1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 12:12:47 +01:00
llvm-mirror/test/CodeGen/Thumb2/mve-pred-loadstore.ll
David Green 17e932c916 [ARM] Create VMOVRRD from adjacent vector extracts
This adds a combine for extract(x, n); extract(x, n+1)  ->
VMOVRRD(extract x, n/2). This allows two vector lanes to be moved at the
same time in a single instruction, and thanks to the other VMOVRRD folds
we have added recently can help reduce the amount of executed
instructions. Floating point types are very similar, but will include a
bitcast to an integer type.

This also adds a shouldRewriteCopySrc, to prevent copy propagation from
DPR to SPR, which can break as not all DPR regs can be extracted from
directly.  Otherwise the machine verifier is unhappy.

Differential Revision: https://reviews.llvm.org/D100244
2021-04-20 15:15:43 +01:00

622 lines
21 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=CHECK-LE
; RUN: llc -mtriple=thumbebv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=CHECK-BE
define arm_aapcs_vfpcc <4 x i32> @load_v4i1(<4 x i1> *%src, <4 x i32> %a) {
; CHECK-LE-LABEL: load_v4i1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: ldrb r0, [r0]
; CHECK-LE-NEXT: vmov.i8 q1, #0x0
; CHECK-LE-NEXT: vmov.i8 q2, #0xff
; CHECK-LE-NEXT: vmsr p0, r0
; CHECK-LE-NEXT: vpsel q1, q2, q1
; CHECK-LE-NEXT: vmov.u8 r0, q1[2]
; CHECK-LE-NEXT: vmov.u8 r1, q1[0]
; CHECK-LE-NEXT: vmov q2[2], q2[0], r1, r0
; CHECK-LE-NEXT: vmov.u8 r0, q1[3]
; CHECK-LE-NEXT: vmov.u8 r1, q1[1]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vmov q2[3], q2[1], r1, r0
; CHECK-LE-NEXT: vcmp.i32 ne, q2, zr
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_v4i1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: ldrb r0, [r0]
; CHECK-BE-NEXT: vmov.i8 q1, #0x0
; CHECK-BE-NEXT: vmov.i8 q2, #0xff
; CHECK-BE-NEXT: rbit r0, r0
; CHECK-BE-NEXT: lsrs r0, r0, #28
; CHECK-BE-NEXT: vmsr p0, r0
; CHECK-BE-NEXT: vpsel q1, q2, q1
; CHECK-BE-NEXT: vmov.u8 r0, q1[2]
; CHECK-BE-NEXT: vmov.u8 r1, q1[0]
; CHECK-BE-NEXT: vmov q2[2], q2[0], r1, r0
; CHECK-BE-NEXT: vmov.u8 r0, q1[3]
; CHECK-BE-NEXT: vmov.u8 r1, q1[1]
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov q2[3], q2[1], r1, r0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vcmp.i32 ne, q2, zr
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
entry:
%c = load <4 x i1>, <4 x i1>* %src
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <8 x i16> @load_v8i1(<8 x i1> *%src, <8 x i16> %a) {
; CHECK-LE-LABEL: load_v8i1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: ldrb r0, [r0]
; CHECK-LE-NEXT: vmov.i8 q1, #0x0
; CHECK-LE-NEXT: vmov.i8 q2, #0xff
; CHECK-LE-NEXT: vmsr p0, r0
; CHECK-LE-NEXT: vpsel q2, q2, q1
; CHECK-LE-NEXT: vmov.u8 r0, q2[0]
; CHECK-LE-NEXT: vmov.16 q1[0], r0
; CHECK-LE-NEXT: vmov.u8 r0, q2[1]
; CHECK-LE-NEXT: vmov.16 q1[1], r0
; CHECK-LE-NEXT: vmov.u8 r0, q2[2]
; CHECK-LE-NEXT: vmov.16 q1[2], r0
; CHECK-LE-NEXT: vmov.u8 r0, q2[3]
; CHECK-LE-NEXT: vmov.16 q1[3], r0
; CHECK-LE-NEXT: vmov.u8 r0, q2[4]
; CHECK-LE-NEXT: vmov.16 q1[4], r0
; CHECK-LE-NEXT: vmov.u8 r0, q2[5]
; CHECK-LE-NEXT: vmov.16 q1[5], r0
; CHECK-LE-NEXT: vmov.u8 r0, q2[6]
