68 lines · plain
1; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -amdgpu-spill-sgpr-to-vgpr=0 -stop-after prologepilog -verify-machineinstrs %s -o - | FileCheck -check-prefix=AFTER-PEI %s2 3; Test that the ScavengeFI is serialized in the SIMachineFunctionInfo.4 5; AFTER-PEI-LABEL: {{^}}name: scavenge_fi6; AFTER-PEI: machineFunctionInfo:7; AFTER-PEI-NEXT: explicitKernArgSize: 128; AFTER-PEI-NEXT: maxKernArgAlign: 89; AFTER-PEI-NEXT: ldsSize: 010; AFTER-PEI-NEXT: gdsSize: 011; AFTER-PEI-NEXT: dynLDSAlign: 112; AFTER-PEI-NEXT: isEntryFunction: true13; AFTER-PEI-NEXT: isChainFunction: false14; AFTER-PEI-NEXT: noSignedZerosFPMath: false15; AFTER-PEI-NEXT: memoryBound: false16; AFTER-PEI-NEXT: waveLimiter: false17; AFTER-PEI-NEXT: hasSpilledSGPRs: true18; AFTER-PEI-NEXT: hasSpilledVGPRs: false19; AFTER-PEI-NEXT: numWaveDispatchSGPRs: 020; AFTER-PEI-NEXT: numWaveDispatchVGPRs: 021; AFTER-PEI-NEXT: scratchRSrcReg: '$sgpr68_sgpr69_sgpr70_sgpr71'22; AFTER-PEI-NEXT: frameOffsetReg: '$fp_reg'23; AFTER-PEI-NEXT: stackPtrOffsetReg: '$sgpr32'24; AFTER-PEI-NEXT: bytesInStackArgArea: 025; AFTER-PEI-NEXT: returnsVoid: true26; AFTER-PEI-NEXT: argumentInfo:27; AFTER-PEI-NEXT: privateSegmentBuffer: { reg: '$sgpr0_sgpr1_sgpr2_sgpr3' }28; AFTER-PEI-NEXT: kernargSegmentPtr: { reg: '$sgpr4_sgpr5' }29; AFTER-PEI-NEXT: workGroupIDX: { reg: '$sgpr6' }30; AFTER-PEI-NEXT: privateSegmentWaveByteOffset: { reg: '$sgpr7' }31; AFTER-PEI-NEXT: workItemIDX: { reg: '$vgpr0' }32; AFTER-PEI-NEXT: psInputAddr: 033; AFTER-PEI-NEXT: psInputEnable: 034; AFTER-PEI-NEXT: maxMemoryClusterDWords: 835; AFTER-PEI-NEXT: mode:36; AFTER-PEI-NEXT: ieee: true37; AFTER-PEI-NEXT: dx10-clamp: true38; AFTER-PEI-NEXT: fp32-input-denormals: true39; AFTER-PEI-NEXT: fp32-output-denormals: true40; AFTER-PEI-NEXT: fp64-fp16-input-denormals: true41; AFTER-PEI-NEXT: fp64-fp16-output-denormals: true42; AFTER-PEI-NEXT: highBitsOf32BitAddress: 043; AFTER-PEI-NEXT: occupancy: 544; AFTER-PEI-NEXT: scavengeFI: '%stack.3'45; AFTER-PEI-NEXT: vgprForAGPRCopy: ''46; AFTER-PEI-NEXT: sgprForEXECCopy: ''47; AFTER-PEI-NEXT: longBranchReservedReg: ''48; AFTER-PEI-NEXT: hasInitWholeWave: false49; AFTER-PEI-NEXT: dynamicVGPRBlockSize: 050; AFTER-PEI-NEXT: scratchReservedForDynamicVGPRs: 051; AFTER-PEI-NEXT: numKernargPreloadSGPRs: 052; AFTER-PEI-NEXT: isWholeWaveFunction: false53; AFTER-PEI-NEXT: body:54define amdgpu_kernel void @scavenge_fi(ptr addrspace(1) %out, i32 %in) #0 {55 %wide.sgpr0 = call <32 x i32> asm sideeffect "; def $0", "=s" () #056 %wide.sgpr1 = call <32 x i32> asm sideeffect "; def $0", "=s" () #057 %wide.sgpr2 = call <32 x i32> asm sideeffect "; def $0", "=s" () #058 %wide.sgpr3 = call <32 x i32> asm sideeffect "; def $0", "=s" () #059 60 call void asm sideeffect "; use $0", "s"(<32 x i32> %wide.sgpr0) #061 call void asm sideeffect "; use $0", "s"(<32 x i32> %wide.sgpr1) #062 call void asm sideeffect "; use $0", "s"(<32 x i32> %wide.sgpr2) #063 call void asm sideeffect "; use $0", "s"(<32 x i32> %wide.sgpr3) #064 ret void65}66 67attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }68