42 lines · plain
1; RUN: llc -mtriple=amdgcn < %s | FileCheck -enable-var-scope --check-prefix=GCN %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -enable-var-scope --check-prefix=GCN %s3 4 5; GCN-LABEL: {{^}}global_truncstore_i32_to_i1:6; GCN: s_load_dword [[LOAD:s[0-9]+]],7; GCN: s_and_b32 [[SREG:s[0-9]+]], [[LOAD]], 18; GCN: v_mov_b32_e32 [[VREG:v[0-9]+]], [[SREG]]9; GCN: buffer_store_byte [[VREG]],10define amdgpu_kernel void @global_truncstore_i32_to_i1(ptr addrspace(1) %out, i32 %val) nounwind {11 %trunc = trunc i32 %val to i112 store i1 %trunc, ptr addrspace(1) %out, align 113 ret void14}15 16; GCN-LABEL: {{^}}global_truncstore_i64_to_i1:17; GCN: buffer_store_byte18define amdgpu_kernel void @global_truncstore_i64_to_i1(ptr addrspace(1) %out, i64 %val) nounwind {19 %trunc = trunc i64 %val to i120 store i1 %trunc, ptr addrspace(1) %out, align 121 ret void22}23 24; FIXME: VGPR on VI25; GCN-LABEL: {{^}}s_arg_global_truncstore_i16_to_i1:26; GCN: s_load_dword [[LOAD:s[0-9]+]],27; GCN: s_and_b32 [[SREG:s[0-9]+]], [[LOAD]], 128; GCN: v_mov_b32_e32 [[VREG:v[0-9]+]], [[SREG]]29; GCN: buffer_store_byte [[VREG]],30define amdgpu_kernel void @s_arg_global_truncstore_i16_to_i1(ptr addrspace(1) %out, i16 %val) nounwind {31 %trunc = trunc i16 %val to i132 store i1 %trunc, ptr addrspace(1) %out, align 133 ret void34}35; GCN-LABEL: {{^}}global_truncstore_i16_to_i1:36define amdgpu_kernel void @global_truncstore_i16_to_i1(ptr addrspace(1) %out, i16 %val0, i16 %val1) nounwind {37 %add = add i16 %val0, %val138 %trunc = trunc i16 %add to i139 store i1 %trunc, ptr addrspace(1) %out, align 140 ret void41}42