; CHECK-LE-NEXT: vmov.16 q1[6], r0
; CHECK-LE-NEXT: vmov.u8 r0, q2[7]
; CHECK-LE-NEXT: vmov.16 q1[7], r0
; CHECK-LE-NEXT: vcmp.i16 ne, q1, zr
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_v8i1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: ldrb r0, [r0]
; CHECK-BE-NEXT: vmov.i8 q1, #0x0
; CHECK-BE-NEXT: vmov.i8 q2, #0xff
; CHECK-BE-NEXT: rbit r0, r0
; CHECK-BE-NEXT: lsrs r0, r0, #24
; CHECK-BE-NEXT: vmsr p0, r0
; CHECK-BE-NEXT: vpsel q2, q2, q1
; CHECK-BE-NEXT: vmov.u8 r0, q2[0]
; CHECK-BE-NEXT: vmov.16 q1[0], r0
; CHECK-BE-NEXT: vmov.u8 r0, q2[1]
; CHECK-BE-NEXT: vmov.16 q1[1], r0
; CHECK-BE-NEXT: vmov.u8 r0, q2[2]
; CHECK-BE-NEXT: vmov.16 q1[2], r0
; CHECK-BE-NEXT: vmov.u8 r0, q2[3]
; CHECK-BE-NEXT: vmov.16 q1[3], r0
; CHECK-BE-NEXT: vmov.u8 r0, q2[4]
; CHECK-BE-NEXT: vmov.16 q1[4], r0
; CHECK-BE-NEXT: vmov.u8 r0, q2[5]
; CHECK-BE-NEXT: vmov.16 q1[5], r0
; CHECK-BE-NEXT: vmov.u8 r0, q2[6]
; CHECK-BE-NEXT: vmov.16 q1[6], r0
; CHECK-BE-NEXT: vmov.u8 r0, q2[7]
; CHECK-BE-NEXT: vmov.16 q1[7], r0
; CHECK-BE-NEXT: vcmp.i16 ne, q1, zr
; CHECK-BE-NEXT: vrev64.16 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vrev32.16 q0, q0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.16 q0, q1
; CHECK-BE-NEXT: bx lr
entry:
%c = load <8 x i1>, <8 x i1>* %src
%s = select <8 x i1> %c, <8 x i16> %a, <8 x i16> zeroinitializer
ret <8 x i16> %s
}
define arm_aapcs_vfpcc <16 x i8> @load_v16i1(<16 x i1> *%src, <16 x i8> %a) {
; CHECK-LE-LABEL: load_v16i1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: ldrh r0, [r0]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vmsr p0, r0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_v16i1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: ldrh r0, [r0]
; CHECK-BE-NEXT: vrev64.8 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: rbit r0, r0
; CHECK-BE-NEXT: vrev32.8 q0, q0
; CHECK-BE-NEXT: lsrs r0, r0, #16
; CHECK-BE-NEXT: vmsr p0, r0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.8 q0, q1
; CHECK-BE-NEXT: bx lr
entry:
%c = load <16 x i1>, <16 x i1>* %src
%s = select <16 x i1> %c, <16 x i8> %a, <16 x i8> zeroinitializer
ret <16 x i8> %s
}
define arm_aapcs_vfpcc <2 x i64> @load_v2i1(<2 x i1> *%src, <2 x i64> %a) {
; CHECK-LE-LABEL: load_v2i1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: ldrb r0, [r0]
; CHECK-LE-NEXT: ubfx r1, r0, #1, #1
; CHECK-LE-NEXT: and r0, r0, #1
; CHECK-LE-NEXT: rsbs r1, r1, #0
; CHECK-LE-NEXT: rsbs r0, r0, #0
; CHECK-LE-NEXT: vmov q1[2], q1[0], r0, r1
; CHECK-LE-NEXT: vmov q1[3], q1[1], r0, r1
; CHECK-LE-NEXT: vand q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_v2i1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: ldrb r0, [r0]
; CHECK-BE-NEXT: and r1, r0, #1
; CHECK-BE-NEXT: ubfx r0, r0, #1, #1
; CHECK-BE-NEXT: rsbs r1, r1, #0
; CHECK-BE-NEXT: rsbs r0, r0, #0
; CHECK-BE-NEXT: vmov q1[2], q1[0], r0, r1
; CHECK-BE-NEXT: vmov q1[3], q1[1], r0, r1
; CHECK-BE-NEXT: vrev64.32 q2, q1
; CHECK-BE-NEXT: vand q0, q0, q2
; CHECK-BE-NEXT: bx lr
entry:
%c = load <2 x i1>, <2 x i1>* %src
%s = select <2 x i1> %c, <2 x i64> %a, <2 x i64> zeroinitializer
ret <2 x i64> %s
}
define arm_aapcs_vfpcc void @store_v4i1(<4 x i1> *%dst, <4 x i32> %a) {
