summaryrefslogtreecommitdiff
path: root/doc/llvm.dox
diff options
context:
space:
mode:
Diffstat (limited to 'doc/llvm.dox')
-rw-r--r--doc/llvm.dox6
1 files changed, 3 insertions, 3 deletions
diff --git a/doc/llvm.dox b/doc/llvm.dox
index fb662ca..8de776a 100644
--- a/doc/llvm.dox
+++ b/doc/llvm.dox
@@ -17,7 +17,7 @@
*
* Once the \c Coal::Program objects holds the source, \c clBuildProgram() can be used to compile it. It does so by invoking \c Coal::Program::build().
*
- * This big function compiles and links the program, so it will be explained later. The actual compilation job is done by \c Coal::Compiler. It does that in \c Coal::Compiler::compile, and then keeps the compilation log and options at hand for future use.
+ * This big function compiles and links the program, so it will be explained later. The actual compilation job is done by \c Coal::Compiler. It does that in \c Coal::Compiler::compile(), and then keeps the compilation log and options at hand for future use.
*
* When a program is compiled, the client application can retrieve it by using \c clGetProgramInfo().
*
@@ -27,7 +27,7 @@
*
* The separation between the unlinked binary and the linked one is the reason for the existence of \c Coal::Program::DeviceDependent::unlinked_binary. The source is compiled to LLVM IR in a module (temporarily stored in linked_module, though it isn't linked yet), that is dumped to unlinked_binary and then linked to form a full executable program.
*
- * So, \c Coal::Program::build() runs its code for every device for which a program must be built. These devices are either given at \c Coal::Program::loadBinaries, or as arguments to \c Coal::Program::build().
+ * So, \c Coal::Program::build() runs its code for every device for which a program must be built. These devices are either given at \c Coal::Program::loadBinaries(), or as arguments to \c Coal::Program::build().
*
* The first step is to see if the program was loaded with sources. If it's the case, they have to be compiled (see \ref compilation).
*
@@ -37,7 +37,7 @@
*
* Then, the device is allowed to add more optimization or analysis passes. \c Coal::CPUProgram::createOptimizationPasses() adds standard link-time optimizations, but hardware-accelerated devices could add autovectorizing, lowering, or analysis passes.
*
- * Finally, \c Coal::DeviceProgram::build is called. It's a no-op function for \c Coal::CPUDevice as it uses directly the module with a LLVM JIT, but hardware devices could use this function to actually compile the program for the target device (LLVM to TGSI transformation for example).
+ * Finally, \c Coal::DeviceProgram::build() is called. It's a no-op function for \c Coal::CPUDevice as it uses directly the module with a LLVM JIT, but hardware devices could use this function to actually compile the program for the target device (LLVM to TGSI transformation for example).
*
* The program is now built and ready to be usable !
*