1954ab83eSAustin Kerbow; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 -filetype=obj < %s | llvm-objdump -s -j .rodata - | FileCheck --check-prefix=OBJDUMP %s 2954ab83eSAustin Kerbow; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx940 < %s | FileCheck --check-prefix=ASM %s 3954ab83eSAustin Kerbow 4954ab83eSAustin Kerbow; OBJDUMP: Contents of section .rodata: 5954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0000 00000000 00000000 10010000 00000000 ................ 6954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0010 00000000 00000000 00000000 00000000 ................ 7954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0020 00000000 00000000 00000000 00000000 ................ 8954ab83eSAustin Kerbow; OBJDUMP-NOT: 0030 0000af00 94130000 1a000400 00000000 ................ 9*c93e001cSShilei Tian; OBJDUMP-NEXT: 0030 8000af00 98130000 1e000400 00000000 ................ 10954ab83eSAustin Kerbow 11954ab83eSAustin Kerbow; ASM-LABEL: amdhsa_kernarg_preload_4_implicit_6: 126548b635SShilei Tian; ASM: .amdhsa_user_sgpr_count 12 136548b635SShilei Tian; ASM: .amdhsa_next_free_sgpr 12 146548b635SShilei Tian; ASM: ; TotalNumSgprs: 18 156548b635SShilei Tian; ASM: ; NumSGPRsForWavesPerEU: 18 16954ab83eSAustin Kerbow 17954ab83eSAustin Kerbow; Test that we include preloaded SGPRs in the GRANULATED_WAVEFRONT_SGPR_COUNT 18954ab83eSAustin Kerbow; feild that are not explicitly referenced in the kernel. This test has 6 implicit 19954ab83eSAustin Kerbow; user SPGRs enabled, 4 preloaded kernarg SGPRs, plus 6 extra SGPRs allocated 20954ab83eSAustin Kerbow; for flat scratch, ect. The total number of allocated SGPRs encoded in the 21954ab83eSAustin Kerbow; kernel descriptor should be 16. That's a 1 in the KD field since the granule 22954ab83eSAustin Kerbow; size is 8 and it's NumGranules - 1. The encoding for that looks like '40'. 23954ab83eSAustin Kerbow 24954ab83eSAustin Kerbowdefine amdgpu_kernel void @amdhsa_kernarg_preload_4_implicit_6(i128 inreg) { ret void } 25954ab83eSAustin Kerbow 26954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0040 00000000 00000000 20010000 00000000 ........ ....... 27954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0050 00000000 00000000 00000000 00000000 ................ 28954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0060 00000000 00000000 00000000 00000000 ................ 29954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0070 4000af00 94000000 08000800 00000000 @............... 30954ab83eSAustin Kerbow 31954ab83eSAustin Kerbow; ASM-LABEL: amdhsa_kernarg_preload_8_implicit_2: 32954ab83eSAustin Kerbow; ASM: .amdhsa_user_sgpr_count 10 33954ab83eSAustin Kerbow; ASM: .amdhsa_next_free_sgpr 10 34c897c13dSJanek van Oirschot; ASM: ; TotalNumSgprs: 16 35954ab83eSAustin Kerbow; ASM: ; NumSGPRsForWavesPerEU: 16 36954ab83eSAustin Kerbow 37954ab83eSAustin Kerbow; Only the kernarg_ptr is enabled so we should have 8 preload kernarg SGPRs, 2 38954ab83eSAustin Kerbow; implicit, and 6 extra. 39954ab83eSAustin Kerbow 40954ab83eSAustin Kerbowdefine amdgpu_kernel void @amdhsa_kernarg_preload_8_implicit_2(i256 inreg) #0 { ret void } 41954ab83eSAustin Kerbow 42954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0080 00000000 00000000 08010000 00000000 ................ 43954ab83eSAustin Kerbow; OBJDUMP-NEXT: 0090 00000000 00000000 00000000 00000000 ................ 44954ab83eSAustin Kerbow; OBJDUMP-NEXT: 00a0 00000000 00000000 00000000 00000000 ................ 45954ab83eSAustin Kerbow; OBJDUMP-NEXT: 00b0 4000af00 86000000 08000100 00000000 @............... 46954ab83eSAustin Kerbow 47954ab83eSAustin Kerbow; ASM-LABEL: amdhsa_kernarg_preload_1_implicit_2: 48954ab83eSAustin Kerbow; ASM: .amdhsa_user_sgpr_count 3 49954ab83eSAustin Kerbow; ASM: .amdhsa_next_free_sgpr 3 50c897c13dSJanek van Oirschot; ASM: ; TotalNumSgprs: 9 51954ab83eSAustin Kerbow; ASM: ; NumSGPRsForWavesPerEU: 9 52954ab83eSAustin Kerbow 53954ab83eSAustin Kerbow; 1 preload, 2 implicit, 6 extra. Rounds up to 16 SGPRs in the KD. 54954ab83eSAustin Kerbow 55954ab83eSAustin Kerbowdefine amdgpu_kernel void @amdhsa_kernarg_preload_1_implicit_2(i32 inreg) #0 { ret void } 56954ab83eSAustin Kerbow 57954ab83eSAustin Kerbow; OBJDUMP-NEXT: 00c0 00000000 00000000 08010000 00000000 ................ 58954ab83eSAustin Kerbow; OBJDUMP-NEXT: 00d0 00000000 00000000 00000000 00000000 ................ 59954ab83eSAustin Kerbow; OBJDUMP-NEXT: 00e0 00000000 00000000 00000000 00000000 ................ 60954ab83eSAustin Kerbow; OBJDUMP-NEXT: 00f0 0000af00 84000000 08000000 00000000 ................ 61954ab83eSAustin Kerbow 62954ab83eSAustin Kerbow; ASM-LABEL: amdhsa_kernarg_preload_0_implicit_2: 63954ab83eSAustin Kerbow; ASM: .amdhsa_user_sgpr_count 2 64954ab83eSAustin Kerbow; ASM: .amdhsa_next_free_sgpr 0 65c897c13dSJanek van Oirschot; ASM: ; TotalNumSgprs: 6 66954ab83eSAustin Kerbow; ASM: ; NumSGPRsForWavesPerEU: 6 67954ab83eSAustin Kerbow 68954ab83eSAustin Kerbow; 0 preload kernarg SGPRs, 2 implicit, 6 extra. Rounds up to 8 SGPRs in the KD. 69954ab83eSAustin Kerbow; Encoded like '00'. 70954ab83eSAustin Kerbow 71954ab83eSAustin Kerbowdefine amdgpu_kernel void @amdhsa_kernarg_preload_0_implicit_2(i32) #0 { ret void } 72954ab83eSAustin Kerbow 73954ab83eSAustin Kerbowattributes #0 = { "amdgpu-no-agpr" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" } 74