializer
%conv6 = trunc <2 x i64> %spec.store.select7 to <2 x i32>
ret <2 x i32> %conv6
}
define <4 x i32> @stest_f32i32(<4 x float> %x) {
; CHECK-LABEL: stest_f32i32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEXT: .pad #4
; CHECK-NEXT: sub sp, sp, #4
; CHECK-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-NEXT: vpush {d8, d9, d10, d11}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, s16
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r7, r0
; CHECK-NEXT: vmov r0, s18
; CHECK-NEXT: mov r8, r1
; CHECK-NEXT: vmov r6, s17
; CHECK-NEXT: vmov r10, s19
; CHECK-NEXT: vmov.32 d8[0], r7
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: vmov.32 d10[0], r0
; CHECK-NEXT: mov r0, r6
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r9, r0
; CHECK-NEXT: vmov.32 d9[0], r0
; CHECK-NEXT: mov r0, r10
; CHECK-NEXT: mov r11, r1
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mvn r6, #-2147483648
; CHECK-NEXT: subs r3, r7, r6
; CHECK-NEXT: sbcs r3, r8, #0
; CHECK-NEXT: vmov.32 d11[0], r0
; CHECK-NEXT: mov r3, #0
; CHECK-NEXT: adr r2, .LCPI3_0
; CHECK-NEXT: movwlt r3, #1
; CHECK-NEXT: subs r7, r5, r6
; CHECK-NEXT: sbcs r7, r4, #0
; CHECK-NEXT: vmov.32 d11[1], r1
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: mvnne r7, #0
; CHECK-NEXT: subs r0, r0, r6
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: vld1.64 {d18, d19}, [r2:128]
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: vmov.32 d10[1], r4
; CHECK-NEXT: vdup.32 d17, r0
; CHECK-NEXT: subs r0, r9, r6
; CHECK-NEXT: sbcs r0, r11, #0
; CHECK-NEXT: vdup.32 d16, r7
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: vbsl q8, q5, q9
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: vmov.32 d9[1], r11
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: mvn r6, #0
; CHECK-NEXT: vdup.32 d21, r0
; CHECK-NEXT: mvnne r3, #0
; CHECK-NEXT: vmov.32 d8[1], r8
; CHECK-NEXT: vmov r0, r1, d16
; CHECK-NEXT: vdup.32 d20, r3
; CHECK-NEXT: vbit q9, q4, q10
; CHECK-NEXT: adr r5, .LCPI3_1
; CHECK-NEXT: vld1.64 {d20, d21}, [r5:128]
; CHECK-NEXT: vmov r5, r4, d17
; CHECK-NEXT: vmov r3, r7, d18
; CHECK-NEXT: rsbs r0, r0, #-2147483648
; CHECK-NEXT: sbcs r0, r6, r1
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: rsbs r1, r3, #-2147483648
; CHECK-NEXT: vmov r1, r3, d19
; CHECK-NEXT: sbcs r7, r6, r7
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: rsbs r5, r5, #-2147483648
; CHECK-NEXT: sbcs r5, r6, r4
; CHECK-NEXT: mov r5, #0
; CHECK-NEXT: movwlt r5, #1
; CHECK-NEXT: rsbs r1, r1, #-2147483648
; CHECK-NEXT: sbcs r1, r6, r3
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: mvnne r5, #0
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: vdup.32 d25, r5
; CHECK-NEXT: mvnne r7, #0
; CHECK-NEXT: vdup.32 d23, r2
; CHECK-NEXT: vdup.32 d24, r0
; CHECK-NEXT: vbif q8, q10, q12
; CHECK-NEXT: vdup.32 d22, r7
; CHECK-NEXT: vbif q9, q10, q11
; CHECK-NEXT: vmovn.i64 d1, q8
; CHECK-NEXT: vmovn.i64 d0, q9
; CHECK-NEXT: vpop {d8, d9, d10, d11}
; CHECK-NEXT: add sp, sp, #4
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, r11, pc}
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: @ %bb.1:
; CHECK-NEXT: .LCPI3_0:
; CHECK-NEXT: .long 2147483647 @ 0x7fffffff
; CHECK-NEXT: .long 0 @ 0x0
; CHECK-NEXT: .long 2147483647 @ 0x7fffffff
; CHECK-NEXT: .long 0 @ 0x0
; CHECK-NEXT: .LCPI3_1:
; CHECK-NEXT: .long 2147483648 @ 0x80000000
; CHECK-NEXT: .long 4294967295 @ 0xffffffff
; CHECK-NEXT: .long 2147483648 @ 0x80000000
; CHECK-NEXT: .long 4294967295 @ 0xffffffff
entry:
%conv = fptosi <4 x float> %x to <4 x i64>
%0 = icmp slt <4 x i64> %conv, <i64 2147483647, i64 2147483647, i64 2147483647, i64 2147483647>
%spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 2147483647, i64 2147483647, i64 2147483647, i64 2147483647>
%1 = icmp sgt <4 x i64> %spec.store.select, <i64 -2147483648, i64 -2147483648, i64 -2147483648, i64 -2147483648>
%spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> <i64 -2147483648, i64 -2147483648, i64 -2147483648, i64 -2147483648>
%conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @utest_f32i32(<4 x float> %x) {
; CHECK-LABEL: utest_f32i32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-NEXT: vpush {d8, d9, d10, d11}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, s17
; CHECK-NEXT: bl __aeabi_f2ulz
; CHECK-NEXT: mov r9, r0
; CHECK-NEXT: vmov r0, s16
; CHECK-NEXT: mov r8, r1
; CHECK-NEXT: vmov r6, s19
; CHECK-NEXT: vmov r7, s18
; CHECK-NEXT: vmov.32 d9[0], r9
; CHECK-NEXT: bl __aeabi_f2ulz
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: vmov.32 d8[0], r0
; CHECK-NEXT: mov r0, r6
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: bl __aeabi_f2ulz
; CHECK-NEXT: mov r6, r0
; CHECK-NEXT: vmov.32 d11[0], r0
; CHECK-NEXT: mov r0, r7
; CHECK-NEXT: mov r10, r1
; CHECK-NEXT: bl __aeabi_f2ulz
; CHECK-NEXT: mvn r7, #0
; CHECK-NEXT: subs r2, r5, r7
; CHECK-NEXT: sbcs r2, r4, #0
; CHECK-NEXT: vmov.32 d10[0], r0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: mov r3, #0
; CHECK-NEXT: movwlo r2, #1
; CHECK-NEXT: subs r0, r0, r7
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: movwlo r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: subs r1, r6, r7
; CHECK-NEXT: sbcs r1, r10, #0
; CHECK-NEXT: mov r1, #0
; CHECK-NEXT: movwlo r1, #1
; CHECK-NEXT: subs r7, r9, r7
; CHECK-NEXT: sbcs r7, r8, #0
; CHECK-NEXT: movwlo r3, #1
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: mvnne r3, #0
; CHECK-NEXT: cmp r1, #0
; CHECK-NEXT: mvnne r1, #0
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: vdup.32 d19, r1
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: vdup.32 d17, r3
; CHECK-NEXT: vdup.32 d18, r0
; CHECK-NEXT: vand q10, q5, q9
; CHECK-NEXT: vdup.32 d16, r2
; CHECK-NEXT: vand q11, q4, q8
; CHECK-NEXT: vorn q9, q10, q9
; CHECK-NEXT: vorn q8, q11, q8
; CHECK-NEXT: vmovn.i64 d1, q9
; CHECK-NEXT: vmovn.i64 d0, q8
; CHECK-NEXT: vpop {d8, d9, d10, d11}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptoui <4 x float> %x to <4 x i64>
%0 = icmp ult <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
%spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
%conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @ustest_f32i32(<4 x float> %x) {
; CHECK-LABEL: ustest_f32i32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: .vsave {d8, d9, d10, d11, d12, d13}
; CHECK-NEXT: vpush {d8, d9, d10, d11, d12, d13}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, s17
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: vmov r0, s16
; CHECK-NEXT: mov r6, r1
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r2, r0
; CHECK-NEXT: vmov r0, s18
; CHECK-NEXT: vmov.32 d16[0], r2
; CHECK-NEXT: mvn r4, #0
; CHECK-NEXT: subs r2, r2, r4
; CHECK-NEXT: vmov r8, s19
; CHECK-NEXT: sbcs r2, r1, #0
; CHECK-NEXT: vmov.32 d17[0], r5
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: vmov.i64 q5, #0xffffffff
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: subs r3, r5, r4
; CHECK-NEXT: sbcs r3, r6, #0
; CHECK-NEXT: vmov.32 d17[1], r6
; CHECK-NEXT: mov r3, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: movwlt r3, #1
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: mvnne r3, #0
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: vdup.32 d19, r3
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: vdup.32 d18, r2
; CHECK-NEXT: vmov.32 d16[1], r1
; CHECK-NEXT: vorr q4, q9, q9
; CHECK-NEXT: vbsl q4, q8, q5
; CHECK-NEXT: vmov r10, r9, d8
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: vmov.32 d12[0], r0
; CHECK-NEXT: mov r0, r8
; CHECK-NEXT: mov r6, r1
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: subs r2, r5, r4
; CHECK-NEXT: vmov.32 d13[0], r0
; CHECK-NEXT: sbcs r2, r6, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: subs r0, r0, r4
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: vmov.32 d13[1], r1
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: vmov r5, r4, d9
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: vmov.32 d12[1], r6
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: vdup.32 d17, r0
; CHECK-NEXT: rsbs r0, r10, #0
; CHECK-NEXT: vdup.32 d16, r2
; CHECK-NEXT: rscs r0, r9, #0
; CHECK-NEXT: vbsl q8, q6, q5
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: vmov r1, r2, d16
; CHECK-NEXT: vmov r3, r6, d17
; CHECK-NEXT: rsbs r1, r1, #0
; CHECK-NEXT: rscs r1, r2, #0
; CHECK-NEXT: mov r1, #0
; CHECK-NEXT: movwlt r1, #1
; CHECK-NEXT: rsbs r2, r3, #0
; CHECK-NEXT: rscs r2, r6, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: rsbs r3, r5, #0
; CHECK-NEXT: rscs r3, r4, #0
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: mvnne r7, #0
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: mvnne r2, #0
; CHECK-NEXT: cmp r1, #0
; CHECK-NEXT: mvnne r1, #0
; CHECK-NEXT: vmov.32 d21[0], r2
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: vmov.32 d20[0], r1
; CHECK-NEXT: mvnne r0, #0
; CHECK-NEXT: vmov.32 d19[0], r7
; CHECK-NEXT: vand q8, q8, q10
; CHECK-NEXT: vmov.32 d18[0], r0
; CHECK-NEXT: vmovn.i64 d1, q8
; CHECK-NEXT: vand q9, q4, q9
; CHECK-NEXT: vmovn.i64 d0, q9
; CHECK-NEXT: vpop {d8, d9, d10, d11, d12, d13}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <4 x float> %x to <4 x i64>
%0 = icmp slt <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
%spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
%1 = icmp sgt <4 x i64> %spec.store.select, zeroinitializer
%spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> zeroinitializer
%conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @stest_f16i32(<4 x half> %x) {
; CHECK-NEON-LABEL: stest_f16i32:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEON-NEXT: .pad #4
; CHECK-NEON-NEXT: sub sp, sp, #4
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11}
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.f32 s16, s3
; CHECK-NEON-NEXT: vmov.f32 s18, s2
; CHECK-NEON-NEXT: vmov.f32 s20, s1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mov r9, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: vmov r10, s16
; CHECK-NEON-NEXT: mov r8, r1
; CHECK-NEON-NEXT: vmov r6, s20
; CHECK-NEON-NEXT: vmov.32 d8[0], r9
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov.32 d10[0], r0
; CHECK-NEON-NEXT: mov r0, r6
; CHECK-NEON-NEXT: mov r4, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mov r11, r0
; CHECK-NEON-NEXT: vmov.32 d9[0], r0
; CHECK-NEON-NEXT: mov r0, r10
; CHECK-NEON-NEXT: mov r7, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mvn r6, #-2147483648
; CHECK-NEON-NEXT: subs r3, r9, r6
; CHECK-NEON-NEXT: sbcs r3, r8, #0
; CHECK-NEON-NEXT: vmov.32 d11[0], r0
; CHECK-NEON-NEXT: mov r3, #0
; CHECK-NEON-NEXT: adr r2, .LCPI6_0
; CHECK-NEON-NEXT: movwlt r3, #1
; CHECK-NEON-NEXT: subs r5, r5, r6
; CHECK-NEON-NEXT: sbcs r5, r4, #0
; CHECK-NEON-NEXT: vmov.32 d11[1], r1
; CHECK-NEON-NEXT: mov r5, #0
; CHECK-NEON-NEXT: movwlt r5, #1
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: mvnne r5, #0
; CHECK-NEON-NEXT: subs r0, r0, r6
; CHECK-NEON-NEXT: sbcs r0, r1, #0
; CHECK-NEON-NEXT: vld1.64 {d18, d19}, [r2:128]
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: mvnne r0, #0
; CHECK-NEON-NEXT: vmov.32 d10[1], r4
; CHECK-NEON-NEXT: vdup.32 d17, r0
; CHECK-NEON-NEXT: subs r0, r11, r6
; CHECK-NEON-NEXT: sbcs r0, r7, #0
; CHECK-NEON-NEXT: vdup.32 d16, r5
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: vbsl q8, q5, q9
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: vmov.32 d9[1], r7
; CHECK-NEON-NEXT: mvnne r0, #0
; CHECK-NEON-NEXT: cmp r3, #0
; CHECK-NEON-NEXT: mvn r6, #0
; CHECK-NEON-NEXT: vdup.32 d21, r0
; CHECK-NEON-NEXT: mvnne r3, #0
; CHECK-NEON-NEXT: vmov.32 d8[1], r8
; CHECK-NEON-NEXT: vmov r0, r1, d16
; CHECK-NEON-NEXT: vdup.32 d20, r3
; CHECK-NEON-NEXT: vbit q9, q4, q10
; CHECK-NEON-NEXT: adr r5, .LCPI6_1
; CHECK-NEON-NEXT: vld1.64 {d20, d21}, [r5:128]
; CHECK-NEON-NEXT: vmov r5, r4, d17
; CHECK-NEON-NEXT: vmov r3, r7, d18
; CHECK-NEON-NEXT: rsbs r0, r0, #-2147483648
; CHECK-NEON-NEXT: sbcs r0, r6, r1
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: mvnne r0, #0
; CHECK-NEON-NEXT: rsbs r1, r3, #-2147483648
; CHECK-NEON-NEXT: vmov r1, r3, d19
; CHECK-NEON-NEXT: sbcs r7, r6, r7
; CHECK-NEON-NEXT: mov r7, #0
; CHECK-NEON-NEXT: movwlt r7, #1
; CHECK-NEON-NEXT: rsbs r5, r5, #-2147483648
; CHECK-NEON-NEXT: sbcs r5, r6, r4
; CHECK-NEON-NEXT: mov r5, #0
; CHECK-NEON-NEXT: movwlt r5, #1
; CHECK-NEON-NEXT: rsbs r1, r1, #-2147483648
; CHECK-NEON-NEXT: sbcs r1, r6, r3
; CHECK-NEON-NEXT: movwlt r2, #1
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: mvnne r2, #0
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: mvnne r5, #0
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: vdup.32 d25, r5
; CHECK-NEON-NEXT: mvnne r7, #0
; CHECK-NEON-NEXT: vdup.32 d23, r2
; CHECK-NEON-NEXT: vdup.32 d24, r0
; CHECK-NEON-NEXT: vbif q8, q10, q12
; CHECK-NEON-NEXT: vdup.32 d22, r7
; CHECK-NEON-NEXT: vbif q9, q10, q11
; CHECK-NEON-NEXT: vmovn.i64 d1, q8
; CHECK-NEON-NEXT: vmovn.i64 d0, q9
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11}
; CHECK-NEON-NEXT: add sp, sp, #4
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, r11, pc}
; CHECK-NEON-NEXT: .p2align 4
; CHECK-NEON-NEXT: @ %bb.1:
; CHECK-NEON-NEXT: .LCPI6_0:
; CHECK-NEON-NEXT: .long 2147483647 @ 0x7fffffff
; CHECK-NEON-NEXT: .long 0 @ 0x0
; CHECK-NEON-NEXT: .long 2147483647 @ 0x7fffffff
; CHECK-NEON-NEXT: .long 0 @ 0x0
; CHECK-NEON-NEXT: .LCPI6_1:
; CHECK-NEON-NEXT: .long 2147483648 @ 0x80000000
; CHECK-NEON-NEXT: .long 4294967295 @ 0xffffffff
; CHECK-NEON-NEXT: .long 2147483648 @ 0x80000000
; CHECK-NEON-NEXT: .long 4294967295 @ 0xffffffff
;
; CHECK-FP16-LABEL: stest_f16i32:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: .vsave {d10, d11, d12, d13}
; CHECK-FP16-NEXT: vpush {d10, d11, d12, d13}
; CHECK-FP16-NEXT: .vsave {d8}
; CHECK-FP16-NEXT: vpush {d8}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[0]
; CHECK-FP16-NEXT: vorr d8, d0, d0
; CHECK-FP16-NEXT: vmov.u16 r6, d0[1]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: mov r4, r0
; CHECK-FP16-NEXT: vmov.u16 r0, d8[2]
; CHECK-FP16-NEXT: mov r8, r1
; CHECK-FP16-NEXT: vmov.32 d10[0], r4
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: vmov s0, r6
; CHECK-FP16-NEXT: mov r5, r0
; CHECK-FP16-NEXT: mov r7, r1
; CHECK-FP16-NEXT: vmov.32 d12[0], r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: mov r9, r0
; CHECK-FP16-NEXT: vmov.u16 r0, d8[3]
; CHECK-FP16-NEXT: mov r10, r1
; CHECK-FP16-NEXT: vmov.32 d11[0], r9
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: mvn r6, #-2147483648
; CHECK-FP16-NEXT: subs r3, r4, r6
; CHECK-FP16-NEXT: sbcs r3, r8, #0
; CHECK-FP16-NEXT: vmov.32 d13[0], r0
; CHECK-FP16-NEXT: mov r3, #0
; CHECK-FP16-NEXT: adr r2, .LCPI6_0
; CHECK-FP16-NEXT: movwlt r3, #1
; CHECK-FP16-NEXT: subs r5, r5, r6
; CHECK-FP16-NEXT: sbcs r5, r7, #0
; CHECK-FP16-NEXT: vmov.32 d13[1], r1
; CHECK-FP16-NEXT: mov r5, #0
; CHECK-FP16-NEXT: movwlt r5, #1
; CHECK-FP16-NEXT: cmp r5, #0
; CHECK-FP16-NEXT: mvnne r5, #0
; CHECK-FP16-NEXT: subs r0, r0, r6
; CHECK-FP16-NEXT: sbcs r0, r1, #0
; CHECK-FP16-NEXT: vld1.64 {d18, d19}, [r2:128]
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: mov r2, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: mvnne r0, #0
; CHECK-FP16-NEXT: vmov.32 d12[1], r7
; CHECK-FP16-NEXT: vdup.32 d17, r0
; CHECK-FP16-NEXT: subs r0, r9, r6
; CHECK-FP16-NEXT: sbcs r0, r10, #0
; CHECK-FP16-NEXT: vdup.32 d16, r5
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: vbsl q8, q6, q9
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: vmov.32 d11[1], r10
; CHECK-FP16-NEXT: mvnne r0, #0
; CHECK-FP16-NEXT: cmp r3, #0
; CHECK-FP16-NEXT: mvn r6, #0
; CHECK-FP16-NEXT: vdup.32 d21, r0
; CHECK-FP16-NEXT: mvnne r3, #0
; CHECK-FP16-NEXT: vmov.32 d10[1], r8
; CHECK-FP16-NEXT: vmov r0, r1, d16
; CHECK-FP16-NEXT: vdup.32 d20, r3
; CHECK-FP16-NEXT: vbit q9, q5, q10
; CHECK-FP16-NEXT: adr r5, .LCPI6_1
; CHECK-FP16-NEXT: vld1.64 {d20, d21}, [r5:128]
; CHECK-FP16-NEXT: vmov r5, r4, d17
; CHECK-FP16-NEXT: vmov r3, r7, d18
; CHECK-FP16-NEXT: rsbs r0, r0, #-2147483648
; CHECK-FP16-NEXT: sbcs r0, r6, r1
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: mvnne r0, #0
; CHECK-FP16-NEXT: rsbs r1, r3, #-2147483648
; CHECK-FP16-NEXT: vmov r1, r3, d19
; CHECK-FP16-NEXT: sbcs r7, r6, r7
; CHECK-FP16-NEXT: mov r7, #0
; CHECK-FP16-NEXT: movwlt r7, #1
; CHECK-FP16-NEXT: rsbs r5, r5, #-2147483648
