1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00

R600/SI: Add a MUBUF store pattern for Reg+Imm offsets

llvm-svn: 200935
This commit is contained in:
Tom Stellard 2014-02-06 18:36:41 +00:00
parent c690406420
commit 1906c48d55
3 changed files with 23 additions and 1 deletions

View File

@ -103,6 +103,11 @@ def IMM12bit : PatLeaf <(imm),
[{return isUInt<12>(N->getZExtValue());}]
>;
def mubuf_vaddr_offset : PatFrag<
(ops node:$ptr, node:$offset, node:$imm_offset),
(add (add node:$ptr, node:$offset), node:$imm_offset)
>;
class InlineImm <ValueType vt> : PatLeaf <(vt imm), [{
return
(*(const SITargetLowering *)getTargetLowering()).analyzeImmediate(N) == 0;

View File

@ -1960,7 +1960,7 @@ defm : SMRD_Pattern <S_LOAD_DWORDX16_IMM, S_LOAD_DWORDX16_SGPR, v16i32>;
multiclass MUBUFLoad_Pattern <MUBUF Instr_ADDR64, ValueType vt,
PatFrag global_ld, PatFrag constant_ld> {
def : Pat <
(vt (global_ld (add (add i64:$ptr, i64:$offset), IMM12bit:$imm_offset))),
(vt (global_ld (mubuf_vaddr_offset i64:$ptr, i64:$offset, IMM12bit:$imm_offset))),
(Instr_ADDR64 (SI_ADDR64_RSRC $ptr), $offset, (as_i16imm $imm_offset))
>;
@ -2006,6 +2006,11 @@ defm : MUBUFLoad_Pattern <BUFFER_LOAD_DWORDX4_ADDR64, v4i32,
multiclass MUBUFStore_Pattern <MUBUF Instr, ValueType vt, PatFrag st> {
def : Pat <
(st vt:$value, (mubuf_vaddr_offset i64:$ptr, i64:$offset, IMM12bit:$imm_offset)),
(Instr $value, (SI_ADDR64_RSRC $ptr), $offset, (as_i16imm $imm_offset))
>;
def : Pat <
(st vt:$value, (add i64:$ptr, IMM12bit:$offset)),
(Instr $value, (SI_ADDR64_RSRC (i64 0)), $ptr, (as_i16imm $offset))

View File

@ -84,3 +84,15 @@ entry:
store i32 0, i32 addrspace(1)* %0
ret void
}
; MUBUF store with a 12-bit immediate offset and a register offset
; CHECK-LABEL: @mubuf_store3
; CHECK-NOT: ADD
; CHECK: BUFFER_STORE_DWORD v{{[0-9]}}, s[{{[0-9]:[0-9]}}] + v[{{[0-9]:[0-9]}}] + 4 ; encoding: [0x04,0x80
define void @mubuf_store3(i32 addrspace(1)* %out, i64 %offset) {
entry:
%0 = getelementptr i32 addrspace(1)* %out, i64 %offset
%1 = getelementptr i32 addrspace(1)* %0, i64 1
store i32 0, i32 addrspace(1)* %1
ret void
}