[PATCH] D65077: AMDGPU: Correct FP atomic patterns
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Jul 22 05:47:50 PDT 2019
arsenm created this revision.
arsenm added a reviewer: rampitec.
Herald added subscribers: jfb, t-tye, tpr, dstuttard, yaxunl, nhaehnle, wdng, jvesely, kzhuravl.
https://reviews.llvm.org/D65077
Files:
lib/Target/AMDGPU/BUFInstructions.td
lib/Target/AMDGPU/FLATInstructions.td
lib/Target/AMDGPU/SIInstrInfo.td
Index: lib/Target/AMDGPU/SIInstrInfo.td
===================================================================
--- lib/Target/AMDGPU/SIInstrInfo.td
+++ lib/Target/AMDGPU/SIInstrInfo.td
@@ -309,6 +309,10 @@
def atomic_load_fmin_local : local_binary_atomic_op<SIatomic_fmin>;
def atomic_load_fmax_local : local_binary_atomic_op<SIatomic_fmax>;
+def atomic_fadd_global_noret : global_binary_atomic_op_frag<SIglobal_atomic_fadd>;
+def atomic_pk_fadd_global_noret : global_binary_atomic_op_frag<SIglobal_atomic_pk_fadd>;
+
+
//===----------------------------------------------------------------------===//
// SDNodes PatFrags for loads/stores with a glue input.
// This is for SDNodes and PatFrag for local loads and stores to
Index: lib/Target/AMDGPU/FLATInstructions.td
===================================================================
--- lib/Target/AMDGPU/FLATInstructions.td
+++ lib/Target/AMDGPU/FLATInstructions.td
@@ -686,10 +686,10 @@
let SubtargetPredicate = HasAtomicFaddInsts, is_flat_global = 1 in {
defm GLOBAL_ATOMIC_ADD_F32 : FLAT_Global_Atomic_Pseudo_NO_RTN <
- "global_atomic_add_f32", VGPR_32, f32, atomic_add_global
+ "global_atomic_add_f32", VGPR_32, f32, atomic_fadd_global_noret
>;
defm GLOBAL_ATOMIC_PK_ADD_F16 : FLAT_Global_Atomic_Pseudo_NO_RTN <
- "global_atomic_pk_add_f16", VGPR_32, v2f16, atomic_add_global
+ "global_atomic_pk_add_f16", VGPR_32, v2f16, atomic_pk_fadd_global_noret
>;
} // End SubtargetPredicate = HasAtomicFaddInsts
@@ -847,9 +847,6 @@
} // End OtherPredicates = [HasFlatAddressSpace]
-def atomic_fadd_global : global_binary_atomic_op_frag<SIglobal_atomic_fadd>;
-def atomic_pk_fadd_global : global_binary_atomic_op_frag<SIglobal_atomic_pk_fadd>;
-
let OtherPredicates = [HasFlatGlobalInsts], AddedComplexity = 10 in {
def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, extloadi8_global, i32>;
@@ -930,8 +927,8 @@
def : FlatSignedAtomicPat <GLOBAL_ATOMIC_CMPSWAP_X2_RTN, AMDGPUatomic_cmp_swap_global, i64, v2i64>;
def : FlatSignedAtomicPat <GLOBAL_ATOMIC_XOR_X2_RTN, atomic_xor_global, i64>;
-def : FlatAtomicPatNoRtn <GLOBAL_ATOMIC_ADD_F32, atomic_fadd_global, f32>;
-def : FlatAtomicPatNoRtn <GLOBAL_ATOMIC_PK_ADD_F16, atomic_pk_fadd_global, v2f16>;
+def : FlatAtomicPatNoRtn <GLOBAL_ATOMIC_ADD_F32, atomic_fadd_global_noret, f32>;
+def : FlatAtomicPatNoRtn <GLOBAL_ATOMIC_PK_ADD_F16, atomic_pk_fadd_global_noret, v2f16>;
} // End OtherPredicates = [HasFlatGlobalInsts], AddedComplexity = 10
Index: lib/Target/AMDGPU/BUFInstructions.td
===================================================================
--- lib/Target/AMDGPU/BUFInstructions.td
+++ lib/Target/AMDGPU/BUFInstructions.td
@@ -1043,10 +1043,10 @@
let SubtargetPredicate = HasAtomicFaddInsts in {
defm BUFFER_ATOMIC_ADD_F32 : MUBUF_Pseudo_Atomics_NO_RTN <
- "buffer_atomic_add_f32", VGPR_32, f32, atomic_add_global
+ "buffer_atomic_add_f32", VGPR_32, f32, atomic_fadd_global_noret
>;
defm BUFFER_ATOMIC_PK_ADD_F16 : MUBUF_Pseudo_Atomics_NO_RTN <
- "buffer_atomic_pk_add_f16", VGPR_32, v2f16, atomic_add_global
+ "buffer_atomic_pk_add_f16", VGPR_32, v2f16, atomic_pk_fadd_global_noret
>;
} // End SubtargetPredicate = HasAtomicFaddInsts
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D65077.211062.patch
Type: text/x-patch
Size: 3198 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190722/31cc7185/attachment.bin>
More information about the llvm-commits
mailing list