; CHECK-FP16-NEXT: sbcs r5, r6, r4
; CHECK-FP16-NEXT: mov r5, #0
; CHECK-FP16-NEXT: movwlt r5, #1
; CHECK-FP16-NEXT: rsbs r1, r1, #-2147483648
; CHECK-FP16-NEXT: sbcs r1, r6, r3
; CHECK-FP16-NEXT: movwlt r2, #1
; CHECK-FP16-NEXT: cmp r2, #0
; CHECK-FP16-NEXT: mvnne r2, #0
; CHECK-FP16-NEXT: cmp r5, #0
; CHECK-FP16-NEXT: mvnne r5, #0
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: vdup.32 d25, r5
; CHECK-FP16-NEXT: mvnne r7, #0
; CHECK-FP16-NEXT: vdup.32 d23, r2
; CHECK-FP16-NEXT: vdup.32 d24, r0
; CHECK-FP16-NEXT: vbif q8, q10, q12
; CHECK-FP16-NEXT: vdup.32 d22, r7
; CHECK-FP16-NEXT: vbif q9, q10, q11
; CHECK-FP16-NEXT: vmovn.i64 d1, q8
; CHECK-FP16-NEXT: vmovn.i64 d0, q9
; CHECK-FP16-NEXT: vpop {d8}
; CHECK-FP16-NEXT: vpop {d10, d11, d12, d13}
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
; CHECK-FP16-NEXT: .p2align 4
; CHECK-FP16-NEXT: @ %bb.1:
; CHECK-FP16-NEXT: .LCPI6_0:
; CHECK-FP16-NEXT: .long 2147483647 @ 0x7fffffff
; CHECK-FP16-NEXT: .long 0 @ 0x0
; CHECK-FP16-NEXT: .long 2147483647 @ 0x7fffffff
; CHECK-FP16-NEXT: .long 0 @ 0x0
; CHECK-FP16-NEXT: .LCPI6_1:
; CHECK-FP16-NEXT: .long 2147483648 @ 0x80000000
; CHECK-FP16-NEXT: .long 4294967295 @ 0xffffffff
; CHECK-FP16-NEXT: .long 2147483648 @ 0x80000000
; CHECK-FP16-NEXT: .long 4294967295 @ 0xffffffff
entry:
%conv = fptosi <4 x half> %x to <4 x i64>
%0 = icmp slt <4 x i64> %conv, <i64 2147483647, i64 2147483647, i64 2147483647, i64 2147483647>
%spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 2147483647, i64 2147483647, i64 2147483647, i64 2147483647>
%1 = icmp sgt <4 x i64> %spec.store.select, <i64 -2147483648, i64 -2147483648, i64 -2147483648, i64 -2147483648>
%spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> <i64 -2147483648, i64 -2147483648, i64 -2147483648, i64 -2147483648>
%conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @utesth_f16i32(<4 x half> %x) {
; CHECK-NEON-LABEL: utesth_f16i32:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEON-NEXT: .vsave {d12, d13}
; CHECK-NEON-NEXT: vpush {d12, d13}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10}
; CHECK-NEON-NEXT: vpush {d8, d9, d10}
; CHECK-NEON-NEXT: vmov r0, s3
; CHECK-NEON-NEXT: vmov.f32 s16, s2
; CHECK-NEON-NEXT: vmov.f32 s18, s1
; CHECK-NEON-NEXT: vmov.f32 s20, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2ulz
; CHECK-NEON-NEXT: mov r10, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: mov r8, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2ulz
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: vmov.32 d13[0], r0
; CHECK-NEON-NEXT: vmov r0, s20
; CHECK-NEON-NEXT: mov r9, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2ulz
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov.32 d12[0], r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: mov r7, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov.32 d9[0], r10
; CHECK-NEON-NEXT: bl __aeabi_f2ulz
; CHECK-NEON-NEXT: mvn r4, #0
; CHECK-NEON-NEXT: subs r2, r5, r4
; CHECK-NEON-NEXT: sbcs r2, r7, #0
; CHECK-NEON-NEXT: vmov.32 d8[0], r0
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: mov r3, #0
; CHECK-NEON-NEXT: movwlo r2, #1
; CHECK-NEON-NEXT: subs r0, r0, r4
; CHECK-NEON-NEXT: sbcs r0, r1, #0
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: movwlo r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: mvnne r0, #0
; CHECK-NEON-NEXT: subs r1, r10, r4
; CHECK-NEON-NEXT: sbcs r1, r8, #0
; CHECK-NEON-NEXT: mov r1, #0
; CHECK-NEON-NEXT: movwlo r1, #1
; CHECK-NEON-NEXT: subs r7, r6, r4
; CHECK-NEON-NEXT: sbcs r7, r9, #0
; CHECK-NEON-NEXT: movwlo r3, #1
; CHECK-NEON-NEXT: cmp r3, #0
; CHECK-NEON-NEXT: mvnne r3, #0
; CHECK-NEON-NEXT: cmp r1, #0
; CHECK-NEON-NEXT: mvnne r1, #0
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: vdup.32 d19, r1
; CHECK-NEON-NEXT: mvnne r2, #0
; CHECK-NEON-NEXT: vdup.32 d17, r3
; CHECK-NEON-NEXT: vdup.32 d18, r0
; CHECK-NEON-NEXT: vand q10, q4, q9
; CHECK-NEON-NEXT: vdup.32 d16, r2
; CHECK-NEON-NEXT: vand q11, q6, q8
; CHECK-NEON-NEXT: vorn q9, q10, q9
; CHECK-NEON-NEXT: vorn q8, q11, q8
; CHECK-NEON-NEXT: vmovn.i64 d1, q9
; CHECK-NEON-NEXT: vmovn.i64 d0, q8
; CHECK-NEON-NEXT: vpop {d8, d9, d10}
; CHECK-NEON-NEXT: vpop {d12, d13}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
;
; CHECK-FP16-LABEL: utesth_f16i32:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-FP16-NEXT: vpush {d8, d9, d10, d11}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[1]
; CHECK-FP16-NEXT: vorr d8, d0, d0
; CHECK-FP16-NEXT: vmov.u16 r5, d0[3]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixunshfdi
; CHECK-FP16-NEXT: mov r10, r0
; CHECK-FP16-NEXT: vmov.u16 r0, d8[0]
; CHECK-FP16-NEXT: mov r8, r1
; CHECK-FP16-NEXT: vmov.32 d11[0], r10
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixunshfdi
; CHECK-FP16-NEXT: vmov s0, r5
; CHECK-FP16-NEXT: mov r6, r0
; CHECK-FP16-NEXT: mov r7, r1
; CHECK-FP16-NEXT: vmov.32 d10[0], r0
; CHECK-FP16-NEXT: bl __fixunshfdi
; CHECK-FP16-NEXT: mov r5, r0
; CHECK-FP16-NEXT: vmov.u16 r0, d8[2]
; CHECK-FP16-NEXT: mov r9, r1
; CHECK-FP16-NEXT: vmov.32 d9[0], r5
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixunshfdi
; CHECK-FP16-NEXT: mvn r4, #0
; CHECK-FP16-NEXT: subs r2, r6, r4
; CHECK-FP16-NEXT: sbcs r2, r7, #0
; CHECK-FP16-NEXT: vmov.32 d8[0], r0
; CHECK-FP16-NEXT: mov r2, #0
; CHECK-FP16-NEXT: mov r3, #0
; CHECK-FP16-NEXT: movwlo r2, #1
; CHECK-FP16-NEXT: subs r0, r0, r4
; CHECK-FP16-NEXT: sbcs r0, r1, #0
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: movwlo r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: mvnne r0, #0
; CHECK-FP16-NEXT: subs r1, r5, r4
; CHECK-FP16-NEXT: sbcs r1, r9, #0
; CHECK-FP16-NEXT: mov r1, #0
; CHECK-FP16-NEXT: movwlo r1, #1
; CHECK-FP16-NEXT: subs r7, r10, r4
; CHECK-FP16-NEXT: sbcs r7, r8, #0
; CHECK-FP16-NEXT: movwlo r3, #1
; CHECK-FP16-NEXT: cmp r3, #0
; CHECK-FP16-NEXT: mvnne r3, #0
; CHECK-FP16-NEXT: cmp r1, #0
; CHECK-FP16-NEXT: mvnne r1, #0
; CHECK-FP16-NEXT: cmp r2, #0
; CHECK-FP16-NEXT: vdup.32 d19, r1
; CHECK-FP16-NEXT: mvnne r2, #0
; CHECK-FP16-NEXT: vdup.32 d17, r3
; CHECK-FP16-NEXT: vdup.32 d18, r0
; CHECK-FP16-NEXT: vand q10, q4, q9
; CHECK-FP16-NEXT: vdup.32 d16, r2
; CHECK-FP16-NEXT: vand q11, q5, q8
; CHECK-FP16-NEXT: vorn q9, q10, q9
; CHECK-FP16-NEXT: vorn q8, q11, q8
; CHECK-FP16-NEXT: vmovn.i64 d1, q9
; CHECK-FP16-NEXT: vmovn.i64 d0, q8
; CHECK-FP16-NEXT: vpop {d8, d9, d10, d11}
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptoui <4 x half> %x to <4 x i64>
%0 = icmp ult <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
%spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
%conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @ustest_f16i32(<4 x half> %x) {
; CHECK-NEON-LABEL: ustest_f16i32:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11, d12, d13}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11, d12, d13}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s3
; CHECK-NEON-NEXT: vmov.f32 s18, s2
; CHECK-NEON-NEXT: vmov.f32 s20, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: vmov r0, s20
; CHECK-NEON-NEXT: mov r7, r1
; CHECK-NEON-NEXT: vmov r5, s18
; CHECK-NEON-NEXT: vmov r8, s16
; CHECK-NEON-NEXT: vmov.32 d9[0], r6
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: vmov.32 d8[0], r0
; CHECK-NEON-NEXT: mvn r9, #0
; CHECK-NEON-NEXT: subs r0, r0, r9
; CHECK-NEON-NEXT: mov r4, #0
; CHECK-NEON-NEXT: sbcs r0, r1, #0
; CHECK-NEON-NEXT: vmov.32 d9[1], r7
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: vmov.32 d8[1], r1
; CHECK-NEON-NEXT: mvnne r0, #0
; CHECK-NEON-NEXT: subs r1, r6, r9
; CHECK-NEON-NEXT: sbcs r1, r7, #0
; CHECK-NEON-NEXT: mov r1, #0
; CHECK-NEON-NEXT: movwlt r1, #1
; CHECK-NEON-NEXT: cmp r1, #0
; CHECK-NEON-NEXT: mvnne r1, #0
; CHECK-NEON-NEXT: vdup.32 d13, r1
; CHECK-NEON-NEXT: vdup.32 d12, r0
; CHECK-NEON-NEXT: mov r0, r5
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov.i64 q5, #0xffffffff
; CHECK-NEON-NEXT: vbif q4, q5, q6
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov.32 d12[0], r0
; CHECK-NEON-NEXT: mov r0, r8
; CHECK-NEON-NEXT: mov r6, r1
; CHECK-NEON-NEXT: vmov r7, r10, d8
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: subs r2, r5, r9
; CHECK-NEON-NEXT: vmov.32 d13[0], r0
; CHECK-NEON-NEXT: sbcs r2, r6, #0
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: movwlt r2, #1
; CHECK-NEON-NEXT: subs r0, r0, r9
; CHECK-NEON-NEXT: sbcs r0, r1, #0
; CHECK-NEON-NEXT: vmov.32 d13[1], r1
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: mvnne r0, #0
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: vmov.32 d12[1], r6
; CHECK-NEON-NEXT: mvnne r2, #0
; CHECK-NEON-NEXT: vdup.32 d17, r0
; CHECK-NEON-NEXT: rsbs r0, r7, #0
; CHECK-NEON-NEXT: vdup.32 d16, r2
; CHECK-NEON-NEXT: vmov r7, r5, d9
; CHECK-NEON-NEXT: vbsl q8, q6, q5
; CHECK-NEON-NEXT: rscs r0, r10, #0
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: vmov r1, r2, d16
; CHECK-NEON-NEXT: vmov r3, r6, d17
; CHECK-NEON-NEXT: rsbs r1, r1, #0
; CHECK-NEON-NEXT: rscs r1, r2, #0
; CHECK-NEON-NEXT: mov r1, #0
; CHECK-NEON-NEXT: movwlt r1, #1
; CHECK-NEON-NEXT: rsbs r2, r3, #0
; CHECK-NEON-NEXT: rscs r2, r6, #0
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: movwlt r2, #1
; CHECK-NEON-NEXT: rsbs r3, r7, #0
; CHECK-NEON-NEXT: rscs r3, r5, #0
; CHECK-NEON-NEXT: movwlt r4, #1
; CHECK-NEON-NEXT: cmp r4, #0
; CHECK-NEON-NEXT: mvnne r4, #0
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: mvnne r2, #0
; CHECK-NEON-NEXT: cmp r1, #0
; CHECK-NEON-NEXT: mvnne r1, #0
; CHECK-NEON-NEXT: vmov.32 d21[0], r2
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: vmov.32 d20[0], r1
; CHECK-NEON-NEXT: mvnne r0, #0
; CHECK-NEON-NEXT: vmov.32 d19[0], r4
; CHECK-NEON-NEXT: vand q8, q8, q10
; CHECK-NEON-NEXT: vmov.32 d18[0], r0
; CHECK-NEON-NEXT: vmovn.i64 d1, q8
; CHECK-NEON-NEXT: vand q9, q4, q9
; CHECK-NEON-NEXT: vmovn.i64 d0, q9
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11, d12, d13}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
;
; CHECK-FP16-LABEL: ustest_f16i32:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-FP16-NEXT: .vsave {d10, d11, d12, d13, d14, d15}
; CHECK-FP16-NEXT: vpush {d10, d11, d12, d13, d14, d15}
; CHECK-FP16-NEXT: .vsave {d8}
; CHECK-FP16-NEXT: vpush {d8}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[1]
; CHECK-FP16-NEXT: vorr d8, d0, d0
; CHECK-FP16-NEXT: vmov.u16 r8, d0[2]
; CHECK-FP16-NEXT: vmov.u16 r9, d0[3]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: mov r4, r0
; CHECK-FP16-NEXT: vmov.u16 r0, d8[0]
; CHECK-FP16-NEXT: mov r5, r1
; CHECK-FP16-NEXT: vmov.32 d11[0], r4
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: vmov.32 d10[0], r0
; CHECK-FP16-NEXT: mvn r7, #0
; CHECK-FP16-NEXT: subs r0, r0, r7
; CHECK-FP16-NEXT: vmov.i64 q6, #0xffffffff
; CHECK-FP16-NEXT: sbcs r0, r1, #0
; CHECK-FP16-NEXT: vmov.32 d11[1], r5
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: vmov s0, r8
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: vmov.32 d10[1], r1
; CHECK-FP16-NEXT: mvnne r0, #0
; CHECK-FP16-NEXT: subs r1, r4, r7
; CHECK-FP16-NEXT: mov r6, #0
; CHECK-FP16-NEXT: sbcs r1, r5, #0
; CHECK-FP16-NEXT: vmov s16, r9
; CHECK-FP16-NEXT: mov r1, #0
; CHECK-FP16-NEXT: movwlt r1, #1
; CHECK-FP16-NEXT: cmp r1, #0
; CHECK-FP16-NEXT: mvnne r1, #0
; CHECK-FP16-NEXT: vdup.32 d17, r1
; CHECK-FP16-NEXT: vdup.32 d16, r0
; CHECK-FP16-NEXT: vbif q5, q6, q8
; CHECK-FP16-NEXT: vmov r9, r8, d10
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: vmov.f32 s0, s16
; CHECK-FP16-NEXT: mov r4, r0
; CHECK-FP16-NEXT: mov r5, r1
; CHECK-FP16-NEXT: vmov.32 d14[0], r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: subs r2, r4, r7
; CHECK-FP16-NEXT: vmov.32 d15[0], r0
; CHECK-FP16-NEXT: sbcs r2, r5, #0
; CHECK-FP16-NEXT: mov r2, #0
; CHECK-FP16-NEXT: movwlt r2, #1
; CHECK-FP16-NEXT: subs r0, r0, r7
; CHECK-FP16-NEXT: sbcs r0, r1, #0
; CHECK-FP16-NEXT: vmov.32 d15[1], r1
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: mvnne r0, #0
; CHECK-FP16-NEXT: cmp r2, #0
; CHECK-FP16-NEXT: vmov.32 d14[1], r5
; CHECK-FP16-NEXT: mvnne r2, #0
; CHECK-FP16-NEXT: vmov r5, r4, d11
; CHECK-FP16-NEXT: vdup.32 d17, r0
; CHECK-FP16-NEXT: rsbs r0, r9, #0
; CHECK-FP16-NEXT: vdup.32 d16, r2
; CHECK-FP16-NEXT: rscs r0, r8, #0
; CHECK-FP16-NEXT: vbsl q8, q7, q6
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: vmov r1, r2, d16
; CHECK-FP16-NEXT: vmov r3, r7, d17
; CHECK-FP16-NEXT: rsbs r1, r1, #0
; CHECK-FP16-NEXT: rscs r1, r2, #0
; CHECK-FP16-NEXT: mov r1, #0
; CHECK-FP16-NEXT: movwlt r1, #1
; CHECK-FP16-NEXT: rsbs r2, r3, #0
; CHECK-FP16-NEXT: rscs r2, r7, #0
; CHECK-FP16-NEXT: mov r2, #0
; CHECK-FP16-NEXT: movwlt r2, #1
; CHECK-FP16-NEXT: rsbs r3, r5, #0
; CHECK-FP16-NEXT: rscs r3, r4, #0
; CHECK-FP16-NEXT: movwlt r6, #1
; CHECK-FP16-NEXT: cmp r6, #0
; CHECK-FP16-NEXT: mvnne r6, #0
; CHECK-FP16-NEXT: cmp r2, #0
; CHECK-FP16-NEXT: mvnne r2, #0
; CHECK-FP16-NEXT: cmp r1, #0
; CHECK-FP16-NEXT: mvnne r1, #0
; CHECK-FP16-NEXT: vmov.32 d21[0], r2
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: vmov.32 d20[0], r1
; CHECK-FP16-NEXT: mvnne r0, #0
; CHECK-FP16-NEXT: vmov.32 d19[0], r6
; CHECK-FP16-NEXT: vand q8, q8, q10
; CHECK-FP16-NEXT: vmov.32 d18[0], r0
; CHECK-FP16-NEXT: vmovn.i64 d1, q8
; CHECK-FP16-NEXT: vand q9, q5, q9
; CHECK-FP16-NEXT: vmovn.i64 d0, q9
; CHECK-FP16-NEXT: vpop {d8}
; CHECK-FP16-NEXT: vpop {d10, d11, d12, d13, d14, d15}
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, r9, r11, pc}
entry:
%conv = fptosi <4 x half> %x to <4 x i64>
%0 = icmp slt <4 x i64> %conv, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
%spec.store.select = select <4 x i1> %0, <4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
%1 = icmp sgt <4 x i64> %spec.store.select, zeroinitializer
%spec.store.select7 = select <4 x i1> %1, <4 x i64> %spec.store.select, <4 x i64> zeroinitializer
%conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
ret <4 x i32> %conv6
}
; i16 saturate
define <2 x i16> @stest_f64i16(<2 x double> %x) {
; CHECK-LABEL: stest_f64i16:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.s32.f64 s4, d0
; CHECK-NEXT: vmov r0, s4
; CHECK-NEXT: vcvt.s32.f64 s0, d1
; CHECK-NEXT: vmov.i32 d17, #0x7fff
; CHECK-NEXT: vmvn.i32 d18, #0x7fff
; CHECK-NEXT: vmov.32 d16[0], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: vmov.32 d16[1], r0
; CHECK-NEXT: vmin.s32 d16, d16, d17
; CHECK-NEXT: vmax.s32 d0, d16, d18
; CHECK-NEXT: bx lr
entry:
%conv = fptosi <2 x double> %x to <2 x i32>
%0 = icmp slt <2 x i32> %conv, <i32 32767, i32 32767>
%spec.store.select = select <2 x i1> %0, <2 x i32> %conv, <2 x i32> <i32 32767, i32 32767>
%1 = icmp sgt <2 x i32> %spec.store.select, <i32 -32768, i32 -32768>
%spec.store.select7 = select <2 x i1> %1, <2 x i32> %spec.store.select, <2 x i32> <i32 -32768, i32 -32768>
%conv6 = trunc <2 x i32> %spec.store.select7 to <2 x i16>
ret <2 x i16> %conv6
}
define <2 x i16> @utest_f64i16(<2 x double> %x) {
; CHECK-LABEL: utest_f64i16:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.u32.f64 s4, d0
; CHECK-NEXT: vmov r0, s4
; CHECK-NEXT: vcvt.u32.f64 s0, d1
; CHECK-NEXT: vmov.i32 d17, #0xffff
; CHECK-NEXT: vmov.32 d16[0], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: vmov.32 d16[1], r0
; CHECK-NEXT: vmin.u32 d0, d16, d17
; CHECK-NEXT: bx lr
entry:
%conv = fptoui <2 x double> %x to <2 x i32>
%0 = icmp ult <2 x i32> %conv, <i32 65535, i32 65535>
%spec.store.select = select <2 x i1> %0, <2 x i32> %conv, <2 x i32> <i32 65535, i32 65535>
%conv6 = trunc <2 x i32> %spec.store.select to <2 x i16>
ret <2 x i16> %conv6
}
define <2 x i16> @ustest_f64i16(<2 x double> %x) {
; CHECK-LABEL: ustest_f64i16:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.s32.f64 s4, d0
; CHECK-NEXT: vmov r0, s4
; CHECK-NEXT: vcvt.s32.f64 s0, d1
; CHECK-NEXT: vmov.i32 d17, #0xffff
; CHECK-NEXT: vmov.i32 d18, #0x0
; CHECK-NEXT: vmov.32 d16[0], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: vmov.32 d16[1], r0
; CHECK-NEXT: vmin.s32 d16, d16, d17
; CHECK-NEXT: vmax.s32 d0, d16, d18
; CHECK-NEXT: bx lr
entry:
%conv = fptosi <2 x double> %x to <2 x i32>
%0 = icmp slt <2 x i32> %conv, <i32 65535, i32 65535>
%spec.store.select = select <2 x i1> %0, <2 x i32> %conv, <2 x i32> <i32 65535, i32 65535>
%1 = icmp sgt <2 x i32> %spec.store.select, zeroinitializer
%spec.store.select7 = select <2 x i1> %1, <2 x i32> %spec.store.select, <2 x i32> zeroinitializer