; CHECK-LE-LABEL: store_v4i1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vcmp.i32 eq, q0, zr
; CHECK-LE-NEXT: movs r3, #0
; CHECK-LE-NEXT: vmrs r1, p0
; CHECK-LE-NEXT: and r2, r1, #1
; CHECK-LE-NEXT: rsbs r2, r2, #0
; CHECK-LE-NEXT: bfi r3, r2, #0, #1
; CHECK-LE-NEXT: ubfx r2, r1, #4, #1
; CHECK-LE-NEXT: rsbs r2, r2, #0
; CHECK-LE-NEXT: bfi r3, r2, #1, #1
; CHECK-LE-NEXT: ubfx r2, r1, #8, #1
; CHECK-LE-NEXT: ubfx r1, r1, #12, #1
; CHECK-LE-NEXT: rsbs r2, r2, #0
; CHECK-LE-NEXT: bfi r3, r2, #2, #1
; CHECK-LE-NEXT: rsbs r1, r1, #0
; CHECK-LE-NEXT: bfi r3, r1, #3, #1
; CHECK-LE-NEXT: strb r3, [r0]
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: store_v4i1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: movs r3, #0
; CHECK-BE-NEXT: vcmp.i32 eq, q1, zr
; CHECK-BE-NEXT: vmrs r1, p0
; CHECK-BE-NEXT: ubfx r2, r1, #12, #1
; CHECK-BE-NEXT: rsbs r2, r2, #0
; CHECK-BE-NEXT: bfi r3, r2, #0, #1
; CHECK-BE-NEXT: ubfx r2, r1, #8, #1
; CHECK-BE-NEXT: rsbs r2, r2, #0
; CHECK-BE-NEXT: bfi r3, r2, #1, #1
; CHECK-BE-NEXT: ubfx r2, r1, #4, #1
; CHECK-BE-NEXT: and r1, r1, #1
; CHECK-BE-NEXT: rsbs r2, r2, #0
; CHECK-BE-NEXT: bfi r3, r2, #2, #1
; CHECK-BE-NEXT: rsbs r1, r1, #0
; CHECK-BE-NEXT: bfi r3, r1, #3, #1
; CHECK-BE-NEXT: strb r3, [r0]
; CHECK-BE-NEXT: bx lr
entry:
%c = icmp eq <4 x i32> %a, zeroinitializer
store <4 x i1> %c, <4 x i1>* %dst
ret void
}
define arm_aapcs_vfpcc void @store_v8i1(<8 x i1> *%dst, <8 x i16> %a) {
; CHECK-LE-LABEL: store_v8i1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vcmp.i16 eq, q0, zr
; CHECK-LE-NEXT: vmrs r2, p0
; CHECK-LE-NEXT: and r1, r2, #1
; CHECK-LE-NEXT: rsbs r3, r1, #0
; CHECK-LE-NEXT: movs r1, #0
; CHECK-LE-NEXT: bfi r1, r3, #0, #1
; CHECK-LE-NEXT: ubfx r3, r2, #2, #1
; CHECK-LE-NEXT: rsbs r3, r3, #0
; CHECK-LE-NEXT: bfi r1, r3, #1, #1
; CHECK-LE-NEXT: ubfx r3, r2, #4, #1
; CHECK-LE-NEXT: rsbs r3, r3, #0
; CHECK-LE-NEXT: bfi r1, r3, #2, #1
; CHECK-LE-NEXT: ubfx r3, r2, #6, #1
; CHECK-LE-NEXT: rsbs r3, r3, #0
; CHECK-LE-NEXT: bfi r1, r3, #3, #1
; CHECK-LE-NEXT: ubfx r3, r2, #8, #1
; CHECK-LE-NEXT: rsbs r3, r3, #0
; CHECK-LE-NEXT: bfi r1, r3, #4, #1
; CHECK-LE-NEXT: ubfx r3, r2, #10, #1
; CHECK-LE-NEXT: rsbs r3, r3, #0
; CHECK-LE-NEXT: bfi r1, r3, #5, #1
; CHECK-LE-NEXT: ubfx r3, r2, #12, #1
; CHECK-LE-NEXT: ubfx r2, r2, #14, #1
; CHECK-LE-NEXT: rsbs r3, r3, #0
; CHECK-LE-NEXT: bfi r1, r3, #6, #1
; CHECK-LE-NEXT: rsbs r2, r2, #0
; CHECK-LE-NEXT: bfi r1, r2, #7, #1
; CHECK-LE-NEXT: strb r1, [r0]
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: store_v8i1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.16 q1, q0
; CHECK-BE-NEXT: vcmp.i16 eq, q1, zr
; CHECK-BE-NEXT: vmrs r2, p0
; CHECK-BE-NEXT: ubfx r1, r2, #14, #1
; CHECK-BE-NEXT: rsbs r3, r1, #0
; CHECK-BE-NEXT: movs r1, #0
; CHECK-BE-NEXT: bfi r1, r3, #0, #1
; CHECK-BE-NEXT: ubfx r3, r2, #12, #1
; CHECK-BE-NEXT: rsbs r3, r3, #0
; CHECK-BE-NEXT: bfi r1, r3, #1, #1
; CHECK-BE-NEXT: ubfx r3, r2, #10, #1
; CHECK-BE-NEXT: rsbs r3, r3, #0
; CHECK-BE-NEXT: bfi r1, r3, #2, #1
; CHECK-BE-NEXT: ubfx r3, r2, #8, #1
; CHECK-BE-NEXT: rsbs r3, r3, #0
; CHECK-BE-NEXT: bfi r1, r3, #3, #1
