26 lines · plain
1; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a < %s | FileCheck -enable-var-scope -check-prefixes=GFX90A %s2; RUN: not llc -global-isel < %s -mtriple=amdgcn -mcpu=gfx908 2>&1 | FileCheck %s -check-prefix=GFX9083 4declare float @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.f32(float, ptr addrspace(8), i32, i32, i32 immarg)5declare <2 x half> @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.v2f16(<2 x half>, ptr addrspace(8), i32, i32, i32 immarg)6 7; GFX908: LLVM ERROR: cannot select: %{{[0-9]+}}:vgpr_32(s32) = G_AMDGPU_BUFFER_ATOMIC_FADD %{{[0-9]+}}:vgpr, %{{[0-9]+}}:sgpr(<4 x s32>), %{{[0-9]+}}:vgpr(s32), %{{[0-9]+}}:vgpr, %{{[0-9]+}}:sgpr, 0, 0, 0 :: (volatile dereferenceable load store (s32) on %ir.rsrc.load, align 1, addrspace 8) (in function: buffer_atomic_add_f32_rtn)8 9; GFX90A-LABEL: {{^}}buffer_atomic_add_f32_rtn:10; GFX90A: buffer_atomic_add_f32 v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9:]+}}], s{{[0-9]+}} offen glc11define amdgpu_kernel void @buffer_atomic_add_f32_rtn(float %val, ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset) {12main_body:13 %ret = call float @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.f32(float %val, ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 0)14 store float %ret, ptr poison15 ret void16}17 18; GFX90A-LABEL: {{^}}buffer_atomic_add_v2f16_rtn:19; GFX90A: buffer_atomic_pk_add_f16 v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9:]+}}], s{{[0-9]+}} offen glc20define amdgpu_kernel void @buffer_atomic_add_v2f16_rtn(<2 x half> %val, ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset) {21main_body:22 %ret = call <2 x half> @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.v2f16(<2 x half> %val, ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 0)23 store <2 x half> %ret, ptr poison24 ret void25}26