%conv6 = trunc <2 x i32> %spec.store.select7 to <2 x i16>
ret <2 x i16> %conv6
}
define <4 x i16> @stest_f32i16(<4 x float> %x) {
; CHECK-LABEL: stest_f32i16:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.s32.f32 q8, q0
; CHECK-NEXT: vmov.i32 q9, #0x7fff
; CHECK-NEXT: vmvn.i32 q10, #0x7fff
; CHECK-NEXT: vmin.s32 q8, q8, q9
; CHECK-NEXT: vmax.s32 q8, q8, q10
; CHECK-NEXT: vmovn.i32 d0, q8
; CHECK-NEXT: bx lr
entry:
%conv = fptosi <4 x float> %x to <4 x i32>
%0 = icmp slt <4 x i32> %conv, <i32 32767, i32 32767, i32 32767, i32 32767>
%spec.store.select = select <4 x i1> %0, <4 x i32> %conv, <4 x i32> <i32 32767, i32 32767, i32 32767, i32 32767>
%1 = icmp sgt <4 x i32> %spec.store.select, <i32 -32768, i32 -32768, i32 -32768, i32 -32768>
%spec.store.select7 = select <4 x i1> %1, <4 x i32> %spec.store.select, <4 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768>
%conv6 = trunc <4 x i32> %spec.store.select7 to <4 x i16>
ret <4 x i16> %conv6
}
define <4 x i16> @utest_f32i16(<4 x float> %x) {
; CHECK-LABEL: utest_f32i16:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.u32.f32 q8, q0
; CHECK-NEXT: vmov.i32 q9, #0xffff
; CHECK-NEXT: vmin.u32 q8, q8, q9
; CHECK-NEXT: vmovn.i32 d0, q8
; CHECK-NEXT: bx lr
entry:
%conv = fptoui <4 x float> %x to <4 x i32>
%0 = icmp ult <4 x i32> %conv, <i32 65535, i32 65535, i32 65535, i32 65535>
%spec.store.select = select <4 x i1> %0, <4 x i32> %conv, <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>
%conv6 = trunc <4 x i32> %spec.store.select to <4 x i16>
ret <4 x i16> %conv6
}
define <4 x i16> @ustest_f32i16(<4 x float> %x) {
; CHECK-LABEL: ustest_f32i16:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.s32.f32 q8, q0
; CHECK-NEXT: vmov.i32 q9, #0xffff
; CHECK-NEXT: vmov.i32 q10, #0x0
; CHECK-NEXT: vmin.s32 q8, q8, q9
; CHECK-NEXT: vmax.s32 q8, q8, q10
; CHECK-NEXT: vmovn.i32 d0, q8
; CHECK-NEXT: bx lr
entry:
%conv = fptosi <4 x float> %x to <4 x i32>
%0 = icmp slt <4 x i32> %conv, <i32 65535, i32 65535, i32 65535, i32 65535>
%spec.store.select = select <4 x i1> %0, <4 x i32> %conv, <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>
%1 = icmp sgt <4 x i32> %spec.store.select, zeroinitializer
%spec.store.select7 = select <4 x i1> %1, <4 x i32> %spec.store.select, <4 x i32> zeroinitializer
%conv6 = trunc <4 x i32> %spec.store.select7 to <4 x i16>
ret <4 x i16> %conv6
}
define <8 x i16> @stest_f16i16(<8 x half> %x) {
; CHECK-NEON-LABEL: stest_f16i16:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s7
; CHECK-NEON-NEXT: vmov.f32 s18, s6
; CHECK-NEON-NEXT: vmov.f32 s20, s5
; CHECK-NEON-NEXT: vmov.f32 s22, s4
; CHECK-NEON-NEXT: vmov.f32 s24, s3
; CHECK-NEON-NEXT: vmov.f32 s26, s2
; CHECK-NEON-NEXT: vmov.f32 s28, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r4, r0
; CHECK-NEON-NEXT: vmov r0, s26
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s22
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: vmov r0, s24
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r7, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[0], r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov s22, r7
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s30, r6
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[1], r0
; CHECK-NEON-NEXT: vmov r0, s28
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r1, s20
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r5
; CHECK-NEON-NEXT: vcvt.s32.f32 s20, s2
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s30
; CHECK-NEON-NEXT: vmov.32 d8[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d12[0], r0
; CHECK-NEON-NEXT: mov r0, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r0, s20
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r4
; CHECK-NEON-NEXT: vmov.i32 q8, #0x7fff
; CHECK-NEON-NEXT: vcvt.s32.f32 s2, s2
; CHECK-NEON-NEXT: vmvn.i32 q9, #0x7fff
; CHECK-NEON-NEXT: vmov.32 d9[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s22
; CHECK-NEON-NEXT: vmov.32 d12[1], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmin.s32 q10, q6, q8
; CHECK-NEON-NEXT: vmax.s32 q10, q10, q9
; CHECK-NEON-NEXT: vmov.32 d9[1], r0
; CHECK-NEON-NEXT: vmov r0, s2
; CHECK-NEON-NEXT: vmovn.i32 d1, q10
; CHECK-NEON-NEXT: vmov.32 d8[1], r0
; CHECK-NEON-NEXT: vmin.s32 q8, q4, q8
; CHECK-NEON-NEXT: vmax.s32 q8, q8, q9
; CHECK-NEON-NEXT: vmovn.i32 d0, q8
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: stest_f16i16:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: vmovx.f16 s4, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s12, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s0, s3
; CHECK-FP16-NEXT: vcvt.s32.f16 s5, s2
; CHECK-FP16-NEXT: vmov r0, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s14, s1
; CHECK-FP16-NEXT: vmovx.f16 s10, s3
; CHECK-FP16-NEXT: vmovx.f16 s8, s2
; CHECK-FP16-NEXT: vcvt.s32.f16 s10, s10
; CHECK-FP16-NEXT: vcvt.s32.f16 s8, s8
; CHECK-FP16-NEXT: vmovx.f16 s6, s1
; CHECK-FP16-NEXT: vcvt.s32.f16 s4, s4
; CHECK-FP16-NEXT: vcvt.s32.f16 s6, s6
; CHECK-FP16-NEXT: vmov.i32 q10, #0x7fff
; CHECK-FP16-NEXT: vmvn.i32 q11, #0x7fff
; CHECK-FP16-NEXT: vmov.32 d17[0], r0
; CHECK-FP16-NEXT: vmov r0, s5
; CHECK-FP16-NEXT: vmov.32 d16[0], r0
; CHECK-FP16-NEXT: vmov r0, s14
; CHECK-FP16-NEXT: vmov.32 d19[0], r0
; CHECK-FP16-NEXT: vmov r0, s12
; CHECK-FP16-NEXT: vmov.32 d18[0], r0
; CHECK-FP16-NEXT: vmov r0, s10
; CHECK-FP16-NEXT: vmov.32 d17[1], r0
; CHECK-FP16-NEXT: vmov r0, s8
; CHECK-FP16-NEXT: vmov.32 d16[1], r0
; CHECK-FP16-NEXT: vmov r0, s6
; CHECK-FP16-NEXT: vmin.s32 q8, q8, q10
; CHECK-FP16-NEXT: vmax.s32 q8, q8, q11
; CHECK-FP16-NEXT: vmovn.i32 d1, q8
; CHECK-FP16-NEXT: vmov.32 d19[1], r0
; CHECK-FP16-NEXT: vmov r0, s4
; CHECK-FP16-NEXT: vmov.32 d18[1], r0
; CHECK-FP16-NEXT: vmin.s32 q9, q9, q10
; CHECK-FP16-NEXT: vmax.s32 q9, q9, q11
; CHECK-FP16-NEXT: vmovn.i32 d0, q9
; CHECK-FP16-NEXT: bx lr
entry:
%conv = fptosi <8 x half> %x to <8 x i32>
%0 = icmp slt <8 x i32> %conv, <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
%spec.store.select = select <8 x i1> %0, <8 x i32> %conv, <8 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>
%1 = icmp sgt <8 x i32> %spec.store.select, <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
%spec.store.select7 = select <8 x i1> %1, <8 x i32> %spec.store.select, <8 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>
%conv6 = trunc <8 x i32> %spec.store.select7 to <8 x i16>
ret <8 x i16> %conv6
}
define <8 x i16> @utesth_f16i16(<8 x half> %x) {
; CHECK-NEON-LABEL: utesth_f16i16:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11, d12, d13, d14}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11, d12, d13, d14}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s7
; CHECK-NEON-NEXT: vmov.f32 s18, s6
; CHECK-NEON-NEXT: vmov.f32 s20, s5
; CHECK-NEON-NEXT: vmov.f32 s22, s4
; CHECK-NEON-NEXT: vmov.f32 s24, s3
; CHECK-NEON-NEXT: vmov.f32 s26, s2
; CHECK-NEON-NEXT: vmov.f32 s28, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r4, r0
; CHECK-NEON-NEXT: vmov r0, s26
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s22
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: vmov r0, s24
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r7, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[0], r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov s16, r7
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s18, r6
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[1], r0
; CHECK-NEON-NEXT: vmov r0, s28
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r1, s20
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r5
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s18
; CHECK-NEON-NEXT: vcvt.u32.f32 s18, s2
; CHECK-NEON-NEXT: vmov.32 d10[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d12[0], r0
; CHECK-NEON-NEXT: mov r0, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r4
; CHECK-NEON-NEXT: vmov.i32 q8, #0xffff
; CHECK-NEON-NEXT: vcvt.u32.f32 s2, s2
; CHECK-NEON-NEXT: vmov.32 d11[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s16
; CHECK-NEON-NEXT: vmov.32 d12[1], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmin.u32 q9, q6, q8
; CHECK-NEON-NEXT: vmov.32 d11[1], r0
; CHECK-NEON-NEXT: vmov r0, s2
; CHECK-NEON-NEXT: vmovn.i32 d1, q9
; CHECK-NEON-NEXT: vmov.32 d10[1], r0
; CHECK-NEON-NEXT: vmin.u32 q8, q5, q8
; CHECK-NEON-NEXT: vmovn.i32 d0, q8
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11, d12, d13, d14}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: utesth_f16i16:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: vmovx.f16 s4, s0
; CHECK-FP16-NEXT: vcvt.u32.f16 s12, s0
; CHECK-FP16-NEXT: vcvt.u32.f16 s0, s3
; CHECK-FP16-NEXT: vcvt.u32.f16 s5, s2
; CHECK-FP16-NEXT: vmov r0, s0
; CHECK-FP16-NEXT: vcvt.u32.f16 s14, s1
; CHECK-FP16-NEXT: vmovx.f16 s10, s3
; CHECK-FP16-NEXT: vmovx.f16 s8, s2
; CHECK-FP16-NEXT: vcvt.u32.f16 s10, s10
; CHECK-FP16-NEXT: vcvt.u32.f16 s8, s8
; CHECK-FP16-NEXT: vmovx.f16 s6, s1
; CHECK-FP16-NEXT: vcvt.u32.f16 s4, s4
; CHECK-FP16-NEXT: vcvt.u32.f16 s6, s6
; CHECK-FP16-NEXT: vmov.i32 q10, #0xffff
; CHECK-FP16-NEXT: vmov.32 d17[0], r0
; CHECK-FP16-NEXT: vmov r0, s5
; CHECK-FP16-NEXT: vmov.32 d16[0], r0
; CHECK-FP16-NEXT: vmov r0, s14
; CHECK-FP16-NEXT: vmov.32 d19[0], r0
; CHECK-FP16-NEXT: vmov r0, s12
; CHECK-FP16-NEXT: vmov.32 d18[0], r0
; CHECK-FP16-NEXT: vmov r0, s10
; CHECK-FP16-NEXT: vmov.32 d17[1], r0
; CHECK-FP16-NEXT: vmov r0, s8
; CHECK-FP16-NEXT: vmov.32 d16[1], r0
; CHECK-FP16-NEXT: vmov r0, s6
; CHECK-FP16-NEXT: vmin.u32 q8, q8, q10
; CHECK-FP16-NEXT: vmovn.i32 d1, q8
; CHECK-FP16-NEXT: vmov.32 d19[1], r0
; CHECK-FP16-NEXT: vmov r0, s4
; CHECK-FP16-NEXT: vmov.32 d18[1], r0
; CHECK-FP16-NEXT: vmin.u32 q9, q9, q10
; CHECK-FP16-NEXT: vmovn.i32 d0, q9
; CHECK-FP16-NEXT: bx lr
entry:
%conv = fptoui <8 x half> %x to <8 x i32>
%0 = icmp ult <8 x i32> %conv, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
%spec.store.select = select <8 x i1> %0, <8 x i32> %conv, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
%conv6 = trunc <8 x i32> %spec.store.select to <8 x i16>
ret <8 x i16> %conv6
}
define <8 x i16> @ustest_f16i16(<8 x half> %x) {
; CHECK-NEON-LABEL: ustest_f16i16:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s7
; CHECK-NEON-NEXT: vmov.f32 s18, s6
; CHECK-NEON-NEXT: vmov.f32 s20, s5
; CHECK-NEON-NEXT: vmov.f32 s22, s4
; CHECK-NEON-NEXT: vmov.f32 s24, s3
; CHECK-NEON-NEXT: vmov.f32 s26, s2
; CHECK-NEON-NEXT: vmov.f32 s28, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r4, r0
; CHECK-NEON-NEXT: vmov r0, s26
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s22
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: vmov r0, s24
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r7, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[0], r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov s22, r7
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s30, r6
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[1], r0
; CHECK-NEON-NEXT: vmov r0, s28
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r1, s20
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r5
; CHECK-NEON-NEXT: vcvt.s32.f32 s20, s2
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s30
; CHECK-NEON-NEXT: vmov.32 d8[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d12[0], r0
; CHECK-NEON-NEXT: mov r0, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r0, s20
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r4
; CHECK-NEON-NEXT: vmov.i32 q8, #0xffff
; CHECK-NEON-NEXT: vcvt.s32.f32 s2, s2
; CHECK-NEON-NEXT: vmov.i32 q9, #0x0
; CHECK-NEON-NEXT: vmov.32 d9[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s22
; CHECK-NEON-NEXT: vmov.32 d12[1], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmin.s32 q10, q6, q8
; CHECK-NEON-NEXT: vmax.s32 q10, q10, q9
; CHECK-NEON-NEXT: vmov.32 d9[1], r0
; CHECK-NEON-NEXT: vmov r0, s2
; CHECK-NEON-NEXT: vmovn.i32 d1, q10
; CHECK-NEON-NEXT: vmov.32 d8[1], r0
; CHECK-NEON-NEXT: vmin.s32 q8, q4, q8
; CHECK-NEON-NEXT: vmax.s32 q8, q8, q9
; CHECK-NEON-NEXT: vmovn.i32 d0, q8
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: ustest_f16i16:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: vmovx.f16 s4, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s12, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s0, s3
; CHECK-FP16-NEXT: vcvt.s32.f16 s5, s2
; CHECK-FP16-NEXT: vmov r0, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s14, s1
; CHECK-FP16-NEXT: vmovx.f16 s10, s3
; CHECK-FP16-NEXT: vmovx.f16 s8, s2
; CHECK-FP16-NEXT: vcvt.s32.f16 s10, s10
; CHECK-FP16-NEXT: vcvt.s32.f16 s8, s8
; CHECK-FP16-NEXT: vmovx.f16 s6, s1
; CHECK-FP16-NEXT: vcvt.s32.f16 s4, s4
; CHECK-FP16-NEXT: vcvt.s32.f16 s6, s6
; CHECK-FP16-NEXT: vmov.i32 q10, #0xffff
; CHECK-FP16-NEXT: vmov.i32 q11, #0x0
; CHECK-FP16-NEXT: vmov.32 d17[0], r0
; CHECK-FP16-NEXT: vmov r0, s5
; CHECK-FP16-NEXT: vmov.32 d16[0], r0
; CHECK-FP16-NEXT: vmov r0, s14
; CHECK-FP16-NEXT: vmov.32 d19[0], r0
; CHECK-FP16-NEXT: vmov r0, s12
; CHECK-FP16-NEXT: vmov.32 d18[0], r0
; CHECK-FP16-NEXT: vmov r0, s10
; CHECK-FP16-NEXT: vmov.32 d17[1], r0
; CHECK-FP16-NEXT: vmov r0, s8
; CHECK-FP16-NEXT: vmov.32 d16[1], r0
; CHECK-FP16-NEXT: vmov r0, s6
; CHECK-FP16-NEXT: vmin.s32 q8, q8, q10
; CHECK-FP16-NEXT: vmax.s32 q8, q8, q11
; CHECK-FP16-NEXT: vmovn.i32 d1, q8
; CHECK-FP16-NEXT: vmov.32 d19[1], r0
; CHECK-FP16-NEXT: vmov r0, s4
; CHECK-FP16-NEXT: vmov.32 d18[1], r0
; CHECK-FP16-NEXT: vmin.s32 q9, q9, q10
; CHECK-FP16-NEXT: vmax.s32 q9, q9, q11
; CHECK-FP16-NEXT: vmovn.i32 d0, q9
; CHECK-FP16-NEXT: bx lr
entry:
%conv = fptosi <8 x half> %x to <8 x i32>
%0 = icmp slt <8 x i32> %conv, <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
%spec.store.select = select <8 x i1> %0, <8 x i32> %conv, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
%1 = icmp sgt <8 x i32> %spec.store.select, zeroinitializer
%spec.store.select7 = select <8 x i1> %1, <8 x i32> %spec.store.select, <8 x i32> zeroinitializer
%conv6 = trunc <8 x i32> %spec.store.select7 to <8 x i16>
ret <8 x i16> %conv6
}
; i64 saturate
define <2 x i64> @stest_f64i64(<2 x double> %x) {
; CHECK-LABEL: stest_f64i64:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vorr d0, d9, d9
; CHECK-NEXT: bl __fixdfti
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: mvn r8, #0
; CHECK-NEXT: subs r0, r0, r8
; CHECK-NEXT: mvn r6, #-2147483648
; CHECK-NEXT: sbcs r0, r1, r6
; CHECK-NEXT: mov r10, r1
; CHECK-NEXT: sbcs r0, r2, #0
; CHECK-NEXT: vorr d0, d8, d8
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: mov r9, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: moveq r3, r0
; CHECK-NEXT: movne r0, r2
; CHECK-NEXT: moveq r10, r6
; CHECK-NEXT: moveq r5, r8
; CHECK-NEXT: rsbs r1, r5, #0
; CHECK-NEXT: rscs r1, r10, #-2147483648
; CHECK-NEXT: sbcs r0, r8, r0
; CHECK-NEXT: sbcs r0, r8, r3
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: bl __fixdfti
; CHECK-NEXT: subs r4, r0, r8
; CHECK-NEXT: vmov.32 d1[0], r5
; CHECK-NEXT: sbcs r4, r1, r6
; CHECK-NEXT: sbcs r4, r2, #0
; CHECK-NEXT: sbcs r4, r3, #0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: moveq r3, r4
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: movne r4, r2
; CHECK-NEXT: moveq r0, r8
; CHECK-NEXT: rsbs r1, r0, #0
; CHECK-NEXT: rscs r1, r6, #-2147483648
; CHECK-NEXT: sbcs r1, r8, r4
; CHECK-NEXT: sbcs r1, r8, r3
; CHECK-NEXT: movwlt r9, #1
; CHECK-NEXT: cmp r9, #0
; CHECK-NEXT: moveq r0, r9
; CHECK-NEXT: mov r1, #-2147483648
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: vmov.32 d0[0], r0
; CHECK-NEXT: moveq r10, r1
; CHECK-NEXT: cmp r9, #0
; CHECK-NEXT: vmov.32 d1[1], r10
; CHECK-NEXT: moveq r6, r1
; CHECK-NEXT: vmov.32 d0[1], r6
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <2 x double> %x to <2 x i128>
%0 = icmp slt <2 x i128> %conv, <i128 9223372036854775807, i128 9223372036854775807>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 9223372036854775807, i128 9223372036854775807>
%1 = icmp sgt <2 x i128> %spec.store.select, <i128 -9223372036854775808, i128 -9223372036854775808>
%spec.store.select7 = select <2 x i1> %1, <2 x i128> %spec.store.select, <2 x i128> <i128 -9223372036854775808, i128 -9223372036854775808>
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @utest_f64i64(<2 x double> %x) {
; CHECK-LABEL: utest_f64i64:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vorr d0, d9, d9
