[llvm] r191156 - Correct the pre-increment load latencies in the PPC A2 itinerary
Hal Finkel
hfinkel at anl.gov
Sat Sep 21 17:08:14 PDT 2013
Author: hfinkel
Date: Sat Sep 21 19:08:14 2013
New Revision: 191156
URL: http://llvm.org/viewvc/llvm-project?rev=191156&view=rev
Log:
Correct the pre-increment load latencies in the PPC A2 itinerary
Pre-increment loads are microcoded on the A2, and the address increment occurs
only after the load completes. As a result, the latency of the GPR address
update is an additional 2 cycles on top of the load latency.
Modified:
llvm/trunk/lib/Target/PowerPC/PPCScheduleA2.td
Modified: llvm/trunk/lib/Target/PowerPC/PPCScheduleA2.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCScheduleA2.td?rev=191156&r1=191155&r2=191156&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCScheduleA2.td (original)
+++ llvm/trunk/lib/Target/PowerPC/PPCScheduleA2.td Sat Sep 21 19:08:14 2013
@@ -70,7 +70,7 @@ def PPCA2Itineraries : ProcessorItinerar
InstrItinData<LdStLoad , [InstrStage<1, [XU]>],
[6, 1, 1]>,
InstrItinData<LdStLoadUpd , [InstrStage<1, [XU]>],
- [6, 2, 1, 1]>,
+ [6, 8, 1, 1]>,
InstrItinData<LdStLDU , [InstrStage<1, [XU]>],
[6, 1, 1]>,
InstrItinData<LdStStore , [InstrStage<1, [XU]>],
@@ -86,11 +86,11 @@ def PPCA2Itineraries : ProcessorItinerar
InstrItinData<LdStLFD , [InstrStage<1, [XU]>],
[7, 1, 1]>,
InstrItinData<LdStLFDU , [InstrStage<1, [XU]>],
- [7, 2, 1, 1]>,
+ [7, 9, 1, 1]>,
InstrItinData<LdStLHA , [InstrStage<1, [XU]>],
[6, 1, 1]>,
InstrItinData<LdStLHAU , [InstrStage<1, [XU]>],
- [6, 2, 1, 1]>,
+ [6, 8, 1, 1]>,
InstrItinData<LdStLWARX , [InstrStage<1, [XU]>],
[82, 1, 1]>, // L2 latency
InstrItinData<LdStSTD , [InstrStage<1, [XU]>],
More information about the llvm-commits
mailing list