Page MenuHomePhabricator
Differential D96340 Diff 323224 llvm/test/CodeGen/AMDGPU/llvm.amdgcn.implicitarg.ptr.ll

Changeset View

Standalone View

View Options

llvm/test/CodeGen/AMDGPU/llvm.amdgcn.implicitarg.ptr.ll

Show First 20 Lines • Show All 53 Lines • ▼ Show 20 Linesdefine amdgpu_kernel void @opencl_kernel_implicitarg_ptr([112 x i8]) #1 {
%implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr() %implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr()
%cast = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)* %cast = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)*
%load = load volatile i32, i32 addrspace(4)* %cast %load = load volatile i32, i32 addrspace(4)* %cast
ret void ret void
} }
; GCN-LABEL: {{^}}func_implicitarg_ptr: ; GCN-LABEL: {{^}}func_implicitarg_ptr:
; GCN: s_waitcnt ; GCN: s_waitcnt
; GCN: s_load_dword s{{[0-9]+}}, s[4:5], 0x0 ; HSA: s_load_dword s{{[0-9]+}}, s[8:9], 0x0
; MESA: s_load_dword s{{[0-9]+}}, s[4:5], 0x0
; GCN-NEXT: s_waitcnt ; GCN-NEXT: s_waitcnt
; GCN-NEXT: s_setpc_b64 ; GCN-NEXT: s_setpc_b64
define void @func_implicitarg_ptr() #0 { define void @func_implicitarg_ptr() #0 {
%implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr() %implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr()
%cast = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)* %cast = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)*
%load = load volatile i32, i32 addrspace(4)* %cast %load = load volatile i32, i32 addrspace(4)* %cast
ret void ret void
} }
; GCN-LABEL: {{^}}opencl_func_implicitarg_ptr: ; GCN-LABEL: {{^}}opencl_func_implicitarg_ptr:
; GCN: s_waitcnt ; GCN: s_waitcnt
; GCN: s_load_dword s{{[0-9]+}}, s[4:5], 0x0 ; HSA: s_load_dword s{{[0-9]+}}, s[8:9], 0x0
; MESA: s_load_dword s{{[0-9]+}}, s[4:5], 0x0
; GCN-NEXT: s_waitcnt ; GCN-NEXT: s_waitcnt
; GCN-NEXT: s_setpc_b64 ; GCN-NEXT: s_setpc_b64
define void @opencl_func_implicitarg_ptr() #0 { define void @opencl_func_implicitarg_ptr() #0 {
%implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr() %implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr()
%cast = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)* %cast = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)*
%load = load volatile i32, i32 addrspace(4)* %cast %load = load volatile i32, i32 addrspace(4)* %cast
ret void ret void
} }
Show All 24 Linesdefine amdgpu_kernel void @opencl_kernel_call_implicitarg_ptr_func_empty() #1 {
ret void ret void
} }
; GCN-LABEL: {{^}}kernel_call_implicitarg_ptr_func: ; GCN-LABEL: {{^}}kernel_call_implicitarg_ptr_func:
; GCN: enable_sgpr_kernarg_segment_ptr = 1 ; GCN: enable_sgpr_kernarg_segment_ptr = 1
; HSA: kernarg_segment_byte_size = 112 ; HSA: kernarg_segment_byte_size = 112
; MESA: kernarg_segment_byte_size = 128 ; MESA: kernarg_segment_byte_size = 128
; HSA: s_add_u32 s4, s4, 0x70 ; HSA: s_add_u32 s8, s8, 0x70
; MESA: s_add_u32 s4, s4, 0x70 ; MESA: s_add_u32 s4, s4, 0x70
; GCN: s_addc_u32 s5, s5, 0{{$}} ; HSA: s_addc_u32 s9, s9, 0{{$}}