; CHECK-NEXT: bl __fixunsdfti
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: subs r0, r2, #1
; CHECK-NEXT: vorr d0, d8, d8
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: movwlo r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: bl __fixunsdfti
; CHECK-NEXT: subs r2, r2, #1
; CHECK-NEXT: vmov.32 d1[0], r5
; CHECK-NEXT: sbcs r2, r3, #0
; CHECK-NEXT: movwlo r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: moveq r0, r6
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: movne r7, r4
; CHECK-NEXT: vmov.32 d0[0], r0
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: vmov.32 d1[1], r7
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: vmov.32 d0[1], r6
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r11, pc}
entry:
%conv = fptoui <2 x double> %x to <2 x i128>
%0 = icmp ult <2 x i128> %conv, <i128 18446744073709551616, i128 18446744073709551616>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>
%conv6 = trunc <2 x i128> %spec.store.select to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @ustest_f64i64(<2 x double> %x) {
; CHECK-LABEL: ustest_f64i64:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vorr d0, d9, d9
; CHECK-NEXT: bl __fixdfti
; CHECK-NEXT: mov r8, r1
; CHECK-NEXT: subs r1, r2, #1
; CHECK-NEXT: sbcs r1, r3, #0
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: movwlt r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: mov r9, #1
; CHECK-NEXT: moveq r3, r6
; CHECK-NEXT: moveq r8, r6
; CHECK-NEXT: moveq r2, r9
; CHECK-NEXT: movne r6, r0
; CHECK-NEXT: rsbs r0, r6, #0
; CHECK-NEXT: rscs r0, r8, #0
; CHECK-NEXT: vorr d0, d8, d8
; CHECK-NEXT: rscs r0, r2, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: rscs r0, r3, #0
; CHECK-NEXT: mov r5, #0
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: moveq r6, r7
; CHECK-NEXT: bl __fixdfti
; CHECK-NEXT: subs r4, r2, #1
; CHECK-NEXT: vmov.32 d1[0], r6
; CHECK-NEXT: sbcs r4, r3, #0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: movne r9, r2
; CHECK-NEXT: moveq r3, r4
; CHECK-NEXT: moveq r1, r4
; CHECK-NEXT: movne r4, r0
; CHECK-NEXT: rsbs r0, r4, #0
; CHECK-NEXT: rscs r0, r1, #0
; CHECK-NEXT: rscs r0, r9, #0
; CHECK-NEXT: rscs r0, r3, #0
; CHECK-NEXT: movwlt r5, #1
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: moveq r4, r5
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: movne r7, r8
; CHECK-NEXT: vmov.32 d0[0], r4
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: vmov.32 d1[1], r7
; CHECK-NEXT: movne r5, r1
; CHECK-NEXT: vmov.32 d0[1], r5
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r11, pc}
entry:
%conv = fptosi <2 x double> %x to <2 x i128>
%0 = icmp slt <2 x i128> %conv, <i128 18446744073709551616, i128 18446744073709551616>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>
%1 = icmp sgt <2 x i128> %spec.store.select, zeroinitializer
%spec.store.select7 = select <2 x i1> %1, <2 x i128> %spec.store.select, <2 x i128> zeroinitializer
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @stest_f32i64(<2 x float> %x) {
; CHECK-LABEL: stest_f32i64:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: .vsave {d8}
; CHECK-NEXT: vpush {d8}
; CHECK-NEXT: vmov.f64 d8, d0
; CHECK-NEXT: vmov.f32 s0, s17
; CHECK-NEXT: bl __fixsfti
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: mvn r8, #0
; CHECK-NEXT: subs r0, r0, r8
; CHECK-NEXT: mvn r6, #-2147483648
; CHECK-NEXT: sbcs r0, r1, r6
; CHECK-NEXT: vmov.f32 s0, s16
; CHECK-NEXT: sbcs r0, r2, #0
; CHECK-NEXT: mov r10, r1
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: mov r9, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: moveq r3, r0
; CHECK-NEXT: movne r0, r2
; CHECK-NEXT: moveq r10, r6
; CHECK-NEXT: moveq r5, r8
; CHECK-NEXT: rsbs r1, r5, #0
; CHECK-NEXT: rscs r1, r10, #-2147483648
; CHECK-NEXT: sbcs r0, r8, r0
; CHECK-NEXT: sbcs r0, r8, r3
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: bl __fixsfti
; CHECK-NEXT: subs r4, r0, r8
; CHECK-NEXT: vmov.32 d1[0], r5
; CHECK-NEXT: sbcs r4, r1, r6
; CHECK-NEXT: sbcs r4, r2, #0
; CHECK-NEXT: sbcs r4, r3, #0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: moveq r3, r4
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: movne r4, r2
; CHECK-NEXT: moveq r0, r8
; CHECK-NEXT: rsbs r1, r0, #0
; CHECK-NEXT: rscs r1, r6, #-2147483648
; CHECK-NEXT: sbcs r1, r8, r4
; CHECK-NEXT: sbcs r1, r8, r3
; CHECK-NEXT: movwlt r9, #1
; CHECK-NEXT: cmp r9, #0
; CHECK-NEXT: moveq r0, r9
; CHECK-NEXT: mov r1, #-2147483648
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: vmov.32 d0[0], r0
; CHECK-NEXT: moveq r10, r1
; CHECK-NEXT: cmp r9, #0
; CHECK-NEXT: vmov.32 d1[1], r10
; CHECK-NEXT: moveq r6, r1
; CHECK-NEXT: vmov.32 d0[1], r6
; CHECK-NEXT: vpop {d8}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <2 x float> %x to <2 x i128>
%0 = icmp slt <2 x i128> %conv, <i128 9223372036854775807, i128 9223372036854775807>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 9223372036854775807, i128 9223372036854775807>
%1 = icmp sgt <2 x i128> %spec.store.select, <i128 -9223372036854775808, i128 -9223372036854775808>
%spec.store.select7 = select <2 x i1> %1, <2 x i128> %spec.store.select, <2 x i128> <i128 -9223372036854775808, i128 -9223372036854775808>
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @utest_f32i64(<2 x float> %x) {
; CHECK-LABEL: utest_f32i64:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: .vsave {d8}
; CHECK-NEXT: vpush {d8}
; CHECK-NEXT: vmov.f64 d8, d0
; CHECK-NEXT: vmov.f32 s0, s17
; CHECK-NEXT: bl __fixunssfti
; CHECK-NEXT: vmov.f32 s0, s16
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: subs r0, r2, #1
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: movwlo r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: bl __fixunssfti
; CHECK-NEXT: subs r2, r2, #1
; CHECK-NEXT: vmov.32 d1[0], r5
; CHECK-NEXT: sbcs r2, r3, #0
; CHECK-NEXT: movwlo r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: moveq r0, r6
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: movne r7, r4
; CHECK-NEXT: vmov.32 d0[0], r0
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: vmov.32 d1[1], r7
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: vmov.32 d0[1], r6
; CHECK-NEXT: vpop {d8}
; CHECK-NEXT: pop {r4, r5, r6, r7, r11, pc}
entry:
%conv = fptoui <2 x float> %x to <2 x i128>
%0 = icmp ult <2 x i128> %conv, <i128 18446744073709551616, i128 18446744073709551616>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>
%conv6 = trunc <2 x i128> %spec.store.select to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @ustest_f32i64(<2 x float> %x) {
; CHECK-LABEL: ustest_f32i64:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-NEXT: .vsave {d8}
; CHECK-NEXT: vpush {d8}
; CHECK-NEXT: vmov.f64 d8, d0
; CHECK-NEXT: vmov.f32 s0, s17
; CHECK-NEXT: bl __fixsfti
; CHECK-NEXT: mov r8, r1
; CHECK-NEXT: subs r1, r2, #1
; CHECK-NEXT: vmov.f32 s0, s16
; CHECK-NEXT: sbcs r1, r3, #0
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: mov r9, #1
; CHECK-NEXT: movwlt r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: moveq r3, r6
; CHECK-NEXT: moveq r8, r6
; CHECK-NEXT: moveq r2, r9
; CHECK-NEXT: movne r6, r0
; CHECK-NEXT: rsbs r0, r6, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: rscs r0, r8, #0
; CHECK-NEXT: mov r5, #0
; CHECK-NEXT: rscs r0, r2, #0
; CHECK-NEXT: rscs r0, r3, #0
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: moveq r6, r7
; CHECK-NEXT: bl __fixsfti
; CHECK-NEXT: subs r4, r2, #1
; CHECK-NEXT: vmov.32 d1[0], r6
; CHECK-NEXT: sbcs r4, r3, #0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: movne r9, r2
; CHECK-NEXT: moveq r3, r4
; CHECK-NEXT: moveq r1, r4
; CHECK-NEXT: movne r4, r0
; CHECK-NEXT: rsbs r0, r4, #0
; CHECK-NEXT: rscs r0, r1, #0
; CHECK-NEXT: rscs r0, r9, #0
; CHECK-NEXT: rscs r0, r3, #0
; CHECK-NEXT: movwlt r5, #1
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: moveq r4, r5
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: movne r7, r8
; CHECK-NEXT: vmov.32 d0[0], r4
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: vmov.32 d1[1], r7
; CHECK-NEXT: movne r5, r1
; CHECK-NEXT: vmov.32 d0[1], r5
; CHECK-NEXT: vpop {d8}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r11, pc}
entry:
%conv = fptosi <2 x float> %x to <2 x i128>
%0 = icmp slt <2 x i128> %conv, <i128 18446744073709551616, i128 18446744073709551616>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>
%1 = icmp sgt <2 x i128> %spec.store.select, zeroinitializer
%spec.store.select7 = select <2 x i1> %1, <2 x i128> %spec.store.select, <2 x i128> zeroinitializer
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @stest_f16i64(<2 x half> %x) {
; CHECK-NEON-LABEL: stest_f16i64:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEON-NEXT: .pad #4
; CHECK-NEON-NEXT: sub sp, sp, #4
; CHECK-NEON-NEXT: .vsave {d8}
; CHECK-NEON-NEXT: vpush {d8}
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.f32 s16, s1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r8, r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: bl __fixsfti
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: mvn r9, #0
; CHECK-NEON-NEXT: subs r0, r0, r9
; CHECK-NEON-NEXT: mvn r7, #-2147483648
; CHECK-NEON-NEXT: sbcs r0, r1, r7
; CHECK-NEON-NEXT: mov r11, r1
; CHECK-NEON-NEXT: sbcs r0, r2, #0
; CHECK-NEON-NEXT: vmov s0, r8
; CHECK-NEON-NEXT: sbcs r0, r3, #0
; CHECK-NEON-NEXT: mov r6, #0
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: mov r10, #0
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: moveq r3, r0
; CHECK-NEON-NEXT: movne r0, r2
; CHECK-NEON-NEXT: moveq r11, r7
; CHECK-NEON-NEXT: moveq r5, r9
; CHECK-NEON-NEXT: rsbs r1, r5, #0
; CHECK-NEON-NEXT: rscs r1, r11, #-2147483648
; CHECK-NEON-NEXT: sbcs r0, r9, r0
; CHECK-NEON-NEXT: sbcs r0, r9, r3
; CHECK-NEON-NEXT: movwlt r6, #1
; CHECK-NEON-NEXT: cmp r6, #0
; CHECK-NEON-NEXT: moveq r5, r6
; CHECK-NEON-NEXT: bl __fixsfti
; CHECK-NEON-NEXT: subs r4, r0, r9
; CHECK-NEON-NEXT: vmov.32 d1[0], r5
; CHECK-NEON-NEXT: sbcs r4, r1, r7
; CHECK-NEON-NEXT: sbcs r4, r2, #0
; CHECK-NEON-NEXT: sbcs r4, r3, #0
; CHECK-NEON-NEXT: mov r4, #0
; CHECK-NEON-NEXT: movwlt r4, #1
; CHECK-NEON-NEXT: cmp r4, #0
; CHECK-NEON-NEXT: moveq r3, r4
; CHECK-NEON-NEXT: movne r7, r1
; CHECK-NEON-NEXT: movne r4, r2
; CHECK-NEON-NEXT: moveq r0, r9
; CHECK-NEON-NEXT: rsbs r1, r0, #0
; CHECK-NEON-NEXT: rscs r1, r7, #-2147483648
; CHECK-NEON-NEXT: sbcs r1, r9, r4
; CHECK-NEON-NEXT: sbcs r1, r9, r3
; CHECK-NEON-NEXT: movwlt r10, #1
; CHECK-NEON-NEXT: cmp r10, #0
; CHECK-NEON-NEXT: moveq r0, r10
; CHECK-NEON-NEXT: mov r1, #-2147483648
; CHECK-NEON-NEXT: cmp r6, #0
; CHECK-NEON-NEXT: vmov.32 d0[0], r0
; CHECK-NEON-NEXT: moveq r11, r1
; CHECK-NEON-NEXT: cmp r10, #0
; CHECK-NEON-NEXT: vmov.32 d1[1], r11
; CHECK-NEON-NEXT: moveq r7, r1
; CHECK-NEON-NEXT: vmov.32 d0[1], r7
; CHECK-NEON-NEXT: vpop {d8}
; CHECK-NEON-NEXT: add sp, sp, #4
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, r11, pc}
;
; CHECK-FP16-LABEL: stest_f16i64:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[1]
; CHECK-FP16-NEXT: vmov.u16 r7, d0[0]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfti
; CHECK-FP16-NEXT: mov r5, r0
; CHECK-FP16-NEXT: mvn r8, #0
; CHECK-FP16-NEXT: subs r0, r0, r8
; CHECK-FP16-NEXT: mvn r6, #-2147483648
; CHECK-FP16-NEXT: sbcs r0, r1, r6
; CHECK-FP16-NEXT: mov r10, r1
; CHECK-FP16-NEXT: sbcs r0, r2, #0
; CHECK-FP16-NEXT: vmov s0, r7
; CHECK-FP16-NEXT: sbcs r0, r3, #0
; CHECK-FP16-NEXT: mov r7, #0
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: mov r9, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: moveq r3, r0
; CHECK-FP16-NEXT: movne r0, r2
; CHECK-FP16-NEXT: moveq r10, r6
; CHECK-FP16-NEXT: moveq r5, r8
; CHECK-FP16-NEXT: rsbs r1, r5, #0
; CHECK-FP16-NEXT: rscs r1, r10, #-2147483648
; CHECK-FP16-NEXT: sbcs r0, r8, r0
; CHECK-FP16-NEXT: sbcs r0, r8, r3
; CHECK-FP16-NEXT: movwlt r7, #1
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: moveq r5, r7
; CHECK-FP16-NEXT: bl __fixhfti
; CHECK-FP16-NEXT: subs r4, r0, r8
; CHECK-FP16-NEXT: vmov.32 d1[0], r5
; CHECK-FP16-NEXT: sbcs r4, r1, r6
; CHECK-FP16-NEXT: sbcs r4, r2, #0
; CHECK-FP16-NEXT: sbcs r4, r3, #0
; CHECK-FP16-NEXT: mov r4, #0
; CHECK-FP16-NEXT: movwlt r4, #1
; CHECK-FP16-NEXT: cmp r4, #0
; CHECK-FP16-NEXT: moveq r3, r4
; CHECK-FP16-NEXT: movne r6, r1
; CHECK-FP16-NEXT: movne r4, r2
; CHECK-FP16-NEXT: moveq r0, r8
; CHECK-FP16-NEXT: rsbs r1, r0, #0
; CHECK-FP16-NEXT: rscs r1, r6, #-2147483648
; CHECK-FP16-NEXT: sbcs r1, r8, r4
; CHECK-FP16-NEXT: sbcs r1, r8, r3
; CHECK-FP16-NEXT: movwlt r9, #1
; CHECK-FP16-NEXT: cmp r9, #0
; CHECK-FP16-NEXT: moveq r0, r9
; CHECK-FP16-NEXT: mov r1, #-2147483648
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: vmov.32 d0[0], r0
; CHECK-FP16-NEXT: moveq r10, r1
; CHECK-FP16-NEXT: cmp r9, #0
; CHECK-FP16-NEXT: vmov.32 d1[1], r10
; CHECK-FP16-NEXT: moveq r6, r1
; CHECK-FP16-NEXT: vmov.32 d0[1], r6
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <2 x half> %x to <2 x i128>
%0 = icmp slt <2 x i128> %conv, <i128 9223372036854775807, i128 9223372036854775807>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 9223372036854775807, i128 9223372036854775807>
%1 = icmp sgt <2 x i128> %spec.store.select, <i128 -9223372036854775808, i128 -9223372036854775808>
%spec.store.select7 = select <2 x i1> %1, <2 x i128> %spec.store.select, <2 x i128> <i128 -9223372036854775808, i128 -9223372036854775808>
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @utesth_f16i64(<2 x half> %x) {
; CHECK-NEON-LABEL: utesth_f16i64:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8}
; CHECK-NEON-NEXT: vpush {d8}
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.f32 s16, s1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: bl __fixunssfti
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: subs r0, r2, #1
; CHECK-NEON-NEXT: vmov s0, r5
; CHECK-NEON-NEXT: sbcs r0, r3, #0
; CHECK-NEON-NEXT: mov r5, #0
; CHECK-NEON-NEXT: mov r4, r1
; CHECK-NEON-NEXT: movwlo r5, #1
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: mov r7, #0
; CHECK-NEON-NEXT: moveq r6, r5
; CHECK-NEON-NEXT: bl __fixunssfti
; CHECK-NEON-NEXT: subs r2, r2, #1
; CHECK-NEON-NEXT: vmov.32 d1[0], r6
; CHECK-NEON-NEXT: sbcs r2, r3, #0
; CHECK-NEON-NEXT: movwlo r7, #1
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: moveq r0, r7
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: movne r5, r4
; CHECK-NEON-NEXT: vmov.32 d0[0], r0
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: vmov.32 d1[1], r5
; CHECK-NEON-NEXT: movne r7, r1
; CHECK-NEON-NEXT: vmov.32 d0[1], r7
; CHECK-NEON-NEXT: vpop {d8}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: utesth_f16i64:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[1]
; CHECK-FP16-NEXT: vmov.u16 r7, d0[0]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixunshfti
; CHECK-FP16-NEXT: mov r5, r0
; CHECK-FP16-NEXT: subs r0, r2, #1
; CHECK-FP16-NEXT: vmov s0, r7
; CHECK-FP16-NEXT: sbcs r0, r3, #0
; CHECK-FP16-NEXT: mov r7, #0
; CHECK-FP16-NEXT: mov r4, r1
; CHECK-FP16-NEXT: movwlo r7, #1
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: mov r6, #0
; CHECK-FP16-NEXT: moveq r5, r7
; CHECK-FP16-NEXT: bl __fixunshfti
; CHECK-FP16-NEXT: subs r2, r2, #1
; CHECK-FP16-NEXT: vmov.32 d1[0], r5
; CHECK-FP16-NEXT: sbcs r2, r3, #0
; CHECK-FP16-NEXT: movwlo r6, #1
; CHECK-FP16-NEXT: cmp r6, #0
; CHECK-FP16-NEXT: moveq r0, r6
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: movne r7, r4
; CHECK-FP16-NEXT: vmov.32 d0[0], r0
; CHECK-FP16-NEXT: cmp r6, #0
; CHECK-FP16-NEXT: vmov.32 d1[1], r7
; CHECK-FP16-NEXT: movne r6, r1
; CHECK-FP16-NEXT: vmov.32 d0[1], r6
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r11, pc}
entry:
%conv = fptoui <2 x half> %x to <2 x i128>
%0 = icmp ult <2 x i128> %conv, <i128 18446744073709551616, i128 18446744073709551616>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>
%conv6 = trunc <2 x i128> %spec.store.select to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @ustest_f16i64(<2 x half> %x) {
; CHECK-NEON-LABEL: ustest_f16i64:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8}
; CHECK-NEON-NEXT: vpush {d8}
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.f32 s16, s1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: bl __fixsfti
; CHECK-NEON-NEXT: mov r8, r1
; CHECK-NEON-NEXT: subs r1, r2, #1
; CHECK-NEON-NEXT: vmov s0, r5
