-
Getting Started
-
FAQs
-
Getting Performance
-
Case Studies
-
Offload™ Library
-
C++ Language Extensions
-
Runtime API
-
Command Line Options
- -snvsips3
- -offloadshowptrcalls
- -nomemcpycacheflush
- -minmemcpysize=[n]
- -memcpydmasize=[n]
- -cachesets=[n]
- -warnonouterreadswrites
- -offloadshowptrcalls
- -BEspul[opt]
- -BEspuL[opt]
- -finline-limit=[n]
- -unroll-factor=[n]
- -outputoffloadcode
- -fno-access-control
- -O[n]
- -BEO[n]
- -BEspuopt:[opt]
- -showcallstackonduplerror
- -ignoretrapwordonspu
- -save-temps
- -ocdir [dir]
- -warnouterthiscalls
-
The liboffload header
-
Software caching library
-
Data locality library
-
Constructs library
-
PS3 SDK Function Overloads
-
Technical Information
-
Compiler Errors
- Summary
- Offload1101
- Offload3001
- Offload3005
- Offload3007
- Offload4000
- Offload4001
- Offload4002
- Offload4003
- Offload4105
- Offload4106
- Offload4107
- Offload2401
- Offload2400
- Offload2302
- Offload1108
- Offload1109
- Offload1115
- Offload1122
- Offload1201
- Offload1202
- Offload2000
- Offload2100
- Offload2200
- Offload2301
- Offload4108
-
Compiler Advice
-
Compiler Warnings
-
-
Features
-
Calling PPU functions from SPU
-
Accessing Global Variables
-
PPU virtual methods on SPU
-
Exploiting SPU Performance
-
Using Altivec Instrinsics
-
Using SPU Instrinsics
-
The Software Cache
-
Type Safe Pointers
-
VMX2SPU Engine
-
Compiler Advice
-
Implicit Function Duplication
-
Structure Type Duplication
-
Helpful command line options
-
Offload (SPU) Job functions
-
Core compiler features
-
FAQs
Offloading Guidelines
Moving data definitions into Offload™ blocks
Using Offload™ Block-local Data Leads To Fast SPU Code
This article was updated on 2011-06-03 03:33:54
Within an Offload™ block, code that does not depend on PPU data (defined outside an offload block) will generate fast SPU code. The compiler provides a number of compiler options (Article 80) to spot PPU data dependencies in order to help the programmer to remove or refactor them for performance.
Moving data local onto the SPU (i.e. defining variables inside an offload block) is the essential idea behind the cache classes (Article 136) and the data locality library (Article 71).