; MESA: s_addc_u32 s5, s5, 0{{$}}
; GCN: s_swappc_b64 ; GCN: s_swappc_b64
define amdgpu_kernel void @kernel_call_implicitarg_ptr_func([112 x i8]) #0 { define amdgpu_kernel void @kernel_call_implicitarg_ptr_func([112 x i8]) #0 {
call void @func_implicitarg_ptr() call void @func_implicitarg_ptr()
ret void ret void
} }
; GCN-LABEL: {{^}}opencl_kernel_call_implicitarg_ptr_func: ; GCN-LABEL: {{^}}opencl_kernel_call_implicitarg_ptr_func:
; GCN: enable_sgpr_kernarg_segment_ptr = 1 ; GCN: enable_sgpr_kernarg_segment_ptr = 1
; HSA: kernarg_segment_byte_size = 160 ; HSA: kernarg_segment_byte_size = 160
; MESA: kernarg_segment_byte_size = 128 ; MESA: kernarg_segment_byte_size = 128
; GCN: s_add_u32 s4, s4, 0x70 ; HSA: s_add_u32 s8, s8, 0x70
; GCN: s_addc_u32 s5, s5, 0{{$}} ; HSA: s_addc_u32 s9, s9, 0{{$}}
; MESA: s_add_u32 s4, s4, 0x70
; MESA: s_addc_u32 s5, s5, 0{{$}}
; GCN: s_swappc_b64 ; GCN: s_swappc_b64
define amdgpu_kernel void @opencl_kernel_call_implicitarg_ptr_func([112 x i8]) #1 { define amdgpu_kernel void @opencl_kernel_call_implicitarg_ptr_func([112 x i8]) #1 {
call void @func_implicitarg_ptr() call void @func_implicitarg_ptr()
ret void ret void
} }
; GCN-LABEL: {{^}}func_call_implicitarg_ptr_func: ; GCN-LABEL: {{^}}func_call_implicitarg_ptr_func:
; GCN-NOT: s4 ; HSA-NOT: s8
; GCN-NOT: s5 ; HSA-NOT: s9
; GCN-NOT: s[4:5] ; HSA-NOT: s[8:9]
; MESA-NOT: s4
; MESA-NOT: s5
; MESA-NOT: s[4:5]
define void @func_call_implicitarg_ptr_func() #0 { define void @func_call_implicitarg_ptr_func() #0 {
call void @func_implicitarg_ptr() call void @func_implicitarg_ptr()
ret void ret void
} }
; GCN-LABEL: {{^}}opencl_func_call_implicitarg_ptr_func: ; GCN-LABEL: {{^}}opencl_func_call_implicitarg_ptr_func:
; GCN-NOT: s4 ; HSA-NOT: s8
; GCN-NOT: s5 ; HSA-NOT: s9
; GCN-NOT: s[4:5] ; HSA-NOT: s[8:9]
; MESA-NOT: s4
; MESA-NOT: s5
; MESA-NOT: s[4:5]
define void @opencl_func_call_implicitarg_ptr_func() #0 { define void @opencl_func_call_implicitarg_ptr_func() #0 {
call void @func_implicitarg_ptr() call void @func_implicitarg_ptr()
ret void ret void
} }
; GCN-LABEL: {{^}}func_kernarg_implicitarg_ptr: ; GCN-LABEL: {{^}}func_kernarg_implicitarg_ptr:
; GCN: s_waitcnt ; GCN: s_waitcnt
; GCN-DAG: s_mov_b64 [[NULL:s\[[0-9]+:[0-9]+\]]], 0 ; GCN-DAG: s_mov_b64 [[NULL:s\[[0-9]+:[0-9]+\]]], 0
; GCN-DAG: s_load_dword s{{[0-9]+}}, [[NULL]], 0x0 ; GCN-DAG: s_load_dword s{{[0-9]+}}, [[NULL]], 0x0
; GCN: s_load_dword s{{[0-9]+}}, s[4:5], 0x0 ; HSA: s_load_dword s{{[0-9]+}}, s[8:9], 0x0
; MESA: s_load_dword s{{[0-9]+}}, s[4:5], 0x0
; GCN: s_waitcnt lgkmcnt(0) ; GCN: s_waitcnt lgkmcnt(0)
define void @func_kernarg_implicitarg_ptr() #0 { define void @func_kernarg_implicitarg_ptr() #0 {
%kernarg.segment.ptr = call i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr() %kernarg.segment.ptr = call i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
%implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr() %implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr()