; CHECK-NEON-NEXT: sbcs r1, r3, #0
; CHECK-NEON-NEXT: mov r5, #0
; CHECK-NEON-NEXT: mov r9, #1
; CHECK-NEON-NEXT: movwlt r5, #1
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: moveq r3, r5
; CHECK-NEON-NEXT: moveq r8, r5
; CHECK-NEON-NEXT: moveq r2, r9
; CHECK-NEON-NEXT: movne r5, r0
; CHECK-NEON-NEXT: rsbs r0, r5, #0
; CHECK-NEON-NEXT: mov r7, #0
; CHECK-NEON-NEXT: rscs r0, r8, #0
; CHECK-NEON-NEXT: mov r6, #0
; CHECK-NEON-NEXT: rscs r0, r2, #0
; CHECK-NEON-NEXT: rscs r0, r3, #0
; CHECK-NEON-NEXT: movwlt r7, #1
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: moveq r5, r7
; CHECK-NEON-NEXT: bl __fixsfti
; CHECK-NEON-NEXT: subs r4, r2, #1
; CHECK-NEON-NEXT: vmov.32 d1[0], r5
; CHECK-NEON-NEXT: sbcs r4, r3, #0
; CHECK-NEON-NEXT: mov r4, #0
; CHECK-NEON-NEXT: movwlt r4, #1
; CHECK-NEON-NEXT: cmp r4, #0
; CHECK-NEON-NEXT: movne r9, r2
; CHECK-NEON-NEXT: moveq r3, r4
; CHECK-NEON-NEXT: moveq r1, r4
; CHECK-NEON-NEXT: movne r4, r0
; CHECK-NEON-NEXT: rsbs r0, r4, #0
; CHECK-NEON-NEXT: rscs r0, r1, #0
; CHECK-NEON-NEXT: rscs r0, r9, #0
; CHECK-NEON-NEXT: rscs r0, r3, #0
; CHECK-NEON-NEXT: movwlt r6, #1
; CHECK-NEON-NEXT: cmp r6, #0
; CHECK-NEON-NEXT: moveq r4, r6
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: movne r7, r8
; CHECK-NEON-NEXT: vmov.32 d0[0], r4
; CHECK-NEON-NEXT: cmp r6, #0
; CHECK-NEON-NEXT: vmov.32 d1[1], r7
; CHECK-NEON-NEXT: movne r6, r1
; CHECK-NEON-NEXT: vmov.32 d0[1], r6
; CHECK-NEON-NEXT: vpop {d8}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, r9, r11, pc}
;
; CHECK-FP16-LABEL: ustest_f16i64:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, r9, r11, lr}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[1]
; CHECK-FP16-NEXT: vmov.u16 r4, d0[0]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfti
; CHECK-FP16-NEXT: mov r8, r1
; CHECK-FP16-NEXT: subs r1, r2, #1
; CHECK-FP16-NEXT: sbcs r1, r3, #0
; CHECK-FP16-NEXT: mov r6, #0
; CHECK-FP16-NEXT: movwlt r6, #1
; CHECK-FP16-NEXT: cmp r6, #0
; CHECK-FP16-NEXT: mov r9, #1
; CHECK-FP16-NEXT: moveq r3, r6
; CHECK-FP16-NEXT: moveq r8, r6
; CHECK-FP16-NEXT: moveq r2, r9
; CHECK-FP16-NEXT: movne r6, r0
; CHECK-FP16-NEXT: rsbs r0, r6, #0
; CHECK-FP16-NEXT: rscs r0, r8, #0
; CHECK-FP16-NEXT: vmov s0, r4
; CHECK-FP16-NEXT: rscs r0, r2, #0
; CHECK-FP16-NEXT: mov r7, #0
; CHECK-FP16-NEXT: rscs r0, r3, #0
; CHECK-FP16-NEXT: mov r5, #0
; CHECK-FP16-NEXT: movwlt r7, #1
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: moveq r6, r7
; CHECK-FP16-NEXT: bl __fixhfti
; CHECK-FP16-NEXT: subs r4, r2, #1
; CHECK-FP16-NEXT: vmov.32 d1[0], r6
; CHECK-FP16-NEXT: sbcs r4, r3, #0
; CHECK-FP16-NEXT: mov r4, #0
; CHECK-FP16-NEXT: movwlt r4, #1
; CHECK-FP16-NEXT: cmp r4, #0
; CHECK-FP16-NEXT: movne r9, r2
; CHECK-FP16-NEXT: moveq r3, r4
; CHECK-FP16-NEXT: moveq r1, r4
; CHECK-FP16-NEXT: movne r4, r0
; CHECK-FP16-NEXT: rsbs r0, r4, #0
; CHECK-FP16-NEXT: rscs r0, r1, #0
; CHECK-FP16-NEXT: rscs r0, r9, #0
; CHECK-FP16-NEXT: rscs r0, r3, #0
; CHECK-FP16-NEXT: movwlt r5, #1
; CHECK-FP16-NEXT: cmp r5, #0
; CHECK-FP16-NEXT: moveq r4, r5
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: movne r7, r8
; CHECK-FP16-NEXT: vmov.32 d0[0], r4
; CHECK-FP16-NEXT: cmp r5, #0
; CHECK-FP16-NEXT: vmov.32 d1[1], r7
; CHECK-FP16-NEXT: movne r5, r1
; CHECK-FP16-NEXT: vmov.32 d0[1], r5
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, r9, r11, pc}
entry:
%conv = fptosi <2 x half> %x to <2 x i128>
%0 = icmp slt <2 x i128> %conv, <i128 18446744073709551616, i128 18446744073709551616>
%spec.store.select = select <2 x i1> %0, <2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>
%1 = icmp sgt <2 x i128> %spec.store.select, zeroinitializer
%spec.store.select7 = select <2 x i1> %1, <2 x i128> %spec.store.select, <2 x i128> zeroinitializer
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
; i32 saturate
define <2 x i32> @stest_f64i32_mm(<2 x double> %x) {
; CHECK-LABEL: stest_f64i32_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, r1, d8
; CHECK-NEXT: bl __aeabi_d2lz
; CHECK-NEXT: mov r4, r0
; CHECK-NEXT: vmov r0, r2, d9
; CHECK-NEXT: mvn r5, #-2147483648
; CHECK-NEXT: subs r3, r4, r5
; CHECK-NEXT: sbcs r3, r1, #0
; CHECK-NEXT: mvn r7, #0
; CHECK-NEXT: mov r3, #0
; CHECK-NEXT: mov r8, #-2147483648
; CHECK-NEXT: movwlt r3, #1
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: movne r3, r1
; CHECK-NEXT: moveq r4, r5
; CHECK-NEXT: rsbs r1, r4, #-2147483648
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: sbcs r1, r7, r3
; CHECK-NEXT: movge r4, r8
; CHECK-NEXT: mov r1, r2
; CHECK-NEXT: bl __aeabi_d2lz
; CHECK-NEXT: subs r2, r0, r5
; CHECK-NEXT: vmov.32 d0[0], r4
; CHECK-NEXT: sbcs r2, r1, #0
; CHECK-NEXT: movwlt r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: movne r5, r0
; CHECK-NEXT: rsbs r0, r5, #-2147483648
; CHECK-NEXT: sbcs r0, r7, r6
; CHECK-NEXT: movge r5, r8
; CHECK-NEXT: vmov.32 d0[1], r5
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, pc}
entry:
%conv = fptosi <2 x double> %x to <2 x i64>
%spec.store.select = call <2 x i64> @llvm.smin.v2i64(<2 x i64> %conv, <2 x i64> <i64 2147483647, i64 2147483647>)
%spec.store.select7 = call <2 x i64> @llvm.smax.v2i64(<2 x i64> %spec.store.select, <2 x i64> <i64 -2147483648, i64 -2147483648>)
%conv6 = trunc <2 x i64> %spec.store.select7 to <2 x i32>
ret <2 x i32> %conv6
}
define <2 x i32> @utest_f64i32_mm(<2 x double> %x) {
; CHECK-LABEL: utest_f64i32_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, lr}
; CHECK-NEXT: push {r4, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, r1, d9
; CHECK-NEXT: bl __aeabi_d2ulz
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: vmov r2, r1, d8
; CHECK-NEXT: vmov.32 d9[0], r0
; CHECK-NEXT: mov r0, r2
; CHECK-NEXT: bl __aeabi_d2ulz
; CHECK-NEXT: vmov.32 d8[0], r0
; CHECK-NEXT: vmov.i64 q8, #0xffffffff
; CHECK-NEXT: vmov.32 d9[1], r4
; CHECK-NEXT: vmov.32 d8[1], r1
; CHECK-NEXT: vqsub.u64 q8, q4, q8
; CHECK-NEXT: vsub.i64 q8, q4, q8
; CHECK-NEXT: vmovn.i64 d0, q8
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, pc}
entry:
%conv = fptoui <2 x double> %x to <2 x i64>
%spec.store.select = call <2 x i64> @llvm.umin.v2i64(<2 x i64> %conv, <2 x i64> <i64 4294967295, i64 4294967295>)
%conv6 = trunc <2 x i64> %spec.store.select to <2 x i32>
ret <2 x i32> %conv6
}
define <2 x i32> @ustest_f64i32_mm(<2 x double> %x) {
; CHECK-LABEL: ustest_f64i32_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, lr}
; CHECK-NEXT: push {r4, r5, r6, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, r1, d8
; CHECK-NEXT: bl __aeabi_d2lz
; CHECK-NEXT: vmov r2, r12, d9
; CHECK-NEXT: mvn r5, #0
; CHECK-NEXT: subs r3, r0, r5
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: sbcs r3, r1, #0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: mov r3, #0
; CHECK-NEXT: movwlt r3, #1
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: movne r3, r1
; CHECK-NEXT: moveq r0, r5
; CHECK-NEXT: rsbs r1, r0, #0
; CHECK-NEXT: rscs r1, r3, #0
; CHECK-NEXT: movwlt r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: movne r6, r0
; CHECK-NEXT: mov r0, r2
; CHECK-NEXT: mov r1, r12
; CHECK-NEXT: bl __aeabi_d2lz
; CHECK-NEXT: subs r2, r0, r5
; CHECK-NEXT: vmov.32 d0[0], r6
; CHECK-NEXT: sbcs r2, r1, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: movne r2, r1
; CHECK-NEXT: movne r5, r0
; CHECK-NEXT: rsbs r0, r5, #0
; CHECK-NEXT: rscs r0, r2, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: movne r4, r5
; CHECK-NEXT: vmov.32 d0[1], r4
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, pc}
entry:
%conv = fptosi <2 x double> %x to <2 x i64>
%spec.store.select = call <2 x i64> @llvm.smin.v2i64(<2 x i64> %conv, <2 x i64> <i64 4294967295, i64 4294967295>)
%spec.store.select7 = call <2 x i64> @llvm.smax.v2i64(<2 x i64> %spec.store.select, <2 x i64> zeroinitializer)
%conv6 = trunc <2 x i64> %spec.store.select7 to <2 x i32>
ret <2 x i32> %conv6
}
define <4 x i32> @stest_f32i32_mm(<4 x float> %x) {
; CHECK-LABEL: stest_f32i32_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: mov r8, #-2147483648
; CHECK-NEXT: mvn r9, #0
; CHECK-NEXT: mov r10, #0
; CHECK-NEXT: vmov r0, s19
; CHECK-NEXT: vmov r5, s16
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r4, r0
; CHECK-NEXT: mvn r7, #-2147483648
; CHECK-NEXT: subs r0, r0, r7
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: movne r0, r1
; CHECK-NEXT: moveq r4, r7
; CHECK-NEXT: rsbs r1, r4, #-2147483648
; CHECK-NEXT: sbcs r0, r9, r0
; CHECK-NEXT: mov r0, r5
; CHECK-NEXT: movge r4, r8
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: subs r0, r0, r7
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: vmov r0, s18
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: movne r2, r1
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: rsbs r1, r5, #-2147483648
; CHECK-NEXT: sbcs r1, r9, r2
; CHECK-NEXT: movge r5, r8
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: mov r6, r0
; CHECK-NEXT: subs r0, r0, r7
; CHECK-NEXT: sbcs r0, r1, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: movne r0, r1
; CHECK-NEXT: moveq r6, r7
; CHECK-NEXT: rsbs r1, r6, #-2147483648
; CHECK-NEXT: sbcs r0, r9, r0
; CHECK-NEXT: vmov r0, s17
; CHECK-NEXT: movge r6, r8
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: subs r2, r0, r7
; CHECK-NEXT: vmov.32 d1[0], r6
; CHECK-NEXT: sbcs r2, r1, #0
; CHECK-NEXT: movwlt r10, #1
; CHECK-NEXT: cmp r10, #0
; CHECK-NEXT: movne r10, r1
; CHECK-NEXT: movne r7, r0
; CHECK-NEXT: rsbs r0, r7, #-2147483648
; CHECK-NEXT: vmov.32 d0[0], r5
; CHECK-NEXT: sbcs r0, r9, r10
; CHECK-NEXT: vmov.32 d1[1], r4
; CHECK-NEXT: movge r7, r8
; CHECK-NEXT: vmov.32 d0[1], r7
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <4 x float> %x to <4 x i64>
%spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 2147483647, i64 2147483647, i64 2147483647, i64 2147483647>)
%spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> <i64 -2147483648, i64 -2147483648, i64 -2147483648, i64 -2147483648>)
%conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @utest_f32i32_mm(<4 x float> %x) {
; CHECK-LABEL: utest_f32i32_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-NEXT: vpush {d8, d9, d10, d11}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, s17
; CHECK-NEXT: bl __aeabi_f2ulz
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: vmov r1, s18
; CHECK-NEXT: vmov r5, s19
; CHECK-NEXT: vmov r6, s16
; CHECK-NEXT: vmov.32 d9[0], r0
; CHECK-NEXT: mov r0, r1
; CHECK-NEXT: bl __aeabi_f2ulz
; CHECK-NEXT: vmov.32 d10[0], r0
; CHECK-NEXT: mov r0, r5
; CHECK-NEXT: mov r7, r1
; CHECK-NEXT: bl __aeabi_f2ulz
; CHECK-NEXT: vmov.32 d11[0], r0
; CHECK-NEXT: mov r0, r6
; CHECK-NEXT: mov r5, r1
; CHECK-NEXT: bl __aeabi_f2ulz
; CHECK-NEXT: vmov.32 d8[0], r0
; CHECK-NEXT: vmov.i64 q8, #0xffffffff
; CHECK-NEXT: vmov.32 d11[1], r5
; CHECK-NEXT: vmov.32 d9[1], r4
; CHECK-NEXT: vmov.32 d10[1], r7
; CHECK-NEXT: vmov.32 d8[1], r1
; CHECK-NEXT: vqsub.u64 q9, q5, q8
; CHECK-NEXT: vqsub.u64 q8, q4, q8
; CHECK-NEXT: vsub.i64 q9, q5, q9
; CHECK-NEXT: vsub.i64 q8, q4, q8
; CHECK-NEXT: vmovn.i64 d1, q9
; CHECK-NEXT: vmovn.i64 d0, q8
; CHECK-NEXT: vpop {d8, d9, d10, d11}
; CHECK-NEXT: pop {r4, r5, r6, r7, r11, pc}
entry:
%conv = fptoui <4 x float> %x to <4 x i64>
%spec.store.select = call <4 x i64> @llvm.umin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
%conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @ustest_f32i32_mm(<4 x float> %x) {
; CHECK-LABEL: ustest_f32i32_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vmov r0, s19
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: vmov r2, s16
; CHECK-NEXT: mvn r7, #0
; CHECK-NEXT: subs r3, r0, r7
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: sbcs r3, r1, #0
; CHECK-NEXT: mov r10, #0
; CHECK-NEXT: mov r3, #0
; CHECK-NEXT: vmov r9, s18
; CHECK-NEXT: movwlt r3, #1
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: movne r3, r1
; CHECK-NEXT: moveq r0, r7
; CHECK-NEXT: rsbs r1, r0, #0
; CHECK-NEXT: vmov r8, s17
; CHECK-NEXT: rscs r1, r3, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: movne r4, r0
; CHECK-NEXT: mov r0, r2
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: subs r2, r0, r7
; CHECK-NEXT: mov r5, #0
; CHECK-NEXT: sbcs r2, r1, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: movne r2, r1
; CHECK-NEXT: moveq r0, r7
; CHECK-NEXT: rsbs r1, r0, #0
; CHECK-NEXT: rscs r1, r2, #0
; CHECK-NEXT: movwlt r5, #1
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: movne r5, r0
; CHECK-NEXT: mov r0, r9
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: subs r2, r0, r7
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: sbcs r2, r1, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: movne r2, r1
; CHECK-NEXT: moveq r0, r7
; CHECK-NEXT: rsbs r1, r0, #0
; CHECK-NEXT: rscs r1, r2, #0
; CHECK-NEXT: movwlt r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: movne r6, r0
; CHECK-NEXT: mov r0, r8
; CHECK-NEXT: bl __aeabi_f2lz
; CHECK-NEXT: subs r2, r0, r7
; CHECK-NEXT: vmov.32 d1[0], r6
; CHECK-NEXT: sbcs r2, r1, #0
; CHECK-NEXT: mov r2, #0
; CHECK-NEXT: vmov.32 d0[0], r5
; CHECK-NEXT: movwlt r2, #1
; CHECK-NEXT: cmp r2, #0
; CHECK-NEXT: movne r2, r1
; CHECK-NEXT: movne r7, r0
; CHECK-NEXT: rsbs r0, r7, #0
; CHECK-NEXT: vmov.32 d1[1], r4
; CHECK-NEXT: rscs r0, r2, #0
; CHECK-NEXT: movwlt r10, #1
; CHECK-NEXT: cmp r10, #0
; CHECK-NEXT: movne r10, r7
; CHECK-NEXT: vmov.32 d0[1], r10
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <4 x float> %x to <4 x i64>
%spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
%spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> zeroinitializer)
%conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @stest_f16i32_mm(<4 x half> %x) {
; CHECK-NEON-LABEL: stest_f16i32_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10}
; CHECK-NEON-NEXT: vpush {d8, d9, d10}
; CHECK-NEON-NEXT: vmov r0, s3
; CHECK-NEON-NEXT: vmov.f32 s18, s2
; CHECK-NEON-NEXT: vmov.f32 s16, s1
; CHECK-NEON-NEXT: vmov.f32 s20, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mov r4, r0
; CHECK-NEON-NEXT: vmov r0, s20
; CHECK-NEON-NEXT: mvn r7, #-2147483648
; CHECK-NEON-NEXT: subs r2, r4, r7
; CHECK-NEON-NEXT: sbcs r2, r1, #0
; CHECK-NEON-NEXT: mov r8, #-2147483648
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: mvn r9, #0
; CHECK-NEON-NEXT: movwlt r2, #1
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: movne r2, r1
; CHECK-NEON-NEXT: moveq r4, r7
; CHECK-NEON-NEXT: rsbs r1, r4, #-2147483648
; CHECK-NEON-NEXT: mov r10, #0
; CHECK-NEON-NEXT: sbcs r1, r9, r2
; CHECK-NEON-NEXT: movge r4, r8
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: subs r0, r0, r7
; CHECK-NEON-NEXT: sbcs r0, r1, #0
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: movwlt r2, #1
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: movne r2, r1
; CHECK-NEON-NEXT: moveq r5, r7
; CHECK-NEON-NEXT: rsbs r1, r5, #-2147483648
; CHECK-NEON-NEXT: sbcs r1, r9, r2
; CHECK-NEON-NEXT: movge r5, r8
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: subs r0, r0, r7
; CHECK-NEON-NEXT: sbcs r0, r1, #0
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: movne r0, r1
; CHECK-NEON-NEXT: moveq r6, r7
; CHECK-NEON-NEXT: rsbs r1, r6, #-2147483648
; CHECK-NEON-NEXT: sbcs r0, r9, r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: movge r6, r8
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: subs r2, r0, r7
; CHECK-NEON-NEXT: vmov.32 d1[0], r6
; CHECK-NEON-NEXT: sbcs r2, r1, #0
; CHECK-NEON-NEXT: movwlt r10, #1
; CHECK-NEON-NEXT: cmp r10, #0
; CHECK-NEON-NEXT: movne r10, r1
; CHECK-NEON-NEXT: movne r7, r0
; CHECK-NEON-NEXT: rsbs r0, r7, #-2147483648
; CHECK-NEON-NEXT: vmov.32 d0[0], r5
; CHECK-NEON-NEXT: sbcs r0, r9, r10
; CHECK-NEON-NEXT: vmov.32 d1[1], r4
; CHECK-NEON-NEXT: movge r7, r8
; CHECK-NEON-NEXT: vmov.32 d0[1], r7
; CHECK-NEON-NEXT: vpop {d8, d9, d10}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
;
; CHECK-FP16-LABEL: stest_f16i32_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: .vsave {d8, d9}
; CHECK-FP16-NEXT: vpush {d8, d9}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[3]
; CHECK-FP16-NEXT: vorr d8, d0, d0
; CHECK-FP16-NEXT: vmov.u16 r5, d0[2]
; CHECK-FP16-NEXT: vmov.u16 r6, d0[0]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: mov r4, r0
; CHECK-FP16-NEXT: mvn r7, #-2147483648
; CHECK-FP16-NEXT: subs r0, r0, r7
; CHECK-FP16-NEXT: vmov s0, r6
; CHECK-FP16-NEXT: sbcs r0, r1, #0