; CHECK-BE-NEXT: ubfx r3, r2, #6, #1
; CHECK-BE-NEXT: rsbs r3, r3, #0
; CHECK-BE-NEXT: bfi r1, r3, #4, #1
; CHECK-BE-NEXT: ubfx r3, r2, #4, #1
; CHECK-BE-NEXT: rsbs r3, r3, #0
; CHECK-BE-NEXT: bfi r1, r3, #5, #1
; CHECK-BE-NEXT: ubfx r3, r2, #2, #1
; CHECK-BE-NEXT: and r2, r2, #1
; CHECK-BE-NEXT: rsbs r3, r3, #0
; CHECK-BE-NEXT: bfi r1, r3, #6, #1
; CHECK-BE-NEXT: rsbs r2, r2, #0
; CHECK-BE-NEXT: bfi r1, r2, #7, #1
; CHECK-BE-NEXT: strb r1, [r0]
; CHECK-BE-NEXT: bx lr
entry:
%c = icmp eq <8 x i16> %a, zeroinitializer
store <8 x i1> %c, <8 x i1>* %dst
ret void
}
define arm_aapcs_vfpcc void @store_v16i1(<16 x i1> *%dst, <16 x i8> %a) {
; CHECK-LE-LABEL: store_v16i1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vcmp.i8 eq, q0, zr
; CHECK-LE-NEXT: vmrs r1, p0
; CHECK-LE-NEXT: strh r1, [r0]
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: store_v16i1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.8 q1, q0
; CHECK-BE-NEXT: vcmp.i8 eq, q1, zr
; CHECK-BE-NEXT: vmrs r1, p0
; CHECK-BE-NEXT: rbit r1, r1
; CHECK-BE-NEXT: lsrs r1, r1, #16
; CHECK-BE-NEXT: strh r1, [r0]
; CHECK-BE-NEXT: bx lr
entry:
%c = icmp eq <16 x i8> %a, zeroinitializer
store <16 x i1> %c, <16 x i1>* %dst
ret void
}
define arm_aapcs_vfpcc void @store_v2i1(<2 x i1> *%dst, <2 x i64> %a) {
; CHECK-LE-LABEL: store_v2i1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vmov r1, r2, d0
; CHECK-LE-NEXT: orrs r1, r2
; CHECK-LE-NEXT: vmov r2, r3, d1
; CHECK-LE-NEXT: cset r1, eq
; CHECK-LE-NEXT: orrs r2, r3
; CHECK-LE-NEXT: cset r2, eq
; CHECK-LE-NEXT: cmp r2, #0
; CHECK-LE-NEXT: it ne
; CHECK-LE-NEXT: mvnne r2, #1
; CHECK-LE-NEXT: bfi r2, r1, #0, #1
; CHECK-LE-NEXT: and r1, r2, #3
; CHECK-LE-NEXT: strb r1, [r0]
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: store_v2i1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov r1, r2, d3
; CHECK-BE-NEXT: orrs r1, r2
; CHECK-BE-NEXT: vmov r2, r3, d2
; CHECK-BE-NEXT: cset r1, eq
; CHECK-BE-NEXT: orrs r2, r3
; CHECK-BE-NEXT: cset r2, eq
; CHECK-BE-NEXT: cmp r2, #0
; CHECK-BE-NEXT: it ne
; CHECK-BE-NEXT: mvnne r2, #1
; CHECK-BE-NEXT: bfi r2, r1, #0, #1
; CHECK-BE-NEXT: and r1, r2, #3
; CHECK-BE-NEXT: strb r1, [r0]
; CHECK-BE-NEXT: bx lr
entry:
%c = icmp eq <2 x i64> %a, zeroinitializer
store <2 x i1> %c, <2 x i1>* %dst
ret void
}
define arm_aapcs_vfpcc <4 x i32> @load_predcastzext(i16* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_predcastzext:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: ldrh r0, [r0]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vmsr p0, r0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_predcastzext:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: ldrh r0, [r0]
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vmsr p0, r0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%l = load i16, i16* %i, align 4
%lz = zext i16 %l to i32
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %lz)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <4 x i32> @load_bc4(i32* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_bc4:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: vldr p0, [r0]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_bc4:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: vldr p0, [r0]