%cast.kernarg.segment.ptr = bitcast i8 addrspace(4)* %kernarg.segment.ptr to i32 addrspace(4)* %cast.kernarg.segment.ptr = bitcast i8 addrspace(4)* %kernarg.segment.ptr to i32 addrspace(4)*
%cast.implicitarg = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)* %cast.implicitarg = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)*
%load0 = load volatile i32, i32 addrspace(4)* %cast.kernarg.segment.ptr %load0 = load volatile i32, i32 addrspace(4)* %cast.kernarg.segment.ptr
%load1 = load volatile i32, i32 addrspace(4)* %cast.implicitarg %load1 = load volatile i32, i32 addrspace(4)* %cast.implicitarg
ret void ret void
} }
; GCN-LABEL: {{^}}opencl_func_kernarg_implicitarg_ptr: ; GCN-LABEL: {{^}}opencl_func_kernarg_implicitarg_ptr:
; GCN: s_waitcnt ; GCN: s_waitcnt
; GCN-DAG: s_mov_b64 [[NULL:s\[[0-9]+:[0-9]+\]]], 0 ; GCN-DAG: s_mov_b64 [[NULL:s\[[0-9]+:[0-9]+\]]], 0
; GCN-DAG: s_load_dword s{{[0-9]+}}, [[NULL]], 0x0 ; GCN-DAG: s_load_dword s{{[0-9]+}}, [[NULL]], 0x0
; GCN: s_load_dword s{{[0-9]+}}, s[4:5], 0x0 ; HSA: s_load_dword s{{[0-9]+}}, s[8:9], 0x0
; MESA: s_load_dword s{{[0-9]+}}, s[4:5], 0x0
; GCN: s_waitcnt lgkmcnt(0) ; GCN: s_waitcnt lgkmcnt(0)
define void @opencl_func_kernarg_implicitarg_ptr() #0 { define void @opencl_func_kernarg_implicitarg_ptr() #0 {
%kernarg.segment.ptr = call i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr() %kernarg.segment.ptr = call i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
%implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr() %implicitarg.ptr = call i8 addrspace(4)* @llvm.amdgcn.implicitarg.ptr()
%cast.kernarg.segment.ptr = bitcast i8 addrspace(4)* %kernarg.segment.ptr to i32 addrspace(4)* %cast.kernarg.segment.ptr = bitcast i8 addrspace(4)* %kernarg.segment.ptr to i32 addrspace(4)*
%cast.implicitarg = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)* %cast.implicitarg = bitcast i8 addrspace(4)* %implicitarg.ptr to i32 addrspace(4)*
%load0 = load volatile i32, i32 addrspace(4)* %cast.kernarg.segment.ptr %load0 = load volatile i32, i32 addrspace(4)* %cast.kernarg.segment.ptr
%load1 = load volatile i32, i32 addrspace(4)* %cast.implicitarg %load1 = load volatile i32, i32 addrspace(4)* %cast.implicitarg
ret void ret void
} }
; GCN-LABEL: {{^}}kernel_call_kernarg_implicitarg_ptr_func: ; GCN-LABEL: {{^}}kernel_call_kernarg_implicitarg_ptr_func:
; GCN: s_add_u32 s4, s4, 0x70 ; HSA: s_add_u32 s8, s8, 0x70
; GCN: s_addc_u32 s5, s5, 0 ; HSA: s_addc_u32 s9, s9, 0
; MESA: s_add_u32 s4, s4, 0x70
; MESA: s_addc_u32 s5, s5, 0
; GCN: s_swappc_b64 ; GCN: s_swappc_b64
define amdgpu_kernel void @kernel_call_kernarg_implicitarg_ptr_func([112 x i8]) #0 { define amdgpu_kernel void @kernel_call_kernarg_implicitarg_ptr_func([112 x i8]) #0 {
call void @func_kernarg_implicitarg_ptr() call void @func_kernarg_implicitarg_ptr()
ret void ret void
} }
; GCN-LABEL: {{^}}kernel_implicitarg_no_struct_align_padding: ; GCN-LABEL: {{^}}kernel_implicitarg_no_struct_align_padding:
; HSA: kernarg_segment_byte_size = 120 ; HSA: kernarg_segment_byte_size = 120
Show All 15 Lines