; CHECK-FP16-NEXT: mov r8, #-2147483648
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: mvn r9, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: movne r0, r1
; CHECK-FP16-NEXT: moveq r4, r7
; CHECK-FP16-NEXT: rsbs r1, r4, #-2147483648
; CHECK-FP16-NEXT: mov r10, #0
; CHECK-FP16-NEXT: sbcs r0, r9, r0
; CHECK-FP16-NEXT: vmov s18, r5
; CHECK-FP16-NEXT: movge r4, r8
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: vmov.f32 s0, s18
; CHECK-FP16-NEXT: mov r5, r0
; CHECK-FP16-NEXT: subs r0, r0, r7
; CHECK-FP16-NEXT: sbcs r0, r1, #0
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: movne r0, r1
; CHECK-FP16-NEXT: moveq r5, r7
; CHECK-FP16-NEXT: rsbs r1, r5, #-2147483648
; CHECK-FP16-NEXT: sbcs r0, r9, r0
; CHECK-FP16-NEXT: movge r5, r8
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: mov r6, r0
; CHECK-FP16-NEXT: subs r0, r0, r7
; CHECK-FP16-NEXT: sbcs r0, r1, #0
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: movne r0, r1
; CHECK-FP16-NEXT: vmov.u16 r1, d8[1]
; CHECK-FP16-NEXT: moveq r6, r7
; CHECK-FP16-NEXT: vmov s0, r1
; CHECK-FP16-NEXT: rsbs r1, r6, #-2147483648
; CHECK-FP16-NEXT: sbcs r0, r9, r0
; CHECK-FP16-NEXT: movge r6, r8
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: subs r2, r0, r7
; CHECK-FP16-NEXT: vmov.32 d1[0], r6
; CHECK-FP16-NEXT: sbcs r2, r1, #0
; CHECK-FP16-NEXT: movwlt r10, #1
; CHECK-FP16-NEXT: cmp r10, #0
; CHECK-FP16-NEXT: movne r10, r1
; CHECK-FP16-NEXT: movne r7, r0
; CHECK-FP16-NEXT: rsbs r0, r7, #-2147483648
; CHECK-FP16-NEXT: vmov.32 d0[0], r5
; CHECK-FP16-NEXT: sbcs r0, r9, r10
; CHECK-FP16-NEXT: vmov.32 d1[1], r4
; CHECK-FP16-NEXT: movge r7, r8
; CHECK-FP16-NEXT: vmov.32 d0[1], r7
; CHECK-FP16-NEXT: vpop {d8, d9}
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <4 x half> %x to <4 x i64>
%spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 2147483647, i64 2147483647, i64 2147483647, i64 2147483647>)
%spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> <i64 -2147483648, i64 -2147483648, i64 -2147483648, i64 -2147483648>)
%conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @utesth_f16i32_mm(<4 x half> %x) {
; CHECK-NEON-LABEL: utesth_f16i32_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s3
; CHECK-NEON-NEXT: vmov.f32 s18, s2
; CHECK-NEON-NEXT: vmov.f32 s20, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2ulz
; CHECK-NEON-NEXT: mov r4, r1
; CHECK-NEON-NEXT: vmov r1, s18
; CHECK-NEON-NEXT: vmov r6, s16
; CHECK-NEON-NEXT: vmov.32 d9[0], r0
; CHECK-NEON-NEXT: vmov r7, s20
; CHECK-NEON-NEXT: mov r0, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2ulz
; CHECK-NEON-NEXT: vmov.32 d10[0], r0
; CHECK-NEON-NEXT: mov r0, r6
; CHECK-NEON-NEXT: mov r5, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2ulz
; CHECK-NEON-NEXT: vmov.32 d11[0], r0
; CHECK-NEON-NEXT: mov r0, r7
; CHECK-NEON-NEXT: mov r6, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2ulz
; CHECK-NEON-NEXT: vmov.32 d8[0], r0
; CHECK-NEON-NEXT: vmov.i64 q8, #0xffffffff
; CHECK-NEON-NEXT: vmov.32 d11[1], r6
; CHECK-NEON-NEXT: vmov.32 d9[1], r4
; CHECK-NEON-NEXT: vmov.32 d10[1], r5
; CHECK-NEON-NEXT: vmov.32 d8[1], r1
; CHECK-NEON-NEXT: vqsub.u64 q9, q5, q8
; CHECK-NEON-NEXT: vqsub.u64 q8, q4, q8
; CHECK-NEON-NEXT: vsub.i64 q9, q5, q9
; CHECK-NEON-NEXT: vsub.i64 q8, q4, q8
; CHECK-NEON-NEXT: vmovn.i64 d1, q9
; CHECK-NEON-NEXT: vmovn.i64 d0, q8
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: utesth_f16i32_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, lr}
; CHECK-FP16-NEXT: .vsave {d10, d11, d12, d13}
; CHECK-FP16-NEXT: vpush {d10, d11, d12, d13}
; CHECK-FP16-NEXT: .vsave {d8}
; CHECK-FP16-NEXT: vpush {d8}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[1]
; CHECK-FP16-NEXT: vorr d8, d0, d0
; CHECK-FP16-NEXT: vmov.u16 r6, d0[3]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixunshfdi
; CHECK-FP16-NEXT: mov r4, r1
; CHECK-FP16-NEXT: vmov.u16 r1, d8[2]
; CHECK-FP16-NEXT: vmov.32 d11[0], r0
; CHECK-FP16-NEXT: vmov s0, r1
; CHECK-FP16-NEXT: bl __fixunshfdi
; CHECK-FP16-NEXT: vmov s0, r6
; CHECK-FP16-NEXT: mov r5, r1
; CHECK-FP16-NEXT: vmov.32 d12[0], r0
; CHECK-FP16-NEXT: bl __fixunshfdi
; CHECK-FP16-NEXT: mov r6, r1
; CHECK-FP16-NEXT: vmov.u16 r1, d8[0]
; CHECK-FP16-NEXT: vmov.32 d13[0], r0
; CHECK-FP16-NEXT: vmov s0, r1
; CHECK-FP16-NEXT: bl __fixunshfdi
; CHECK-FP16-NEXT: vmov.32 d10[0], r0
; CHECK-FP16-NEXT: vmov.i64 q8, #0xffffffff
; CHECK-FP16-NEXT: vmov.32 d13[1], r6
; CHECK-FP16-NEXT: vmov.32 d11[1], r4
; CHECK-FP16-NEXT: vmov.32 d12[1], r5
; CHECK-FP16-NEXT: vmov.32 d10[1], r1
; CHECK-FP16-NEXT: vqsub.u64 q9, q6, q8
; CHECK-FP16-NEXT: vqsub.u64 q8, q5, q8
; CHECK-FP16-NEXT: vsub.i64 q9, q6, q9
; CHECK-FP16-NEXT: vsub.i64 q8, q5, q8
; CHECK-FP16-NEXT: vmovn.i64 d1, q9
; CHECK-FP16-NEXT: vmovn.i64 d0, q8
; CHECK-FP16-NEXT: vpop {d8}
; CHECK-FP16-NEXT: vpop {d10, d11, d12, d13}
; CHECK-FP16-NEXT: pop {r4, r5, r6, pc}
entry:
%conv = fptoui <4 x half> %x to <4 x i64>
%spec.store.select = call <4 x i64> @llvm.umin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
%conv6 = trunc <4 x i64> %spec.store.select to <4 x i32>
ret <4 x i32> %conv6
}
define <4 x i32> @ustest_f16i32_mm(<4 x half> %x) {
; CHECK-NEON-LABEL: ustest_f16i32_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10}
; CHECK-NEON-NEXT: vpush {d8, d9, d10}
; CHECK-NEON-NEXT: vmov r0, s3
; CHECK-NEON-NEXT: vmov.f32 s16, s2
; CHECK-NEON-NEXT: vmov.f32 s18, s1
; CHECK-NEON-NEXT: vmov.f32 s20, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: vmov r2, s20
; CHECK-NEON-NEXT: mvn r7, #0
; CHECK-NEON-NEXT: subs r3, r0, r7
; CHECK-NEON-NEXT: mov r4, #0
; CHECK-NEON-NEXT: sbcs r3, r1, #0
; CHECK-NEON-NEXT: mov r10, #0
; CHECK-NEON-NEXT: mov r3, #0
; CHECK-NEON-NEXT: vmov r8, s18
; CHECK-NEON-NEXT: movwlt r3, #1
; CHECK-NEON-NEXT: cmp r3, #0
; CHECK-NEON-NEXT: movne r3, r1
; CHECK-NEON-NEXT: moveq r0, r7
; CHECK-NEON-NEXT: rsbs r1, r0, #0
; CHECK-NEON-NEXT: vmov r9, s16
; CHECK-NEON-NEXT: rscs r1, r3, #0
; CHECK-NEON-NEXT: movwlt r4, #1
; CHECK-NEON-NEXT: cmp r4, #0
; CHECK-NEON-NEXT: movne r4, r0
; CHECK-NEON-NEXT: mov r0, r2
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: subs r2, r0, r7
; CHECK-NEON-NEXT: mov r5, #0
; CHECK-NEON-NEXT: sbcs r2, r1, #0
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: movwlt r2, #1
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: movne r2, r1
; CHECK-NEON-NEXT: moveq r0, r7
; CHECK-NEON-NEXT: rsbs r1, r0, #0
; CHECK-NEON-NEXT: rscs r1, r2, #0
; CHECK-NEON-NEXT: movwlt r5, #1
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: movne r5, r0
; CHECK-NEON-NEXT: mov r0, r9
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: subs r2, r0, r7
; CHECK-NEON-NEXT: mov r6, #0
; CHECK-NEON-NEXT: sbcs r2, r1, #0
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: movwlt r2, #1
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: movne r2, r1
; CHECK-NEON-NEXT: moveq r0, r7
; CHECK-NEON-NEXT: rsbs r1, r0, #0
; CHECK-NEON-NEXT: rscs r1, r2, #0
; CHECK-NEON-NEXT: movwlt r6, #1
; CHECK-NEON-NEXT: cmp r6, #0
; CHECK-NEON-NEXT: movne r6, r0
; CHECK-NEON-NEXT: mov r0, r8
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: bl __aeabi_f2lz
; CHECK-NEON-NEXT: subs r2, r0, r7
; CHECK-NEON-NEXT: vmov.32 d1[0], r6
; CHECK-NEON-NEXT: sbcs r2, r1, #0
; CHECK-NEON-NEXT: mov r2, #0
; CHECK-NEON-NEXT: vmov.32 d0[0], r5
; CHECK-NEON-NEXT: movwlt r2, #1
; CHECK-NEON-NEXT: cmp r2, #0
; CHECK-NEON-NEXT: movne r2, r1
; CHECK-NEON-NEXT: movne r7, r0
; CHECK-NEON-NEXT: rsbs r0, r7, #0
; CHECK-NEON-NEXT: vmov.32 d1[1], r4
; CHECK-NEON-NEXT: rscs r0, r2, #0
; CHECK-NEON-NEXT: movwlt r10, #1
; CHECK-NEON-NEXT: cmp r10, #0
; CHECK-NEON-NEXT: movne r10, r7
; CHECK-NEON-NEXT: vmov.32 d0[1], r10
; CHECK-NEON-NEXT: vpop {d8, d9, d10}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
;
; CHECK-FP16-LABEL: ustest_f16i32_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, lr}
; CHECK-FP16-NEXT: .vsave {d8, d9}
; CHECK-FP16-NEXT: vpush {d8, d9}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[3]
; CHECK-FP16-NEXT: vorr d8, d0, d0
; CHECK-FP16-NEXT: vmov.u16 r6, d0[0]
; CHECK-FP16-NEXT: vmov.u16 r7, d0[2]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: vmov.u16 r2, d8[1]
; CHECK-FP16-NEXT: mvn r5, #0
; CHECK-FP16-NEXT: vmov s0, r6
; CHECK-FP16-NEXT: mov r6, #0
; CHECK-FP16-NEXT: mov r8, #0
; CHECK-FP16-NEXT: vmov s18, r7
; CHECK-FP16-NEXT: vmov s16, r2
; CHECK-FP16-NEXT: subs r2, r0, r5
; CHECK-FP16-NEXT: sbcs r2, r1, #0
; CHECK-FP16-NEXT: mov r2, #0
; CHECK-FP16-NEXT: movwlt r2, #1
; CHECK-FP16-NEXT: cmp r2, #0
; CHECK-FP16-NEXT: movne r2, r1
; CHECK-FP16-NEXT: moveq r0, r5
; CHECK-FP16-NEXT: rsbs r1, r0, #0
; CHECK-FP16-NEXT: rscs r1, r2, #0
; CHECK-FP16-NEXT: movwlt r6, #1
; CHECK-FP16-NEXT: cmp r6, #0
; CHECK-FP16-NEXT: movne r6, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: subs r2, r0, r5
; CHECK-FP16-NEXT: vmov.f32 s0, s18
; CHECK-FP16-NEXT: sbcs r2, r1, #0
; CHECK-FP16-NEXT: mov r7, #0
; CHECK-FP16-NEXT: mov r2, #0
; CHECK-FP16-NEXT: movwlt r2, #1
; CHECK-FP16-NEXT: cmp r2, #0
; CHECK-FP16-NEXT: movne r2, r1
; CHECK-FP16-NEXT: moveq r0, r5
; CHECK-FP16-NEXT: rsbs r1, r0, #0
; CHECK-FP16-NEXT: rscs r1, r2, #0
; CHECK-FP16-NEXT: movwlt r7, #1
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: movne r7, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: subs r2, r0, r5
; CHECK-FP16-NEXT: vmov.f32 s0, s16
; CHECK-FP16-NEXT: sbcs r2, r1, #0
; CHECK-FP16-NEXT: mov r4, #0
; CHECK-FP16-NEXT: mov r2, #0
; CHECK-FP16-NEXT: movwlt r2, #1
; CHECK-FP16-NEXT: cmp r2, #0
; CHECK-FP16-NEXT: movne r2, r1
; CHECK-FP16-NEXT: moveq r0, r5
; CHECK-FP16-NEXT: rsbs r1, r0, #0
; CHECK-FP16-NEXT: rscs r1, r2, #0
; CHECK-FP16-NEXT: movwlt r4, #1
; CHECK-FP16-NEXT: cmp r4, #0
; CHECK-FP16-NEXT: movne r4, r0
; CHECK-FP16-NEXT: bl __fixhfdi
; CHECK-FP16-NEXT: subs r2, r0, r5
; CHECK-FP16-NEXT: vmov.32 d1[0], r4
; CHECK-FP16-NEXT: sbcs r2, r1, #0
; CHECK-FP16-NEXT: mov r2, #0
; CHECK-FP16-NEXT: vmov.32 d0[0], r7
; CHECK-FP16-NEXT: movwlt r2, #1
; CHECK-FP16-NEXT: cmp r2, #0
; CHECK-FP16-NEXT: movne r2, r1
; CHECK-FP16-NEXT: movne r5, r0
; CHECK-FP16-NEXT: rsbs r0, r5, #0
; CHECK-FP16-NEXT: vmov.32 d1[1], r6
; CHECK-FP16-NEXT: rscs r0, r2, #0
; CHECK-FP16-NEXT: movwlt r8, #1
; CHECK-FP16-NEXT: cmp r8, #0
; CHECK-FP16-NEXT: movne r8, r5
; CHECK-FP16-NEXT: vmov.32 d0[1], r8
; CHECK-FP16-NEXT: vpop {d8, d9}
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, pc}
entry:
%conv = fptosi <4 x half> %x to <4 x i64>
%spec.store.select = call <4 x i64> @llvm.smin.v4i64(<4 x i64> %conv, <4 x i64> <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>)
%spec.store.select7 = call <4 x i64> @llvm.smax.v4i64(<4 x i64> %spec.store.select, <4 x i64> zeroinitializer)
%conv6 = trunc <4 x i64> %spec.store.select7 to <4 x i32>
ret <4 x i32> %conv6
}
; i16 saturate
define <2 x i16> @stest_f64i16_mm(<2 x double> %x) {
; CHECK-LABEL: stest_f64i16_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.s32.f64 s4, d0
; CHECK-NEXT: vmov r0, s4
; CHECK-NEXT: vcvt.s32.f64 s0, d1
; CHECK-NEXT: vmov.i32 d17, #0x7fff
; CHECK-NEXT: vmvn.i32 d18, #0x7fff
; CHECK-NEXT: vmov.32 d16[0], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: vmov.32 d16[1], r0
; CHECK-NEXT: vmin.s32 d16, d16, d17
; CHECK-NEXT: vmax.s32 d0, d16, d18
; CHECK-NEXT: bx lr
entry:
%conv = fptosi <2 x double> %x to <2 x i32>
%spec.store.select = call <2 x i32> @llvm.smin.v2i32(<2 x i32> %conv, <2 x i32> <i32 32767, i32 32767>)
%spec.store.select7 = call <2 x i32> @llvm.smax.v2i32(<2 x i32> %spec.store.select, <2 x i32> <i32 -32768, i32 -32768>)
%conv6 = trunc <2 x i32> %spec.store.select7 to <2 x i16>
ret <2 x i16> %conv6
}
define <2 x i16> @utest_f64i16_mm(<2 x double> %x) {
; CHECK-LABEL: utest_f64i16_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.u32.f64 s4, d0
; CHECK-NEXT: vmov r0, s4
; CHECK-NEXT: vcvt.u32.f64 s0, d1
; CHECK-NEXT: vmov.i32 d17, #0xffff
; CHECK-NEXT: vmov.32 d16[0], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: vmov.32 d16[1], r0
; CHECK-NEXT: vmin.u32 d0, d16, d17
; CHECK-NEXT: bx lr
entry:
%conv = fptoui <2 x double> %x to <2 x i32>
%spec.store.select = call <2 x i32> @llvm.umin.v2i32(<2 x i32> %conv, <2 x i32> <i32 65535, i32 65535>)
%conv6 = trunc <2 x i32> %spec.store.select to <2 x i16>
ret <2 x i16> %conv6
}
define <2 x i16> @ustest_f64i16_mm(<2 x double> %x) {
; CHECK-LABEL: ustest_f64i16_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.s32.f64 s4, d0
; CHECK-NEXT: vmov r0, s4
; CHECK-NEXT: vcvt.s32.f64 s0, d1
; CHECK-NEXT: vmov.i32 d17, #0xffff
; CHECK-NEXT: vmov.i32 d18, #0x0
; CHECK-NEXT: vmov.32 d16[0], r0
; CHECK-NEXT: vmov r0, s0
; CHECK-NEXT: vmov.32 d16[1], r0
; CHECK-NEXT: vmin.s32 d16, d16, d17
; CHECK-NEXT: vmax.s32 d0, d16, d18
; CHECK-NEXT: bx lr
entry:
%conv = fptosi <2 x double> %x to <2 x i32>
%spec.store.select = call <2 x i32> @llvm.smin.v2i32(<2 x i32> %conv, <2 x i32> <i32 65535, i32 65535>)
%spec.store.select7 = call <2 x i32> @llvm.smax.v2i32(<2 x i32> %spec.store.select, <2 x i32> zeroinitializer)
%conv6 = trunc <2 x i32> %spec.store.select7 to <2 x i16>
ret <2 x i16> %conv6
}
define <4 x i16> @stest_f32i16_mm(<4 x float> %x) {
; CHECK-LABEL: stest_f32i16_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.s32.f32 q8, q0
; CHECK-NEXT: vmov.i32 q9, #0x7fff
; CHECK-NEXT: vmvn.i32 q10, #0x7fff
; CHECK-NEXT: vmin.s32 q8, q8, q9
; CHECK-NEXT: vmax.s32 q8, q8, q10
; CHECK-NEXT: vmovn.i32 d0, q8
; CHECK-NEXT: bx lr
entry:
%conv = fptosi <4 x float> %x to <4 x i32>
%spec.store.select = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %conv, <4 x i32> <i32 32767, i32 32767, i32 32767, i32 32767>)
%spec.store.select7 = call <4 x i32> @llvm.smax.v4i32(<4 x i32> %spec.store.select, <4 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768>)
%conv6 = trunc <4 x i32> %spec.store.select7 to <4 x i16>
ret <4 x i16> %conv6
}
define <4 x i16> @utest_f32i16_mm(<4 x float> %x) {
; CHECK-LABEL: utest_f32i16_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.u32.f32 q8, q0
; CHECK-NEXT: vmov.i32 q9, #0xffff
; CHECK-NEXT: vmin.u32 q8, q8, q9
; CHECK-NEXT: vmovn.i32 d0, q8
; CHECK-NEXT: bx lr
entry:
%conv = fptoui <4 x float> %x to <4 x i32>
%spec.store.select = call <4 x i32> @llvm.umin.v4i32(<4 x i32> %conv, <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>)
%conv6 = trunc <4 x i32> %spec.store.select to <4 x i16>
ret <4 x i16> %conv6
}
define <4 x i16> @ustest_f32i16_mm(<4 x float> %x) {
; CHECK-LABEL: ustest_f32i16_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vcvt.s32.f32 q8, q0
; CHECK-NEXT: vmov.i32 q9, #0xffff
; CHECK-NEXT: vmov.i32 q10, #0x0
; CHECK-NEXT: vmin.s32 q8, q8, q9
; CHECK-NEXT: vmax.s32 q8, q8, q10
; CHECK-NEXT: vmovn.i32 d0, q8
; CHECK-NEXT: bx lr
entry:
%conv = fptosi <4 x float> %x to <4 x i32>
%spec.store.select = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %conv, <4 x i32> <i32 65535, i32 65535, i32 65535, i32 65535>)
%spec.store.select7 = call <4 x i32> @llvm.smax.v4i32(<4 x i32> %spec.store.select, <4 x i32> zeroinitializer)
%conv6 = trunc <4 x i32> %spec.store.select7 to <4 x i16>
ret <4 x i16> %conv6
}
define <8 x i16> @stest_f16i16_mm(<8 x half> %x) {
; CHECK-NEON-LABEL: stest_f16i16_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s7
; CHECK-NEON-NEXT: vmov.f32 s18, s6
; CHECK-NEON-NEXT: vmov.f32 s20, s5
; CHECK-NEON-NEXT: vmov.f32 s22, s4
; CHECK-NEON-NEXT: vmov.f32 s24, s3
; CHECK-NEON-NEXT: vmov.f32 s26, s2
; CHECK-NEON-NEXT: vmov.f32 s28, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r4, r0
; CHECK-NEON-NEXT: vmov r0, s26
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s22
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: vmov r0, s24
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r7, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[0], r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov s22, r7
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s30, r6
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[1], r0
; CHECK-NEON-NEXT: vmov r0, s28
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r1, s20
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r5
; CHECK-NEON-NEXT: vcvt.s32.f32 s20, s2
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s30
; CHECK-NEON-NEXT: vmov.32 d8[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d12[0], r0
