77 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize64 < %s | FileCheck -check-prefix=GFX12 %s3; RUN: llc -global-isel=1 -global-isel-abort=2 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+wavefrontsize64 < %s | FileCheck -check-prefix=GFX12 %s4 5declare i32 @llvm.amdgcn.global.load.tr.b64.i32.p1(ptr addrspace(1))6declare <4 x i16> @llvm.amdgcn.global.load.tr.b128.v4i16.p1(ptr addrspace(1))7declare <4 x half> @llvm.amdgcn.global.load.tr.b128.v4f16.p1(ptr addrspace(1))8declare <4 x bfloat> @llvm.amdgcn.global.load.tr.b128.v4bf16.p1(ptr addrspace(1))9 10define amdgpu_kernel void @global_load_tr_b64_i32(ptr addrspace(1) %addr, ptr addrspace(1) %use) {11; GFX12-LABEL: global_load_tr_b64_i32:12; GFX12: ; %bb.0: ; %entry13; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x2414; GFX12-NEXT: v_mov_b32_e32 v0, 015; GFX12-NEXT: s_wait_kmcnt 0x016; GFX12-NEXT: global_load_tr_b64 v1, v0, s[0:1] offset:3217; GFX12-NEXT: s_wait_loadcnt 0x018; GFX12-NEXT: global_store_b32 v0, v1, s[2:3]19; GFX12-NEXT: s_endpgm20entry:21 %gep = getelementptr i64, ptr addrspace(1) %addr, i32 422 %val = call i32 @llvm.amdgcn.global.load.tr.b64.i32.p1(ptr addrspace(1) %gep)23 store i32 %val, ptr addrspace(1) %use24 ret void25}26 27define amdgpu_kernel void @global_load_tr_b128_v4i16(ptr addrspace(1) %addr, ptr addrspace(1) %use) {28; GFX12-LABEL: global_load_tr_b128_v4i16:29; GFX12: ; %bb.0: ; %entry30; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x2431; GFX12-NEXT: v_mov_b32_e32 v2, 032; GFX12-NEXT: s_wait_kmcnt 0x033; GFX12-NEXT: global_load_tr_b128 v[0:1], v2, s[0:1] offset:3234; GFX12-NEXT: s_wait_loadcnt 0x035; GFX12-NEXT: global_store_b64 v2, v[0:1], s[2:3]36; GFX12-NEXT: s_endpgm37entry:38 %gep = getelementptr i64, ptr addrspace(1) %addr, i32 439 %val = call <4 x i16> @llvm.amdgcn.global.load.tr.b128.v4i16.p1(ptr addrspace(1) %gep)40 store <4 x i16> %val, ptr addrspace(1) %use41 ret void42}43 44define amdgpu_kernel void @global_load_tr_b128_v4f16(ptr addrspace(1) %addr, ptr addrspace(1) %use) {45; GFX12-LABEL: global_load_tr_b128_v4f16:46; GFX12: ; %bb.0: ; %entry47; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x2448; GFX12-NEXT: v_mov_b32_e32 v2, 049; GFX12-NEXT: s_wait_kmcnt 0x050; GFX12-NEXT: global_load_tr_b128 v[0:1], v2, s[0:1] offset:3251; GFX12-NEXT: s_wait_loadcnt 0x052; GFX12-NEXT: global_store_b64 v2, v[0:1], s[2:3]53; GFX12-NEXT: s_endpgm54entry:55 %gep = getelementptr i64, ptr addrspace(1) %addr, i32 456 %val = call <4 x half> @llvm.amdgcn.global.load.tr.b128.v4f16.p1(ptr addrspace(1) %gep)57 store <4 x half> %val, ptr addrspace(1) %use58 ret void59}60 61define amdgpu_kernel void @global_load_tr_b128_v4bf16(ptr addrspace(1) %addr, ptr addrspace(1) %use) {62; GFX12-LABEL: global_load_tr_b128_v4bf16:63; GFX12: ; %bb.0: ; %entry64; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x2465; GFX12-NEXT: v_mov_b32_e32 v2, 066; GFX12-NEXT: s_wait_kmcnt 0x067; GFX12-NEXT: global_load_tr_b128 v[0:1], v2, s[0:1] offset:3268; GFX12-NEXT: s_wait_loadcnt 0x069; GFX12-NEXT: global_store_b64 v2, v[0:1], s[2:3]70; GFX12-NEXT: s_endpgm71entry:72 %gep = getelementptr i64, ptr addrspace(1) %addr, i32 473 %val = call <4 x bfloat> @llvm.amdgcn.global.load.tr.b128.v4bf16.p1(ptr addrspace(1) %gep)74 store <4 x bfloat> %val, ptr addrspace(1) %use75 ret void76}77