Prepared by: | Gustavo Serra Scalet |
---|---|
Workspace: | /home/gut/hs/hotspot |
Compare against: | http://hg.openjdk.java.net/jdk9/hs/hotspot |
Compare against version: | 12309 |
Summary of changes: | 100 lines changed: 64 ins; 8 del; 28 mod; 20503 unchg |
Changeset: | hotspot.changeset |
Bug id: | JDK-8170094 : PPC64: Keep immediate value 0 cached into a register to improve performance |
Legend: |
Modified file Deleted file New file |
Cdiffs
Udiffs
Sdiffs
Frames
Old
New
-----
Raw
src/cpu/ppc/vm/c1_Defs_ppc.hpp
rev 12310 : Reserve R30 to a cleared content register on C1 and C2 code Several times a 0 is loaded to a register as a temporary value. This can be improved by caching a 0 into a register. I didn't notice a performance drop since only applying this patch showed no drop of performance, hence there are more registers available than normally needed and this caching technique can be applied. Despite setting R30_zero as a dedicated register and initialized with 0 for the C1 and C2 code, new rules for storing 0 related to stb,sth,stw,std were added.2 lines changed: 0 ins; 0 del; 2 mod; 74 unchg
Cdiffs
Udiffs
Sdiffs
Frames
Old
New
-----
Raw
src/cpu/ppc/vm/c1_FrameMap_ppc.cpp
rev 12310 : Reserve R30 to a cleared content register on C1 and C2 code Several times a 0 is loaded to a register as a temporary value. This can be improved by caching a 0 into a register. I didn't notice a performance drop since only applying this patch showed no drop of performance, hence there are more registers available than normally needed and this caching technique can be applied. Despite setting R30_zero as a dedicated register and initialized with 0 for the C1 and C2 code, new rules for storing 0 related to stb,sth,stw,std were added.4 lines changed: 0 ins; 0 del; 4 mod; 390 unchg
Cdiffs
Udiffs
Sdiffs
Frames
Old
New
-----
Raw
src/cpu/ppc/vm/c1_FrameMap_ppc.hpp
rev 12310 : Reserve R30 to a cleared content register on C1 and C2 code Several times a 0 is loaded to a register as a temporary value. This can be improved by caching a 0 into a register. I didn't notice a performance drop since only applying this patch showed no drop of performance, hence there are more registers available than normally needed and this caching technique can be applied. Despite setting R30_zero as a dedicated register and initialized with 0 for the C1 and C2 code, new rules for storing 0 related to stb,sth,stw,std were added.4 lines changed: 1 ins; 0 del; 3 mod; 199 unchg
Cdiffs
Udiffs
Sdiffs
Frames
Old
New
-----
Raw
src/cpu/ppc/vm/ppc.ad
rev 12310 : Reserve R30 to a cleared content register on C1 and C2 code Several times a 0 is loaded to a register as a temporary value. This can be improved by caching a 0 into a register. I didn't notice a performance drop since only applying this patch showed no drop of performance, hence there are more registers available than normally needed and this caching technique can be applied. Despite setting R30_zero as a dedicated register and initialized with 0 for the C1 and C2 code, new rules for storing 0 related to stb,sth,stw,std were added.84 lines changed: 57 ins; 8 del; 19 mod; 13283 unchg
Cdiffs
Udiffs
Sdiffs
Frames
Old
New
-----
Raw
src/cpu/ppc/vm/register_ppc.hpp
rev 12310 : Reserve R30 to a cleared content register on C1 and C2 code Several times a 0 is loaded to a register as a temporary value. This can be improved by caching a 0 into a register. I didn't notice a performance drop since only applying this patch showed no drop of performance, hence there are more registers available than normally needed and this caching technique can be applied. Despite setting R30_zero as a dedicated register and initialized with 0 for the C1 and C2 code, new rules for storing 0 related to stb,sth,stw,std were added.2 lines changed: 2 ins; 0 del; 0 mod; 752 unchg
Cdiffs
Udiffs
Sdiffs
Frames
Old
New
-----
Raw
src/cpu/ppc/vm/stubGenerator_ppc.cpp
rev 12310 : Reserve R30 to a cleared content register on C1 and C2 code Several times a 0 is loaded to a register as a temporary value. This can be improved by caching a 0 into a register. I didn't notice a performance drop since only applying this patch showed no drop of performance, hence there are more registers available than normally needed and this caching technique can be applied. Despite setting R30_zero as a dedicated register and initialized with 0 for the C1 and C2 code, new rules for storing 0 related to stb,sth,stw,std were added.3 lines changed: 3 ins; 0 del; 0 mod; 3448 unchg
Cdiffs
Udiffs
Sdiffs
Frames
Old
New
-----
Raw
src/cpu/ppc/vm/templateInterpreterGenerator_ppc.cpp
rev 12310 : Reserve R30 to a cleared content register on C1 and C2 code Several times a 0 is loaded to a register as a temporary value. This can be improved by caching a 0 into a register. I didn't notice a performance drop since only applying this patch showed no drop of performance, hence there are more registers available than normally needed and this caching technique can be applied. Despite setting R30_zero as a dedicated register and initialized with 0 for the C1 and C2 code, new rules for storing 0 related to stb,sth,stw,std were added.1 line changed: 1 ins; 0 del; 0 mod; 2357 unchg
This code review page was prepared using ./webrev.ksh (vers 25.16-hg+openjdk.java.net).