; CHECK-NEON-NEXT: mov r0, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r0, s20
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r4
; CHECK-NEON-NEXT: vmov.i32 q8, #0x7fff
; CHECK-NEON-NEXT: vcvt.s32.f32 s2, s2
; CHECK-NEON-NEXT: vmvn.i32 q9, #0x7fff
; CHECK-NEON-NEXT: vmov.32 d9[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s22
; CHECK-NEON-NEXT: vmov.32 d12[1], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmin.s32 q10, q6, q8
; CHECK-NEON-NEXT: vmax.s32 q10, q10, q9
; CHECK-NEON-NEXT: vmov.32 d9[1], r0
; CHECK-NEON-NEXT: vmov r0, s2
; CHECK-NEON-NEXT: vmovn.i32 d1, q10
; CHECK-NEON-NEXT: vmov.32 d8[1], r0
; CHECK-NEON-NEXT: vmin.s32 q8, q4, q8
; CHECK-NEON-NEXT: vmax.s32 q8, q8, q9
; CHECK-NEON-NEXT: vmovn.i32 d0, q8
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: stest_f16i16_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: vmovx.f16 s4, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s12, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s0, s3
; CHECK-FP16-NEXT: vcvt.s32.f16 s5, s2
; CHECK-FP16-NEXT: vmov r0, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s14, s1
; CHECK-FP16-NEXT: vmovx.f16 s10, s3
; CHECK-FP16-NEXT: vmovx.f16 s8, s2
; CHECK-FP16-NEXT: vcvt.s32.f16 s10, s10
; CHECK-FP16-NEXT: vcvt.s32.f16 s8, s8
; CHECK-FP16-NEXT: vmovx.f16 s6, s1
; CHECK-FP16-NEXT: vcvt.s32.f16 s4, s4
; CHECK-FP16-NEXT: vcvt.s32.f16 s6, s6
; CHECK-FP16-NEXT: vmov.i32 q10, #0x7fff
; CHECK-FP16-NEXT: vmvn.i32 q11, #0x7fff
; CHECK-FP16-NEXT: vmov.32 d17[0], r0
; CHECK-FP16-NEXT: vmov r0, s5
; CHECK-FP16-NEXT: vmov.32 d16[0], r0
; CHECK-FP16-NEXT: vmov r0, s14
; CHECK-FP16-NEXT: vmov.32 d19[0], r0
; CHECK-FP16-NEXT: vmov r0, s12
; CHECK-FP16-NEXT: vmov.32 d18[0], r0
; CHECK-FP16-NEXT: vmov r0, s10
; CHECK-FP16-NEXT: vmov.32 d17[1], r0
; CHECK-FP16-NEXT: vmov r0, s8
; CHECK-FP16-NEXT: vmov.32 d16[1], r0
; CHECK-FP16-NEXT: vmov r0, s6
; CHECK-FP16-NEXT: vmin.s32 q8, q8, q10
; CHECK-FP16-NEXT: vmax.s32 q8, q8, q11
; CHECK-FP16-NEXT: vmovn.i32 d1, q8
; CHECK-FP16-NEXT: vmov.32 d19[1], r0
; CHECK-FP16-NEXT: vmov r0, s4
; CHECK-FP16-NEXT: vmov.32 d18[1], r0
; CHECK-FP16-NEXT: vmin.s32 q9, q9, q10
; CHECK-FP16-NEXT: vmax.s32 q9, q9, q11
; CHECK-FP16-NEXT: vmovn.i32 d0, q9
; CHECK-FP16-NEXT: bx lr
entry:
%conv = fptosi <8 x half> %x to <8 x i32>
%spec.store.select = call <8 x i32> @llvm.smin.v8i32(<8 x i32> %conv, <8 x i32> <i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767, i32 32767>)
%spec.store.select7 = call <8 x i32> @llvm.smax.v8i32(<8 x i32> %spec.store.select, <8 x i32> <i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768, i32 -32768>)
%conv6 = trunc <8 x i32> %spec.store.select7 to <8 x i16>
ret <8 x i16> %conv6
}
define <8 x i16> @utesth_f16i16_mm(<8 x half> %x) {
; CHECK-NEON-LABEL: utesth_f16i16_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11, d12, d13, d14}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11, d12, d13, d14}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s7
; CHECK-NEON-NEXT: vmov.f32 s18, s6
; CHECK-NEON-NEXT: vmov.f32 s20, s5
; CHECK-NEON-NEXT: vmov.f32 s22, s4
; CHECK-NEON-NEXT: vmov.f32 s24, s3
; CHECK-NEON-NEXT: vmov.f32 s26, s2
; CHECK-NEON-NEXT: vmov.f32 s28, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r4, r0
; CHECK-NEON-NEXT: vmov r0, s26
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s22
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: vmov r0, s24
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r7, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[0], r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov s16, r7
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s18, r6
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[1], r0
; CHECK-NEON-NEXT: vmov r0, s28
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r1, s20
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r5
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s18
; CHECK-NEON-NEXT: vcvt.u32.f32 s18, s2
; CHECK-NEON-NEXT: vmov.32 d10[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d12[0], r0
; CHECK-NEON-NEXT: mov r0, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r4
; CHECK-NEON-NEXT: vmov.i32 q8, #0xffff
; CHECK-NEON-NEXT: vcvt.u32.f32 s2, s2
; CHECK-NEON-NEXT: vmov.32 d11[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.u32.f32 s0, s16
; CHECK-NEON-NEXT: vmov.32 d12[1], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmin.u32 q9, q6, q8
; CHECK-NEON-NEXT: vmov.32 d11[1], r0
; CHECK-NEON-NEXT: vmov r0, s2
; CHECK-NEON-NEXT: vmovn.i32 d1, q9
; CHECK-NEON-NEXT: vmov.32 d10[1], r0
; CHECK-NEON-NEXT: vmin.u32 q8, q5, q8
; CHECK-NEON-NEXT: vmovn.i32 d0, q8
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11, d12, d13, d14}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: utesth_f16i16_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: vmovx.f16 s4, s0
; CHECK-FP16-NEXT: vcvt.u32.f16 s12, s0
; CHECK-FP16-NEXT: vcvt.u32.f16 s0, s3
; CHECK-FP16-NEXT: vcvt.u32.f16 s5, s2
; CHECK-FP16-NEXT: vmov r0, s0
; CHECK-FP16-NEXT: vcvt.u32.f16 s14, s1
; CHECK-FP16-NEXT: vmovx.f16 s10, s3
; CHECK-FP16-NEXT: vmovx.f16 s8, s2
; CHECK-FP16-NEXT: vcvt.u32.f16 s10, s10
; CHECK-FP16-NEXT: vcvt.u32.f16 s8, s8
; CHECK-FP16-NEXT: vmovx.f16 s6, s1
; CHECK-FP16-NEXT: vcvt.u32.f16 s4, s4
; CHECK-FP16-NEXT: vcvt.u32.f16 s6, s6
; CHECK-FP16-NEXT: vmov.i32 q10, #0xffff
; CHECK-FP16-NEXT: vmov.32 d17[0], r0
; CHECK-FP16-NEXT: vmov r0, s5
; CHECK-FP16-NEXT: vmov.32 d16[0], r0
; CHECK-FP16-NEXT: vmov r0, s14
; CHECK-FP16-NEXT: vmov.32 d19[0], r0
; CHECK-FP16-NEXT: vmov r0, s12
; CHECK-FP16-NEXT: vmov.32 d18[0], r0
; CHECK-FP16-NEXT: vmov r0, s10
; CHECK-FP16-NEXT: vmov.32 d17[1], r0
; CHECK-FP16-NEXT: vmov r0, s8
; CHECK-FP16-NEXT: vmov.32 d16[1], r0
; CHECK-FP16-NEXT: vmov r0, s6
; CHECK-FP16-NEXT: vmin.u32 q8, q8, q10
; CHECK-FP16-NEXT: vmovn.i32 d1, q8
; CHECK-FP16-NEXT: vmov.32 d19[1], r0
; CHECK-FP16-NEXT: vmov r0, s4
; CHECK-FP16-NEXT: vmov.32 d18[1], r0
; CHECK-FP16-NEXT: vmin.u32 q9, q9, q10
; CHECK-FP16-NEXT: vmovn.i32 d0, q9
; CHECK-FP16-NEXT: bx lr
entry:
%conv = fptoui <8 x half> %x to <8 x i32>
%spec.store.select = call <8 x i32> @llvm.umin.v8i32(<8 x i32> %conv, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>)
%conv6 = trunc <8 x i32> %spec.store.select to <8 x i16>
ret <8 x i16> %conv6
}
define <8 x i16> @ustest_f16i16_mm(<8 x half> %x) {
; CHECK-NEON-LABEL: ustest_f16i16_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: vpush {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s7
; CHECK-NEON-NEXT: vmov.f32 s18, s6
; CHECK-NEON-NEXT: vmov.f32 s20, s5
; CHECK-NEON-NEXT: vmov.f32 s22, s4
; CHECK-NEON-NEXT: vmov.f32 s24, s3
; CHECK-NEON-NEXT: vmov.f32 s26, s2
; CHECK-NEON-NEXT: vmov.f32 s28, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r4, r0
; CHECK-NEON-NEXT: vmov r0, s26
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s22
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: vmov r0, s24
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r7, r0
; CHECK-NEON-NEXT: vmov r0, s18
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[0], r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov s22, r7
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s30, r6
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d13[1], r0
; CHECK-NEON-NEXT: vmov r0, s28
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r1, s20
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r5
; CHECK-NEON-NEXT: vcvt.s32.f32 s20, s2
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s30
; CHECK-NEON-NEXT: vmov.32 d8[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.32 d12[0], r0
; CHECK-NEON-NEXT: mov r0, r1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: vmov r0, s20
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s0
; CHECK-NEON-NEXT: vmov s2, r4
; CHECK-NEON-NEXT: vmov.i32 q8, #0xffff
; CHECK-NEON-NEXT: vcvt.s32.f32 s2, s2
; CHECK-NEON-NEXT: vmov.i32 q9, #0x0
; CHECK-NEON-NEXT: vmov.32 d9[0], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vcvt.s32.f32 s0, s22
; CHECK-NEON-NEXT: vmov.32 d12[1], r0
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmin.s32 q10, q6, q8
; CHECK-NEON-NEXT: vmax.s32 q10, q10, q9
; CHECK-NEON-NEXT: vmov.32 d9[1], r0
; CHECK-NEON-NEXT: vmov r0, s2
; CHECK-NEON-NEXT: vmovn.i32 d1, q10
; CHECK-NEON-NEXT: vmov.32 d8[1], r0
; CHECK-NEON-NEXT: vmin.s32 q8, q4, q8
; CHECK-NEON-NEXT: vmax.s32 q8, q8, q9
; CHECK-NEON-NEXT: vmovn.i32 d0, q8
; CHECK-NEON-NEXT: vpop {d8, d9, d10, d11, d12, d13, d14, d15}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: ustest_f16i16_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: vmovx.f16 s4, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s12, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s0, s3
; CHECK-FP16-NEXT: vcvt.s32.f16 s5, s2
; CHECK-FP16-NEXT: vmov r0, s0
; CHECK-FP16-NEXT: vcvt.s32.f16 s14, s1
; CHECK-FP16-NEXT: vmovx.f16 s10, s3
; CHECK-FP16-NEXT: vmovx.f16 s8, s2
; CHECK-FP16-NEXT: vcvt.s32.f16 s10, s10
; CHECK-FP16-NEXT: vcvt.s32.f16 s8, s8
; CHECK-FP16-NEXT: vmovx.f16 s6, s1
; CHECK-FP16-NEXT: vcvt.s32.f16 s4, s4
; CHECK-FP16-NEXT: vcvt.s32.f16 s6, s6
; CHECK-FP16-NEXT: vmov.i32 q10, #0xffff
; CHECK-FP16-NEXT: vmov.i32 q11, #0x0
; CHECK-FP16-NEXT: vmov.32 d17[0], r0
; CHECK-FP16-NEXT: vmov r0, s5
; CHECK-FP16-NEXT: vmov.32 d16[0], r0
; CHECK-FP16-NEXT: vmov r0, s14
; CHECK-FP16-NEXT: vmov.32 d19[0], r0
; CHECK-FP16-NEXT: vmov r0, s12
; CHECK-FP16-NEXT: vmov.32 d18[0], r0
; CHECK-FP16-NEXT: vmov r0, s10
; CHECK-FP16-NEXT: vmov.32 d17[1], r0
; CHECK-FP16-NEXT: vmov r0, s8
; CHECK-FP16-NEXT: vmov.32 d16[1], r0
; CHECK-FP16-NEXT: vmov r0, s6
; CHECK-FP16-NEXT: vmin.s32 q8, q8, q10
; CHECK-FP16-NEXT: vmax.s32 q8, q8, q11
; CHECK-FP16-NEXT: vmovn.i32 d1, q8
; CHECK-FP16-NEXT: vmov.32 d19[1], r0
; CHECK-FP16-NEXT: vmov r0, s4
; CHECK-FP16-NEXT: vmov.32 d18[1], r0
; CHECK-FP16-NEXT: vmin.s32 q9, q9, q10
; CHECK-FP16-NEXT: vmax.s32 q9, q9, q11
; CHECK-FP16-NEXT: vmovn.i32 d0, q9
; CHECK-FP16-NEXT: bx lr
entry:
%conv = fptosi <8 x half> %x to <8 x i32>
%spec.store.select = call <8 x i32> @llvm.smin.v8i32(<8 x i32> %conv, <8 x i32> <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>)
%spec.store.select7 = call <8 x i32> @llvm.smax.v8i32(<8 x i32> %spec.store.select, <8 x i32> zeroinitializer)
%conv6 = trunc <8 x i32> %spec.store.select7 to <8 x i16>
ret <8 x i16> %conv6
}
; i64 saturate
define <2 x i64> @stest_f64i64_mm(<2 x double> %x) {
; CHECK-LABEL: stest_f64i64_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vorr d0, d9, d9
; CHECK-NEXT: bl __fixdfti
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: mvn r8, #0
; CHECK-NEXT: subs r0, r0, r8
; CHECK-NEXT: mvn r6, #-2147483648
; CHECK-NEXT: sbcs r0, r1, r6
; CHECK-NEXT: mov r10, r1
; CHECK-NEXT: sbcs r0, r2, #0
; CHECK-NEXT: vorr d0, d8, d8
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: mov r9, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: moveq r3, r0
; CHECK-NEXT: movne r0, r2
; CHECK-NEXT: moveq r10, r6
; CHECK-NEXT: moveq r5, r8
; CHECK-NEXT: rsbs r1, r5, #0
; CHECK-NEXT: rscs r1, r10, #-2147483648
; CHECK-NEXT: sbcs r0, r8, r0
; CHECK-NEXT: sbcs r0, r8, r3
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: bl __fixdfti
; CHECK-NEXT: subs r4, r0, r8
; CHECK-NEXT: vmov.32 d1[0], r5
; CHECK-NEXT: sbcs r4, r1, r6
; CHECK-NEXT: sbcs r4, r2, #0
; CHECK-NEXT: sbcs r4, r3, #0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: moveq r3, r4
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: movne r4, r2
; CHECK-NEXT: moveq r0, r8
; CHECK-NEXT: rsbs r1, r0, #0
; CHECK-NEXT: rscs r1, r6, #-2147483648
; CHECK-NEXT: sbcs r1, r8, r4
; CHECK-NEXT: sbcs r1, r8, r3
; CHECK-NEXT: movwlt r9, #1
; CHECK-NEXT: cmp r9, #0
; CHECK-NEXT: moveq r0, r9
; CHECK-NEXT: mov r1, #-2147483648
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: vmov.32 d0[0], r0
; CHECK-NEXT: moveq r10, r1
; CHECK-NEXT: cmp r9, #0
; CHECK-NEXT: vmov.32 d1[1], r10
; CHECK-NEXT: moveq r6, r1
; CHECK-NEXT: vmov.32 d0[1], r6
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <2 x double> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.smin.v2i128(<2 x i128> %conv, <2 x i128> <i128 9223372036854775807, i128 9223372036854775807>)
%spec.store.select7 = call <2 x i128> @llvm.smax.v2i128(<2 x i128> %spec.store.select, <2 x i128> <i128 -9223372036854775808, i128 -9223372036854775808>)
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @utest_f64i64_mm(<2 x double> %x) {
; CHECK-LABEL: utest_f64i64_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: vorr d0, d9, d9
; CHECK-NEXT: bl __fixunsdfti
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: subs r0, r2, #1
; CHECK-NEXT: vorr d0, d8, d8
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: movwlo r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: bl __fixunsdfti
; CHECK-NEXT: subs r2, r2, #1
; CHECK-NEXT: vmov.32 d1[0], r5
; CHECK-NEXT: sbcs r2, r3, #0
; CHECK-NEXT: movwlo r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: moveq r0, r6
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: movne r7, r4
; CHECK-NEXT: vmov.32 d0[0], r0
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: vmov.32 d1[1], r7
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: vmov.32 d0[1], r6
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r11, pc}
entry:
%conv = fptoui <2 x double> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.umin.v2i128(<2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>)
%conv6 = trunc <2 x i128> %spec.store.select to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @ustest_f64i64_mm(<2 x double> %x) {
; CHECK-LABEL: ustest_f64i64_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, lr}
; CHECK-NEXT: .vsave {d8, d9}
; CHECK-NEXT: vpush {d8, d9}
; CHECK-NEXT: vorr q4, q0, q0
; CHECK-NEXT: bl __fixdfti
; CHECK-NEXT: mov r6, r0
; CHECK-NEXT: subs r0, r2, #1
; CHECK-NEXT: vorr d0, d9, d9
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: mov r5, r3
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: moveq r6, r7
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: mov r8, r1
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: movwmi r6, #0
; CHECK-NEXT: bl __fixdfti
; CHECK-NEXT: subs r2, r2, #1
; CHECK-NEXT: vmov.32 d0[0], r6
; CHECK-NEXT: sbcs r2, r3, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: moveq r0, r4
; CHECK-NEXT: moveq r3, r4
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: movwmi r0, #0
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: movne r4, r1
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: vmov.32 d1[0], r0
; CHECK-NEXT: movwmi r4, #0
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: movne r7, r8
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: vmov.32 d1[1], r4
; CHECK-NEXT: movwmi r7, #0
; CHECK-NEXT: vmov.32 d0[1], r7
; CHECK-NEXT: vpop {d8, d9}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, pc}
entry:
%conv = fptosi <2 x double> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.smin.v2i128(<2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>)
%spec.store.select7 = call <2 x i128> @llvm.smax.v2i128(<2 x i128> %spec.store.select, <2 x i128> zeroinitializer)
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @stest_f32i64_mm(<2 x float> %x) {
; CHECK-LABEL: stest_f32i64_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-NEXT: .vsave {d8}
; CHECK-NEXT: vpush {d8}
; CHECK-NEXT: vmov.f64 d8, d0
; CHECK-NEXT: vmov.f32 s0, s17
; CHECK-NEXT: bl __fixsfti
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: mvn r8, #0
; CHECK-NEXT: subs r0, r0, r8
; CHECK-NEXT: mvn r6, #-2147483648
; CHECK-NEXT: sbcs r0, r1, r6
; CHECK-NEXT: vmov.f32 s0, s16
; CHECK-NEXT: sbcs r0, r2, #0
; CHECK-NEXT: mov r10, r1
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: mov r0, #0
; CHECK-NEXT: mov r9, #0
; CHECK-NEXT: movwlt r0, #1
; CHECK-NEXT: cmp r0, #0
; CHECK-NEXT: moveq r3, r0
; CHECK-NEXT: movne r0, r2
; CHECK-NEXT: moveq r10, r6
; CHECK-NEXT: moveq r5, r8
; CHECK-NEXT: rsbs r1, r5, #0
; CHECK-NEXT: rscs r1, r10, #-2147483648
; CHECK-NEXT: sbcs r0, r8, r0
; CHECK-NEXT: sbcs r0, r8, r3
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: bl __fixsfti
; CHECK-NEXT: subs r4, r0, r8
; CHECK-NEXT: vmov.32 d1[0], r5
; CHECK-NEXT: sbcs r4, r1, r6
; CHECK-NEXT: sbcs r4, r2, #0
; CHECK-NEXT: sbcs r4, r3, #0
