1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 04:32:44 +01:00
llvm-mirror/test/CodeGen/AMDGPU/scratch-buffer.ll
Scott Linder a70016c8d5 [AMDGPU] Add Scratch Wave Offset to Scratch Buffer Descriptor in entry functions
Add the scratch wave offset to the scratch buffer descriptor (SRSrc) in
the entry function prologue. This allows us to removes the scratch wave
offset register from the calling convention ABI.

As part of this change, allow the use of an inline constant zero for the
SOffset of MUBUF instructions accessing the stack in entry functions
when a frame pointer is not requested/required. Entry functions with
calls still need to set up the calling convention ABI stack pointer
register, and reference it in order to address arguments of called
functions. The ABI stack pointer register remains unswizzled, but is now
wave-relative instead of queue-relative.

Non-entry functions also use an inline constant zero SOffset for
wave-relative scratch access, but continue to use the stack and frame
pointers as before. When the stack or frame pointer is converted to a
swizzled offset it is now scaled directly, as the scratch wave offset no
longer needs to be subtracted first.

Update llvm/docs/AMDGPUUsage.rst to reflect these changes to the calling
convention.

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D75138
2020-03-19 15:35:16 -04:00

125 lines
5.1 KiB
LLVM

; RUN: llc -amdgpu-scalarize-global-loads=false -verify-machineinstrs -march=amdgcn < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
; RUN: llc -amdgpu-scalarize-global-loads=false -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
; When a frame index offset is more than 12-bits, make sure we don't store
; it in mubuf's offset field.
; Also, make sure we use the same register for storing the scratch buffer addresss
; for both stores. This register is allocated by the register scavenger, so we
; should be able to reuse the same regiser for each scratch buffer access.
; GCN-LABEL: {{^}}legal_offset_fi:
; GCN: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+}}:{{[0-9]+}}], 0 offset:4{{$}}
; GCN: v_mov_b32_e32 [[OFFSET:v[0-9]+]], 0x8004
; GCN: buffer_store_dword v{{[0-9]+}}, [[OFFSET]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offen{{$}}
define amdgpu_kernel void @legal_offset_fi(i32 addrspace(1)* %out, i32 %cond, i32 %if_offset, i32 %else_offset) {
entry:
%scratch0 = alloca [8192 x i32], addrspace(5)
%scratch1 = alloca [8192 x i32], addrspace(5)
%scratchptr0 = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %scratch0, i32 0, i32 0
store i32 1, i32 addrspace(5)* %scratchptr0
%scratchptr1 = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %scratch1, i32 0, i32 0
store i32 2, i32 addrspace(5)* %scratchptr1
%cmp = icmp eq i32 %cond, 0
br i1 %cmp, label %if, label %else
if:
%if_ptr = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %scratch0, i32 0, i32 %if_offset
%if_value = load i32, i32 addrspace(5)* %if_ptr
br label %done
else:
%else_ptr = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %scratch1, i32 0, i32 %else_offset
%else_value = load i32, i32 addrspace(5)* %else_ptr
br label %done
done:
%value = phi i32 [%if_value, %if], [%else_value, %else]
store i32 %value, i32 addrspace(1)* %out
ret void
ret void
}
; GCN-LABEL: {{^}}legal_offset_fi_offset:
; GCN-DAG: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 0 offen{{$}}
; This constant isn't folded, because it has multiple uses.
; GCN-DAG: v_mov_b32_e32 [[K8000:v[0-9]+]], 0x8004
; GCN-DAG: v_add_{{[iu]}}32_e32 [[OFFSET:v[0-9]+]], vcc, [[K8000]]
; GCN: buffer_store_dword v{{[0-9]+}}, [[OFFSET]], s[{{[0-9]+}}:{{[0-9]+}}], 0 offen{{$}}
define amdgpu_kernel void @legal_offset_fi_offset(i32 addrspace(1)* %out, i32 %cond, i32 addrspace(1)* %offsets, i32 %if_offset, i32 %else_offset) {
entry:
%scratch0 = alloca [8192 x i32], addrspace(5)
%scratch1 = alloca [8192 x i32], addrspace(5)
%offset0 = load i32, i32 addrspace(1)* %offsets
%scratchptr0 = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %scratch0, i32 0, i32 %offset0
store i32 %offset0, i32 addrspace(5)* %scratchptr0
%offsetptr1 = getelementptr i32, i32 addrspace(1)* %offsets, i32 1
%offset1 = load i32, i32 addrspace(1)* %offsetptr1
%scratchptr1 = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %scratch1, i32 0, i32 %offset1
store i32 %offset1, i32 addrspace(5)* %scratchptr1
%cmp = icmp eq i32 %cond, 0
br i1 %cmp, label %if, label %else
if:
%if_ptr = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %scratch0, i32 0, i32 %if_offset
%if_value = load i32, i32 addrspace(5)* %if_ptr
br label %done
else:
%else_ptr = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %scratch1, i32 0, i32 %else_offset
%else_value = load i32, i32 addrspace(5)* %else_ptr
br label %done
done:
%value = phi i32 [%if_value, %if], [%else_value, %else]
store i32 %value, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}neg_vaddr_offset_inbounds:
; GCN: v_add_{{[iu]}}32_e32 [[ADD:v[0-9]+]], vcc, 16, v{{[0-9]+}}
; GCN: buffer_store_dword v{{[0-9]+}}, [[ADD]], s[{{[0-9]+:[0-9]+}}], 0 offen{{$}}
define amdgpu_kernel void @neg_vaddr_offset_inbounds(i32 %offset) {
entry:
%array = alloca [8192 x i32], addrspace(5)
%ptr_offset = add i32 %offset, 4
%ptr = getelementptr inbounds [8192 x i32], [8192 x i32] addrspace(5)* %array, i32 0, i32 %ptr_offset
store i32 0, i32 addrspace(5)* %ptr
ret void
}
; GCN-LABEL: {{^}}neg_vaddr_offset:
; GCN: v_add_{{[iu]}}32_e32 [[ADD:v[0-9]+]], vcc, 16, v{{[0-9]+}}
; GCN: buffer_store_dword v{{[0-9]+}}, [[ADD]], s[{{[0-9]+:[0-9]+}}], 0 offen{{$}}
define amdgpu_kernel void @neg_vaddr_offset(i32 %offset) {
entry:
%array = alloca [8192 x i32], addrspace(5)
%ptr_offset = add i32 %offset, 4
%ptr = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %array, i32 0, i32 %ptr_offset
store i32 0, i32 addrspace(5)* %ptr
ret void
}
; GCN-LABEL: {{^}}pos_vaddr_offset:
; GCN: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], 0 offset:20
define amdgpu_kernel void @pos_vaddr_offset(i32 addrspace(1)* %out, i32 %offset) {
entry:
%array = alloca [8192 x i32], addrspace(5)
%ptr = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %array, i32 0, i32 4
store i32 0, i32 addrspace(5)* %ptr
%load_ptr = getelementptr [8192 x i32], [8192 x i32] addrspace(5)* %array, i32 0, i32 %offset
%val = load i32, i32 addrspace(5)* %load_ptr
store i32 %val, i32 addrspace(1)* %out
ret void
}