6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-14 11:32:34 +00:00
Code Issues Projects Releases Wiki Activity

R600/SI: Add other LDS atomic operations

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210676 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Matt Arsenault 2014-06-11 18:08:42 +00:00
parent 507d8cad7b
commit 2da1a85cbb
2 changed files with 220 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
lib/Target/R600/SIInstructions.td
View File

@ -2210,7 +2210,7 @@ defm : DSWritePat <DS_WRITE_B16, i32, truncstorei16_local>;
defm : DSWritePat <DS_WRITE_B32, i32, local_store>; defm : DSWritePat <DS_WRITE_B32, i32, local_store>;
defm : DSWritePat <DS_WRITE_B64, i64, local_store>; defm : DSWritePat <DS_WRITE_B64, i64, local_store>;
multiclass DSAtomicPat<DS inst, ValueType vt, PatFrag frag> { multiclass DSAtomicRetPat<DS inst, ValueType vt, PatFrag frag> {
def : Pat < def : Pat <
(frag (add i32:$ptr, (i32 IMM16bit:$offset)), vt:$value), (frag (add i32:$ptr, (i32 IMM16bit:$offset)), vt:$value),
(inst (i1 0), $ptr, $value, (as_i16imm $offset)) (inst (i1 0), $ptr, $value, (as_i16imm $offset))
@ -2222,8 +2222,17 @@ multiclass DSAtomicPat<DS inst, ValueType vt, PatFrag frag> {
>; >;
} }
defm : DSAtomicPat<DS_ADD_RTN_U32, i32, atomic_load_add_local>; defm : DSAtomicRetPat<DS_WRXCHG_RTN_B32, i32, atomic_swap_local>;
defm : DSAtomicPat<DS_SUB_RTN_U32, i32, atomic_load_sub_local>; defm : DSAtomicRetPat<DS_ADD_RTN_U32, i32, atomic_load_add_local>;
defm : DSAtomicRetPat<DS_SUB_RTN_U32, i32, atomic_load_sub_local>;
defm : DSAtomicRetPat<DS_AND_RTN_B32, i32, atomic_load_and_local>;
defm : DSAtomicRetPat<DS_OR_RTN_B32, i32, atomic_load_or_local>;
defm : DSAtomicRetPat<DS_XOR_RTN_B32, i32, atomic_load_xor_local>;
defm : DSAtomicRetPat<DS_MIN_RTN_I32, i32, atomic_load_min_local>;
defm : DSAtomicRetPat<DS_MAX_RTN_I32, i32, atomic_load_max_local>;
defm : DSAtomicRetPat<DS_MIN_RTN_U32, i32, atomic_load_umin_local>;
defm : DSAtomicRetPat<DS_MAX_RTN_U32, i32, atomic_load_umax_local>;
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
// MUBUF Patterns // MUBUF Patterns

208
test/CodeGen/R600/local-atomics.ll Normal file
View File

@ -0,0 +1,208 @@
; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
; FUNC-LABEL: @lds_atomic_xchg_ret_i32:
; SI: S_LOAD_DWORD [[SPTR:s[0-9]+]],
; SI: V_MOV_B32_e32 [[DATA:v[0-9]+]], 4
; SI: V_MOV_B32_e32 [[VPTR:v[0-9]+]], [[SPTR]]
; SI: DS_WRXCHG_RTN_B32 [[RESULT:v[0-9]+]], [[VPTR]], [[DATA]], 0x0, [M0]
; SI: BUFFER_STORE_DWORD [[RESULT]],
; SI: S_ENDPGM
define void @lds_atomic_xchg_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw xchg i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_xchg_ret_i32_offset:
; SI: DS_WRXCHG_RTN_B32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_xchg_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw xchg i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; XXX - Is it really necessary to load 4 into VGPR?
; FUNC-LABEL: @lds_atomic_add_ret_i32:
; SI: S_LOAD_DWORD [[SPTR:s[0-9]+]],
; SI: V_MOV_B32_e32 [[DATA:v[0-9]+]], 4
; SI: V_MOV_B32_e32 [[VPTR:v[0-9]+]], [[SPTR]]
; SI: DS_ADD_RTN_U32 [[RESULT:v[0-9]+]], [[VPTR]], [[DATA]], 0x0, [M0]
; SI: BUFFER_STORE_DWORD [[RESULT]],
; SI: S_ENDPGM
define void @lds_atomic_add_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw add i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_add_ret_i32_offset:
; SI: DS_ADD_RTN_U32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_add_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw add i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_sub_ret_i32:
; SI: DS_SUB_RTN_U32
; SI: S_ENDPGM
define void @lds_atomic_sub_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw sub i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_sub_ret_i32_offset:
; SI: DS_SUB_RTN_U32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_sub_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw sub i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_and_ret_i32:
; SI: DS_AND_RTN_B32
; SI: S_ENDPGM
define void @lds_atomic_and_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw and i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_and_ret_i32_offset:
; SI: DS_AND_RTN_B32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_and_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw and i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_or_ret_i32:
; SI: DS_OR_RTN_B32
; SI: S_ENDPGM
define void @lds_atomic_or_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw or i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_or_ret_i32_offset:
; SI: DS_OR_RTN_B32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_or_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw or i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_xor_ret_i32:
; SI: DS_XOR_RTN_B32
; SI: S_ENDPGM
define void @lds_atomic_xor_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw xor i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_xor_ret_i32_offset:
; SI: DS_XOR_RTN_B32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_xor_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw xor i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FIXME: There is no atomic nand instr
; XFUNC-LABEL: @lds_atomic_nand_ret_i32:uction, so we somehow need to expand this.
; define void @lds_atomic_nand_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
; %result = atomicrmw nand i32 addrspace(3)* %ptr, i32 4 seq_cst
; store i32 %result, i32 addrspace(1)* %out, align 4
; ret void
; }
; FUNC-LABEL: @lds_atomic_min_ret_i32:
; SI: DS_MIN_RTN_I32
; SI: S_ENDPGM
define void @lds_atomic_min_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw min i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_min_ret_i32_offset:
; SI: DS_MIN_RTN_I32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_min_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw min i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_max_ret_i32:
; SI: DS_MAX_RTN_I32
; SI: S_ENDPGM
define void @lds_atomic_max_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw max i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_max_ret_i32_offset:
; SI: DS_MAX_RTN_I32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_max_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw max i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_umin_ret_i32:
; SI: DS_MIN_RTN_U32
; SI: S_ENDPGM
define void @lds_atomic_umin_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw umin i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_umin_ret_i32_offset:
; SI: DS_MIN_RTN_U32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_umin_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw umin i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_umax_ret_i32:
; SI: DS_MAX_RTN_U32
; SI: S_ENDPGM
define void @lds_atomic_umax_ret_i32(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%result = atomicrmw umax i32 addrspace(3)* %ptr, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: @lds_atomic_umax_ret_i32_offset:
; SI: DS_MAX_RTN_U32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 0x10
; SI: S_ENDPGM
define void @lds_atomic_umax_ret_i32_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %ptr) nounwind {
%gep = getelementptr i32 addrspace(3)* %ptr, i32 4
%result = atomicrmw umax i32 addrspace(3)* %gep, i32 4 seq_cst
store i32 %result, i32 addrspace(1)* %out, align 4
ret void
}