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: moveq r3, r4
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: movne r4, r2
; CHECK-NEXT: moveq r0, r8
; CHECK-NEXT: rsbs r1, r0, #0
; CHECK-NEXT: rscs r1, r6, #-2147483648
; CHECK-NEXT: sbcs r1, r8, r4
; CHECK-NEXT: sbcs r1, r8, r3
; CHECK-NEXT: movwlt r9, #1
; CHECK-NEXT: cmp r9, #0
; CHECK-NEXT: moveq r0, r9
; CHECK-NEXT: mov r1, #-2147483648
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: vmov.32 d0[0], r0
; CHECK-NEXT: moveq r10, r1
; CHECK-NEXT: cmp r9, #0
; CHECK-NEXT: vmov.32 d1[1], r10
; CHECK-NEXT: moveq r6, r1
; CHECK-NEXT: vmov.32 d0[1], r6
; CHECK-NEXT: vpop {d8}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <2 x float> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.smin.v2i128(<2 x i128> %conv, <2 x i128> <i128 9223372036854775807, i128 9223372036854775807>)
%spec.store.select7 = call <2 x i128> @llvm.smax.v2i128(<2 x i128> %spec.store.select, <2 x i128> <i128 -9223372036854775808, i128 -9223372036854775808>)
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @utest_f32i64_mm(<2 x float> %x) {
; CHECK-LABEL: utest_f32i64_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEXT: .vsave {d8}
; CHECK-NEXT: vpush {d8}
; CHECK-NEXT: vmov.f64 d8, d0
; CHECK-NEXT: vmov.f32 s0, s17
; CHECK-NEXT: bl __fixunssfti
; CHECK-NEXT: vmov.f32 s0, s16
; CHECK-NEXT: mov r5, r0
; CHECK-NEXT: subs r0, r2, #1
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r4, r1
; CHECK-NEXT: movwlo r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: mov r6, #0
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: bl __fixunssfti
; CHECK-NEXT: subs r2, r2, #1
; CHECK-NEXT: vmov.32 d1[0], r5
; CHECK-NEXT: sbcs r2, r3, #0
; CHECK-NEXT: movwlo r6, #1
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: moveq r0, r6
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: movne r7, r4
; CHECK-NEXT: vmov.32 d0[0], r0
; CHECK-NEXT: cmp r6, #0
; CHECK-NEXT: vmov.32 d1[1], r7
; CHECK-NEXT: movne r6, r1
; CHECK-NEXT: vmov.32 d0[1], r6
; CHECK-NEXT: vpop {d8}
; CHECK-NEXT: pop {r4, r5, r6, r7, r11, pc}
entry:
%conv = fptoui <2 x float> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.umin.v2i128(<2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>)
%conv6 = trunc <2 x i128> %spec.store.select to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @ustest_f32i64_mm(<2 x float> %x) {
; CHECK-LABEL: ustest_f32i64_mm:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: .save {r4, r5, r6, r7, r8, lr}
; CHECK-NEXT: push {r4, r5, r6, r7, r8, lr}
; CHECK-NEXT: .vsave {d8}
; CHECK-NEXT: vpush {d8}
; CHECK-NEXT: vmov.f64 d8, d0
; CHECK-NEXT: bl __fixsfti
; CHECK-NEXT: vmov.f32 s0, s17
; CHECK-NEXT: mov r6, r0
; CHECK-NEXT: subs r0, r2, #1
; CHECK-NEXT: mov r7, #0
; CHECK-NEXT: sbcs r0, r3, #0
; CHECK-NEXT: mov r5, r3
; CHECK-NEXT: movwlt r7, #1
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: moveq r6, r7
; CHECK-NEXT: moveq r5, r7
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: mov r8, r1
; CHECK-NEXT: mov r4, #0
; CHECK-NEXT: movwmi r6, #0
; CHECK-NEXT: bl __fixsfti
; CHECK-NEXT: subs r2, r2, #1
; CHECK-NEXT: vmov.32 d0[0], r6
; CHECK-NEXT: sbcs r2, r3, #0
; CHECK-NEXT: movwlt r4, #1
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: moveq r0, r4
; CHECK-NEXT: moveq r3, r4
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: movwmi r0, #0
; CHECK-NEXT: cmp r4, #0
; CHECK-NEXT: movne r4, r1
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: vmov.32 d1[0], r0
; CHECK-NEXT: movwmi r4, #0
; CHECK-NEXT: cmp r7, #0
; CHECK-NEXT: movne r7, r8
; CHECK-NEXT: cmp r5, #0
; CHECK-NEXT: vmov.32 d1[1], r4
; CHECK-NEXT: movwmi r7, #0
; CHECK-NEXT: vmov.32 d0[1], r7
; CHECK-NEXT: vpop {d8}
; CHECK-NEXT: pop {r4, r5, r6, r7, r8, pc}
entry:
%conv = fptosi <2 x float> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.smin.v2i128(<2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>)
%spec.store.select7 = call <2 x i128> @llvm.smax.v2i128(<2 x i128> %spec.store.select, <2 x i128> zeroinitializer)
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @stest_f16i64_mm(<2 x half> %x) {
; CHECK-NEON-LABEL: stest_f16i64_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; CHECK-NEON-NEXT: .pad #4
; CHECK-NEON-NEXT: sub sp, sp, #4
; CHECK-NEON-NEXT: .vsave {d8}
; CHECK-NEON-NEXT: vpush {d8}
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.f32 s16, s1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r8, r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: bl __fixsfti
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: mvn r9, #0
; CHECK-NEON-NEXT: subs r0, r0, r9
; CHECK-NEON-NEXT: mvn r7, #-2147483648
; CHECK-NEON-NEXT: sbcs r0, r1, r7
; CHECK-NEON-NEXT: mov r11, r1
; CHECK-NEON-NEXT: sbcs r0, r2, #0
; CHECK-NEON-NEXT: vmov s0, r8
; CHECK-NEON-NEXT: sbcs r0, r3, #0
; CHECK-NEON-NEXT: mov r6, #0
; CHECK-NEON-NEXT: mov r0, #0
; CHECK-NEON-NEXT: mov r10, #0
; CHECK-NEON-NEXT: movwlt r0, #1
; CHECK-NEON-NEXT: cmp r0, #0
; CHECK-NEON-NEXT: moveq r3, r0
; CHECK-NEON-NEXT: movne r0, r2
; CHECK-NEON-NEXT: moveq r11, r7
; CHECK-NEON-NEXT: moveq r5, r9
; CHECK-NEON-NEXT: rsbs r1, r5, #0
; CHECK-NEON-NEXT: rscs r1, r11, #-2147483648
; CHECK-NEON-NEXT: sbcs r0, r9, r0
; CHECK-NEON-NEXT: sbcs r0, r9, r3
; CHECK-NEON-NEXT: movwlt r6, #1
; CHECK-NEON-NEXT: cmp r6, #0
; CHECK-NEON-NEXT: moveq r5, r6
; CHECK-NEON-NEXT: bl __fixsfti
; CHECK-NEON-NEXT: subs r4, r0, r9
; CHECK-NEON-NEXT: vmov.32 d1[0], r5
; CHECK-NEON-NEXT: sbcs r4, r1, r7
; CHECK-NEON-NEXT: sbcs r4, r2, #0
; CHECK-NEON-NEXT: sbcs r4, r3, #0
; CHECK-NEON-NEXT: mov r4, #0
; CHECK-NEON-NEXT: movwlt r4, #1
; CHECK-NEON-NEXT: cmp r4, #0
; CHECK-NEON-NEXT: moveq r3, r4
; CHECK-NEON-NEXT: movne r7, r1
; CHECK-NEON-NEXT: movne r4, r2
; CHECK-NEON-NEXT: moveq r0, r9
; CHECK-NEON-NEXT: rsbs r1, r0, #0
; CHECK-NEON-NEXT: rscs r1, r7, #-2147483648
; CHECK-NEON-NEXT: sbcs r1, r9, r4
; CHECK-NEON-NEXT: sbcs r1, r9, r3
; CHECK-NEON-NEXT: movwlt r10, #1
; CHECK-NEON-NEXT: cmp r10, #0
; CHECK-NEON-NEXT: moveq r0, r10
; CHECK-NEON-NEXT: mov r1, #-2147483648
; CHECK-NEON-NEXT: cmp r6, #0
; CHECK-NEON-NEXT: vmov.32 d0[0], r0
; CHECK-NEON-NEXT: moveq r11, r1
; CHECK-NEON-NEXT: cmp r10, #0
; CHECK-NEON-NEXT: vmov.32 d1[1], r11
; CHECK-NEON-NEXT: moveq r7, r1
; CHECK-NEON-NEXT: vmov.32 d0[1], r7
; CHECK-NEON-NEXT: vpop {d8}
; CHECK-NEON-NEXT: add sp, sp, #4
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, r11, pc}
;
; CHECK-FP16-LABEL: stest_f16i64_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, r9, r10, lr}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[1]
; CHECK-FP16-NEXT: vmov.u16 r7, d0[0]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfti
; CHECK-FP16-NEXT: mov r5, r0
; CHECK-FP16-NEXT: mvn r8, #0
; CHECK-FP16-NEXT: subs r0, r0, r8
; CHECK-FP16-NEXT: mvn r6, #-2147483648
; CHECK-FP16-NEXT: sbcs r0, r1, r6
; CHECK-FP16-NEXT: mov r10, r1
; CHECK-FP16-NEXT: sbcs r0, r2, #0
; CHECK-FP16-NEXT: vmov s0, r7
; CHECK-FP16-NEXT: sbcs r0, r3, #0
; CHECK-FP16-NEXT: mov r7, #0
; CHECK-FP16-NEXT: mov r0, #0
; CHECK-FP16-NEXT: mov r9, #0
; CHECK-FP16-NEXT: movwlt r0, #1
; CHECK-FP16-NEXT: cmp r0, #0
; CHECK-FP16-NEXT: moveq r3, r0
; CHECK-FP16-NEXT: movne r0, r2
; CHECK-FP16-NEXT: moveq r10, r6
; CHECK-FP16-NEXT: moveq r5, r8
; CHECK-FP16-NEXT: rsbs r1, r5, #0
; CHECK-FP16-NEXT: rscs r1, r10, #-2147483648
; CHECK-FP16-NEXT: sbcs r0, r8, r0
; CHECK-FP16-NEXT: sbcs r0, r8, r3
; CHECK-FP16-NEXT: movwlt r7, #1
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: moveq r5, r7
; CHECK-FP16-NEXT: bl __fixhfti
; CHECK-FP16-NEXT: subs r4, r0, r8
; CHECK-FP16-NEXT: vmov.32 d1[0], r5
; CHECK-FP16-NEXT: sbcs r4, r1, r6
; CHECK-FP16-NEXT: sbcs r4, r2, #0
; CHECK-FP16-NEXT: sbcs r4, r3, #0
; CHECK-FP16-NEXT: mov r4, #0
; CHECK-FP16-NEXT: movwlt r4, #1
; CHECK-FP16-NEXT: cmp r4, #0
; CHECK-FP16-NEXT: moveq r3, r4
; CHECK-FP16-NEXT: movne r6, r1
; CHECK-FP16-NEXT: movne r4, r2
; CHECK-FP16-NEXT: moveq r0, r8
; CHECK-FP16-NEXT: rsbs r1, r0, #0
; CHECK-FP16-NEXT: rscs r1, r6, #-2147483648
; CHECK-FP16-NEXT: sbcs r1, r8, r4
; CHECK-FP16-NEXT: sbcs r1, r8, r3
; CHECK-FP16-NEXT: movwlt r9, #1
; CHECK-FP16-NEXT: cmp r9, #0
; CHECK-FP16-NEXT: moveq r0, r9
; CHECK-FP16-NEXT: mov r1, #-2147483648
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: vmov.32 d0[0], r0
; CHECK-FP16-NEXT: moveq r10, r1
; CHECK-FP16-NEXT: cmp r9, #0
; CHECK-FP16-NEXT: vmov.32 d1[1], r10
; CHECK-FP16-NEXT: moveq r6, r1
; CHECK-FP16-NEXT: vmov.32 d0[1], r6
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, r9, r10, pc}
entry:
%conv = fptosi <2 x half> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.smin.v2i128(<2 x i128> %conv, <2 x i128> <i128 9223372036854775807, i128 9223372036854775807>)
%spec.store.select7 = call <2 x i128> @llvm.smax.v2i128(<2 x i128> %spec.store.select, <2 x i128> <i128 -9223372036854775808, i128 -9223372036854775808>)
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @utesth_f16i64_mm(<2 x half> %x) {
; CHECK-NEON-LABEL: utesth_f16i64_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-NEON-NEXT: .vsave {d8}
; CHECK-NEON-NEXT: vpush {d8}
; CHECK-NEON-NEXT: vmov r0, s0
; CHECK-NEON-NEXT: vmov.f32 s16, s1
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r5, r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: bl __fixunssfti
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: subs r0, r2, #1
; CHECK-NEON-NEXT: vmov s0, r5
; CHECK-NEON-NEXT: sbcs r0, r3, #0
; CHECK-NEON-NEXT: mov r5, #0
; CHECK-NEON-NEXT: mov r4, r1
; CHECK-NEON-NEXT: movwlo r5, #1
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: mov r7, #0
; CHECK-NEON-NEXT: moveq r6, r5
; CHECK-NEON-NEXT: bl __fixunssfti
; CHECK-NEON-NEXT: subs r2, r2, #1
; CHECK-NEON-NEXT: vmov.32 d1[0], r6
; CHECK-NEON-NEXT: sbcs r2, r3, #0
; CHECK-NEON-NEXT: movwlo r7, #1
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: moveq r0, r7
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: movne r5, r4
; CHECK-NEON-NEXT: vmov.32 d0[0], r0
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: vmov.32 d1[1], r5
; CHECK-NEON-NEXT: movne r7, r1
; CHECK-NEON-NEXT: vmov.32 d0[1], r7
; CHECK-NEON-NEXT: vpop {d8}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r11, pc}
;
; CHECK-FP16-LABEL: utesth_f16i64_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r11, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r11, lr}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[1]
; CHECK-FP16-NEXT: vmov.u16 r7, d0[0]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixunshfti
; CHECK-FP16-NEXT: mov r5, r0
; CHECK-FP16-NEXT: subs r0, r2, #1
; CHECK-FP16-NEXT: vmov s0, r7
; CHECK-FP16-NEXT: sbcs r0, r3, #0
; CHECK-FP16-NEXT: mov r7, #0
; CHECK-FP16-NEXT: mov r4, r1
; CHECK-FP16-NEXT: movwlo r7, #1
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: mov r6, #0
; CHECK-FP16-NEXT: moveq r5, r7
; CHECK-FP16-NEXT: bl __fixunshfti
; CHECK-FP16-NEXT: subs r2, r2, #1
; CHECK-FP16-NEXT: vmov.32 d1[0], r5
; CHECK-FP16-NEXT: sbcs r2, r3, #0
; CHECK-FP16-NEXT: movwlo r6, #1
; CHECK-FP16-NEXT: cmp r6, #0
; CHECK-FP16-NEXT: moveq r0, r6
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: movne r7, r4
; CHECK-FP16-NEXT: vmov.32 d0[0], r0
; CHECK-FP16-NEXT: cmp r6, #0
; CHECK-FP16-NEXT: vmov.32 d1[1], r7
; CHECK-FP16-NEXT: movne r6, r1
; CHECK-FP16-NEXT: vmov.32 d0[1], r6
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r11, pc}
entry:
%conv = fptoui <2 x half> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.umin.v2i128(<2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>)
%conv6 = trunc <2 x i128> %spec.store.select to <2 x i64>
ret <2 x i64> %conv6
}
define <2 x i64> @ustest_f16i64_mm(<2 x half> %x) {
; CHECK-NEON-LABEL: ustest_f16i64_mm:
; CHECK-NEON: @ %bb.0: @ %entry
; CHECK-NEON-NEXT: .save {r4, r5, r6, r7, r8, lr}
; CHECK-NEON-NEXT: push {r4, r5, r6, r7, r8, lr}
; CHECK-NEON-NEXT: .vsave {d8}
; CHECK-NEON-NEXT: vpush {d8}
; CHECK-NEON-NEXT: vmov r0, s1
; CHECK-NEON-NEXT: vmov.f32 s16, s0
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: mov r7, r0
; CHECK-NEON-NEXT: vmov r0, s16
; CHECK-NEON-NEXT: bl __aeabi_h2f
; CHECK-NEON-NEXT: vmov s0, r0
; CHECK-NEON-NEXT: bl __fixsfti
; CHECK-NEON-NEXT: mov r6, r0
; CHECK-NEON-NEXT: subs r0, r2, #1
; CHECK-NEON-NEXT: vmov s0, r7
; CHECK-NEON-NEXT: sbcs r0, r3, #0
; CHECK-NEON-NEXT: mov r7, #0
; CHECK-NEON-NEXT: mov r5, r3
; CHECK-NEON-NEXT: movwlt r7, #1
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: moveq r6, r7
; CHECK-NEON-NEXT: moveq r5, r7
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: mov r8, r1
; CHECK-NEON-NEXT: mov r4, #0
; CHECK-NEON-NEXT: movwmi r6, #0
; CHECK-NEON-NEXT: bl __fixsfti
; CHECK-NEON-NEXT: subs r2, r2, #1
; CHECK-NEON-NEXT: vmov.32 d0[0], r6
; CHECK-NEON-NEXT: sbcs r2, r3, #0
; CHECK-NEON-NEXT: movwlt r4, #1
; CHECK-NEON-NEXT: cmp r4, #0
; CHECK-NEON-NEXT: moveq r0, r4
; CHECK-NEON-NEXT: moveq r3, r4
; CHECK-NEON-NEXT: cmp r3, #0
; CHECK-NEON-NEXT: movwmi r0, #0
; CHECK-NEON-NEXT: cmp r4, #0
; CHECK-NEON-NEXT: movne r4, r1
; CHECK-NEON-NEXT: cmp r3, #0
; CHECK-NEON-NEXT: vmov.32 d1[0], r0
; CHECK-NEON-NEXT: movwmi r4, #0
; CHECK-NEON-NEXT: cmp r7, #0
; CHECK-NEON-NEXT: movne r7, r8
; CHECK-NEON-NEXT: cmp r5, #0
; CHECK-NEON-NEXT: vmov.32 d1[1], r4
; CHECK-NEON-NEXT: movwmi r7, #0
; CHECK-NEON-NEXT: vmov.32 d0[1], r7
; CHECK-NEON-NEXT: vpop {d8}
; CHECK-NEON-NEXT: pop {r4, r5, r6, r7, r8, pc}
;
; CHECK-FP16-LABEL: ustest_f16i64_mm:
; CHECK-FP16: @ %bb.0: @ %entry
; CHECK-FP16-NEXT: .save {r4, r5, r6, r7, r8, lr}
; CHECK-FP16-NEXT: push {r4, r5, r6, r7, r8, lr}
; CHECK-FP16-NEXT: vmov.u16 r0, d0[0]
; CHECK-FP16-NEXT: vmov.u16 r7, d0[1]
; CHECK-FP16-NEXT: vmov s0, r0
; CHECK-FP16-NEXT: bl __fixhfti
; CHECK-FP16-NEXT: mov r6, r0
; CHECK-FP16-NEXT: subs r0, r2, #1
; CHECK-FP16-NEXT: vmov s0, r7
; CHECK-FP16-NEXT: sbcs r0, r3, #0
; CHECK-FP16-NEXT: mov r7, #0
; CHECK-FP16-NEXT: mov r5, r3
; CHECK-FP16-NEXT: movwlt r7, #1
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: moveq r6, r7
; CHECK-FP16-NEXT: moveq r5, r7
; CHECK-FP16-NEXT: cmp r5, #0
; CHECK-FP16-NEXT: mov r8, r1
; CHECK-FP16-NEXT: mov r4, #0
; CHECK-FP16-NEXT: movwmi r6, #0
; CHECK-FP16-NEXT: bl __fixhfti
; CHECK-FP16-NEXT: subs r2, r2, #1
; CHECK-FP16-NEXT: vmov.32 d0[0], r6
; CHECK-FP16-NEXT: sbcs r2, r3, #0
; CHECK-FP16-NEXT: movwlt r4, #1
; CHECK-FP16-NEXT: cmp r4, #0
; CHECK-FP16-NEXT: moveq r0, r4
; CHECK-FP16-NEXT: moveq r3, r4
; CHECK-FP16-NEXT: cmp r3, #0
; CHECK-FP16-NEXT: movwmi r0, #0
; CHECK-FP16-NEXT: cmp r4, #0
; CHECK-FP16-NEXT: movne r4, r1
; CHECK-FP16-NEXT: cmp r3, #0
; CHECK-FP16-NEXT: vmov.32 d1[0], r0
; CHECK-FP16-NEXT: movwmi r4, #0
; CHECK-FP16-NEXT: cmp r7, #0
; CHECK-FP16-NEXT: movne r7, r8
; CHECK-FP16-NEXT: cmp r5, #0
; CHECK-FP16-NEXT: vmov.32 d1[1], r4
; CHECK-FP16-NEXT: movwmi r7, #0
; CHECK-FP16-NEXT: vmov.32 d0[1], r7
; CHECK-FP16-NEXT: pop {r4, r5, r6, r7, r8, pc}
entry:
%conv = fptosi <2 x half> %x to <2 x i128>
%spec.store.select = call <2 x i128> @llvm.smin.v2i128(<2 x i128> %conv, <2 x i128> <i128 18446744073709551616, i128 18446744073709551616>)
%spec.store.select7 = call <2 x i128> @llvm.smax.v2i128(<2 x i128> %spec.store.select, <2 x i128> zeroinitializer)
%conv6 = trunc <2 x i128> %spec.store.select7 to <2 x i64>
ret <2 x i64> %conv6
}
declare <2 x i32> @llvm.smin.v2i32(<2 x i32>, <2 x i32>)
declare <2 x i32> @llvm.smax.v2i32(<2 x i32>, <2 x i32>)
declare <2 x i32> @llvm.umin.v2i32(<2 x i32>, <2 x i32>)
declare <4 x i32> @llvm.smin.v4i32(<4 x i32>, <4 x i32>)
declare <4 x i32> @llvm.smax.v4i32(<4 x i32>, <4 x i32>)
declare <4 x i32> @llvm.umin.v4i32(<4 x i32>, <4 x i32>)
declare <8 x i32> @llvm.smin.v8i32(<8 x i32>, <8 x i32>)
declare <8 x i32> @llvm.smax.v8i32(<8 x i32>, <8 x i32>)
declare <8 x i32> @llvm.umin.v8i32(<8 x i32>, <8 x i32>)
declare <2 x i64> @llvm.smin.v2i64(<2 x i64>, <2 x i64>)
declare <2 x i64> @llvm.smax.v2i64(<2 x i64>, <2 x i64>)
declare <2 x i64> @llvm.umin.v2i64(<2 x i64>, <2 x i64>)
declare <4 x i64> @llvm.smin.v4i64(<4 x i64>, <4 x i64>)
declare <4 x i64> @llvm.smax.v4i64(<4 x i64>, <4 x i64>)
declare <4 x i64> @llvm.umin.v4i64(<4 x i64>, <4 x i64>)
declare <2 x i128> @llvm.smin.v2i128(<2 x i128>, <2 x i128>)
declare <2 x i128> @llvm.smax.v2i128(<2 x i128>, <2 x i128>)
declare <2 x i128> @llvm.umin.v2i128(<2 x i128>, <2 x i128>)
Codebase Browser
llvm