87 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp -enable-arm-maskedgatscat=false %s -o - | FileCheck --check-prefix NOGATSCAT %s3; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=-mve %s -o - | FileCheck --check-prefix NOMVE %s4 5define arm_aapcs_vfpcc <4 x i32> @unscaled_i32_i32_gather(ptr %base, ptr %offptr) {6; NOGATSCAT-LABEL: unscaled_i32_i32_gather:7; NOGATSCAT: @ %bb.0: @ %entry8; NOGATSCAT-NEXT: vldrw.u32 q0, [r1]9; NOGATSCAT-NEXT: vadd.i32 q0, q0, r010; NOGATSCAT-NEXT: vmov r0, r1, d111; NOGATSCAT-NEXT: vmov r2, r3, d012; NOGATSCAT-NEXT: ldr r0, [r0]13; NOGATSCAT-NEXT: ldr r2, [r2]14; NOGATSCAT-NEXT: ldr r1, [r1]15; NOGATSCAT-NEXT: ldr r3, [r3]16; NOGATSCAT-NEXT: vmov q0[2], q0[0], r2, r017; NOGATSCAT-NEXT: vmov q0[3], q0[1], r3, r118; NOGATSCAT-NEXT: bx lr19;20; NOMVE-LABEL: unscaled_i32_i32_gather:21; NOMVE: @ %bb.0: @ %entry22; NOMVE-NEXT: .save {r4, lr}23; NOMVE-NEXT: push {r4, lr}24; NOMVE-NEXT: ldm.w r1, {r2, r3, lr}25; NOMVE-NEXT: ldr r4, [r1, #12]26; NOMVE-NEXT: ldr.w r12, [r0, r2]27; NOMVE-NEXT: ldr r1, [r0, r3]28; NOMVE-NEXT: ldr.w r2, [r0, lr]29; NOMVE-NEXT: ldr r3, [r0, r4]30; NOMVE-NEXT: mov r0, r1231; NOMVE-NEXT: pop {r4, pc}32 33entry:34 %offs = load <4 x i32>, ptr %offptr, align 435 %byte_ptrs = getelementptr inbounds i8, ptr %base, <4 x i32> %offs36 %ptrs = bitcast <4 x ptr> %byte_ptrs to <4 x ptr>37 %gather = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)38 ret <4 x i32> %gather39}40 41declare <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr>, i32, <4 x i1>, <4 x i32>)42 43 44define arm_aapcs_vfpcc void @unscaled_i32_i8_scatter(ptr %base, ptr %offptr, <4 x i32> %input) {45; NOGATSCAT-LABEL: unscaled_i32_i8_scatter:46; NOGATSCAT: @ %bb.0: @ %entry47; NOGATSCAT-NEXT: .save {r4, r5, r7, lr}48; NOGATSCAT-NEXT: push {r4, r5, r7, lr}49; NOGATSCAT-NEXT: vldrb.u32 q1, [r1]50; NOGATSCAT-NEXT: vmov r1, r3, d051; NOGATSCAT-NEXT: vmov r4, r5, d152; NOGATSCAT-NEXT: vadd.i32 q1, q1, r053; NOGATSCAT-NEXT: vmov r0, r12, d254; NOGATSCAT-NEXT: vmov r2, lr, d355; NOGATSCAT-NEXT: str r1, [r0]56; NOGATSCAT-NEXT: str.w r3, [r12]57; NOGATSCAT-NEXT: str r4, [r2]58; NOGATSCAT-NEXT: str.w r5, [lr]59; NOGATSCAT-NEXT: pop {r4, r5, r7, pc}60;61; NOMVE-LABEL: unscaled_i32_i8_scatter:62; NOMVE: @ %bb.0: @ %entry63; NOMVE-NEXT: .save {r4, lr}64; NOMVE-NEXT: push {r4, lr}65; NOMVE-NEXT: ldrb.w r12, [r1]66; NOMVE-NEXT: ldrb.w lr, [r1, #1]67; NOMVE-NEXT: ldrb r4, [r1, #2]68; NOMVE-NEXT: ldrb r1, [r1, #3]69; NOMVE-NEXT: str.w r2, [r0, r12]70; NOMVE-NEXT: ldr r2, [sp, #8]71; NOMVE-NEXT: str.w r3, [r0, lr]72; NOMVE-NEXT: str r2, [r0, r4]73; NOMVE-NEXT: ldr r2, [sp, #12]74; NOMVE-NEXT: str r2, [r0, r1]75; NOMVE-NEXT: pop {r4, pc}76 77entry:78 %offs = load <4 x i8>, ptr %offptr, align 179 %offs.zext = zext <4 x i8> %offs to <4 x i32>80 %byte_ptrs = getelementptr inbounds i8, ptr %base, <4 x i32> %offs.zext81 %ptrs = bitcast <4 x ptr> %byte_ptrs to <4 x ptr>82 call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> %input, <4 x ptr> %ptrs, i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>)83 ret void84}85 86declare void @llvm.masked.scatter.v4i32.v4p0(<4 x i32>, <4 x ptr>, i32, <4 x i1>)87