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%l = load i32, i32* %i, align 4
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %l)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <8 x i16> @load_predcast8(i32* %i, <8 x i16> %a) {
; CHECK-LE-LABEL: load_predcast8:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: vldr p0, [r0]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_predcast8:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: vrev64.16 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vldr p0, [r0]
; CHECK-BE-NEXT: vrev32.16 q0, q0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.16 q0, q1
; CHECK-BE-NEXT: bx lr
%l = load i32, i32* %i, align 4
%c = tail call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 %l)
%s = select <8 x i1> %c, <8 x i16> %a, <8 x i16> zeroinitializer
ret <8 x i16> %s
}
define arm_aapcs_vfpcc <16 x i8> @load_predcast16(i32* %i, <16 x i8> %a) {
; CHECK-LE-LABEL: load_predcast16:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: vldr p0, [r0]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_predcast16:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: vrev64.8 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vldr p0, [r0]
; CHECK-BE-NEXT: vrev32.8 q0, q0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.8 q0, q1
; CHECK-BE-NEXT: bx lr
%l = load i32, i32* %i, align 4
%c = tail call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 %l)
%s = select <16 x i1> %c, <16 x i8> %a, <16 x i8> zeroinitializer
ret <16 x i8> %s
}
define arm_aapcs_vfpcc <4 x i32> @load_bc4_align2(i32* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_bc4_align2:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: ldr r0, [r0]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vmsr p0, r0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_bc4_align2:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: ldr r0, [r0]
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vmsr p0, r0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%l = load i32, i32* %i, align 2
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %l)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <4 x i32> @load_bc4_offset(i16* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_bc4_offset:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: adds r0, #6
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vldr p0, [r0]
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_bc4_offset:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: adds r0, #6
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vldr p0, [r0]
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%g = getelementptr inbounds i16, i16* %i, i32 3
%gb = bitcast i16* %g to i32*
%l = load i32, i32* %gb, align 4
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %l)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <4 x i32> @load_bc4_range4(i32* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_bc4_range4:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: vldr p0, [r0, #4]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_bc4_range4:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: vldr p0, [r0, #4]
