91 lines · plain
1; RUN: llc -mtriple=amdgcn -mcpu=gfx90a < %s | FileCheck -check-prefixes=GFX90A,SDAG %s2; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -early-live-intervals < %s | FileCheck -check-prefixes=GFX90A,SDAG %s3; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx90a < %s | FileCheck -check-prefixes=GFX90A,GISEL %s4 5; GFX90A-LABEL: {{^}}sample_1d:6; GFX90A-NOT: s_wqm_b647; GFX90A: image_sample v[{{[0-9:]+}}], v{{[0-9]+}}, s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf8define amdgpu_ps <4 x float> @sample_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {9main_body:10 %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)11 ret <4 x float> %v12}13 14; GFX90A-LABEL: {{^}}sample_1d_lwe:15; GFX90A-NOT: s_wqm_b6416; GFX90A: image_sample v[{{[0-9:]+}}], v{{[0-9]+}}, s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf lwe17define amdgpu_ps <4 x float> @sample_1d_lwe(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, ptr addrspace(1) inreg %out, float %s) {18main_body:19 %v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 2, i32 0)20 %v.vec = extractvalue {<4 x float>, i32} %v, 021 %v.err = extractvalue {<4 x float>, i32} %v, 122 store i32 %v.err, ptr addrspace(1) %out, align 423 ret <4 x float> %v.vec24}25 26; GFX90A-LABEL: {{^}}sample_2d:27; GFX90A-NOT: s_wqm_b6428; GFX90A: image_sample v[{{[0-9:]+}}], v[{{[0-9:]+}}], s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf29define amdgpu_ps <4 x float> @sample_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t) {30main_body:31 %v = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)32 ret <4 x float> %v33}34 35; GFX90A-LABEL: {{^}}sample_3d:36; GFX90A-NOT: s_wqm_b6437; GFX90A: image_sample v[{{[0-9:]+}}], v[{{[0-9:]+}}], s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf38define amdgpu_ps <4 x float> @sample_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %r) {39main_body:40 %v = call <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f32(i32 15, float %s, float %t, float %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)41 ret <4 x float> %v42}43 44; GFX90A-LABEL: {{^}}sample_cube:45; GFX90A-NOT: s_wqm_b6446; GFX90A: image_sample v[{{[0-9:]+}}], v[{{[0-9:]+}}], s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf da47define amdgpu_ps <4 x float> @sample_cube(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %face) {48main_body:49 %v = call <4 x float> @llvm.amdgcn.image.sample.cube.v4f32.f32(i32 15, float %s, float %t, float %face, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)50 ret <4 x float> %v51}52 53; GFX90A-LABEL: {{^}}sample_1darray:54; GFX90A-NOT: s_wqm_b6455; GFX90A: image_sample v[{{[0-9:]+}}], v[{{[0-9:]+}}], s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf da56define amdgpu_ps <4 x float> @sample_1darray(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %slice) {57main_body:58 %v = call <4 x float> @llvm.amdgcn.image.sample.1darray.v4f32.f32(i32 15, float %s, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)59 ret <4 x float> %v60}61 62; GFX90A-LABEL: {{^}}sample_1d_unorm:63; GFX90A-NOT: s_wqm_b6464; GFX90A: image_sample v[{{[0-9:]+}}], v{{[0-9]+}}, s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf unorm65define amdgpu_ps <4 x float> @sample_1d_unorm(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {66main_body:67 %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 1, i32 0, i32 0)68 ret <4 x float> %v69}70 71; Address register must be even aligned.72 73; GFX90A-LABEL: {{^}}sample_1d_addr_align:74; GFX90A: v_mov_b32_e32 [[VADDR:v[0-9]?[02468]]], v175; SDAG: image_sample v{{[0-9]+}}, [[VADDR]], s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0x176; GISEL: image_sample v[{{[0-9:]+}}], [[VADDR]], s[{{[0-9:]+}}], s[{{[0-9:]+}}] dmask:0xf77define amdgpu_ps float @sample_1d_addr_align(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, <2 x float> %s) {78main_body:79 %s1 = extractelement <2 x float> %s, i32 180 %v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s1, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)81 %v1 = extractelement <4 x float> %v, i32 082 ret float %v183}84 85declare <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32)86declare {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32)87declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)88declare <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)89declare <4 x float> @llvm.amdgcn.image.sample.cube.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)90declare <4 x float> @llvm.amdgcn.image.sample.1darray.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)91