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%g = getelementptr inbounds i32, i32* %i, i32 1
%l = load i32, i32* %g, align 4
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %l)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <4 x i32> @load_bc4_range(i32* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_bc4_range:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: vldr p0, [r0, #508]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_bc4_range:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: vldr p0, [r0, #508]
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%g = getelementptr inbounds i32, i32* %i, i32 127
%l = load i32, i32* %g, align 4
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %l)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <4 x i32> @load_bc4_range2(i32* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_bc4_range2:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: vldr p0, [r0, #-508]
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_bc4_range2:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: vldr p0, [r0, #-508]
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%g = getelementptr inbounds i32, i32* %i, i32 -127
%l = load i32, i32* %g, align 4
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %l)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <4 x i32> @load_bc4_range3(i32* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_bc4_range3:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: add.w r0, r0, #512
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vldr p0, [r0]
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_bc4_range3:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: add.w r0, r0, #512
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vldr p0, [r0]
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%g = getelementptr inbounds i32, i32* %i, i32 128
%l = load i32, i32* %g, align 4
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %l)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
define arm_aapcs_vfpcc <4 x i32> @load_bc4_range5(i32* %i, <4 x i32> %a) {
; CHECK-LE-LABEL: load_bc4_range5:
; CHECK-LE: @ %bb.0:
; CHECK-LE-NEXT: sub.w r0, r0, #512
; CHECK-LE-NEXT: vmov.i32 q1, #0x0
; CHECK-LE-NEXT: vldr p0, [r0]
; CHECK-LE-NEXT: vpsel q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_bc4_range5:
; CHECK-BE: @ %bb.0:
; CHECK-BE-NEXT: sub.w r0, r0, #512
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vldr p0, [r0]
; CHECK-BE-NEXT: vmov.i32 q0, #0x0
; CHECK-BE-NEXT: vpsel q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
%g = getelementptr inbounds i32, i32* %i, i32 -128
%l = load i32, i32* %g, align 4
%c = tail call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %l)
%s = select <4 x i1> %c, <4 x i32> %a, <4 x i32> zeroinitializer
ret <4 x i32> %s
}
declare <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32)
declare <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32)
declare <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32)