[Beignet] [PATCH] GBE: Do topological sorting of basicblocks.
Zhigang Gong
zhigang.gong at linux.intel.com
Sun Nov 9 22:38:41 PST 2014
Nice patch. Pushed, thanks.
And just as we discussed, still need an assert check before liveness anaylsis to
make sure there is no backward jump which doesn't belong to any loop.
On Mon, Nov 10, 2014 at 11:31:34AM +0800, Ruiling Song wrote:
> Toplogical sorting have two big advantages:
> 1. Sorted basicblocks will reduce unneccesary register pressure.
> 2. Sorted basicblocks will make liveness analysis easier.
>
> This patch fix opencv failures:
> ./opencv_test_video --gtest_filter=OCL_OCL_Video/Mog2_Update.Accuracy/1
> ./opencv_test_imgproc --gtest_filter=OCL_ImageProc/Filter2D.Mat/482
>
> Signed-off-by: Ruiling Song <ruiling.song at intel.com>
> ---
> backend/src/llvm/llvm_gen_backend.cpp | 53 ++++++++++++++++++++++++++++++++-
> 1 file changed, 52 insertions(+), 1 deletion(-)
>
> diff --git a/backend/src/llvm/llvm_gen_backend.cpp b/backend/src/llvm/llvm_gen_backend.cpp
> index b1b9382..cb0511a 100644
> --- a/backend/src/llvm/llvm_gen_backend.cpp
> +++ b/backend/src/llvm/llvm_gen_backend.cpp
> @@ -112,6 +112,7 @@
> #include "llvm/Target/Mangler.h"
> #endif
>
> +#include "llvm/ADT/PostOrderIterator.h"
> #include "llvm/Transforms/Scalar.h"
> #include "llvm/MC/MCAsmInfo.h"
> #include "llvm/MC/MCContext.h"
> @@ -542,6 +543,8 @@ namespace gbe
> void allocateGlobalVariableRegister(Function &F);
> /*! gather all the loops in the function and add them to ir::Function */
> void gatherLoopInfo(ir::Function &fn);
> + /*! do topological sorting of basicblocks */
> + void sortBasicBlock(Function &F);
> /*! Emit the complete function code and declaration */
> void emitFunction(Function &F);
> /*! Handle input and output function parameters */
> @@ -1893,7 +1896,53 @@ namespace gbe
> fn.addLoop(loopBBs, loopExits);
> }
> }
> -
> +/*!
> +
> + Sorting Basic blocks is mainly used to solve register liveness issue, take a
> + look at below CFG:
> +
> + -<--1--
> + | |
> + | ->2
> + -- 3 <--- |
> + | ^ | -->4--
> + | | | | |
> + | | -----5<-- |
> + | | |
> + | ----------6<-----
> + |
> + -->7
> +
> + A register %10 defined in bb4, and used in bb5 & bb6. In normal liveness
> + analysis, %10 is not alive in bb3. But under simd execution model, after
> + executing bb4, some channel jump through bb5 to bb3, other channel may jump
> + to bb6, we must execute bb3 first, then bb6, to avoid missing instructions.
> + The physical register of %10 was assigned some value in bb4, but when
> + executing bb3, its content may be over-written as it is dead in bb3. When
> + jumping back to execute bb6, it will get polluted data. What a disaster!
> + What we do here is do a topological sorting of basic blocks, For this case
> + we can see the bb3 will be placed after bb5 & bb6. The liveness calculation
> + is just as normal and will be correct.
> +
> + Another advantage of sorting basic blocks is reducing register pressure.
> + In the above CFG, a register defined in bb3 and used in bb7 will be
> + alive through 3,4,5,6,7. But in fact it should be only alive in bb3 and bb7.
> + After topological sorting, this kind of register would be only alive in bb3
> + and bb7. Register pressure in 4,5,6 is reduced.
> +*/
> +
> + void GenWriter::sortBasicBlock(Function &F) {
> + typedef ReversePostOrderTraversal<Function*> RPOTType;
> + RPOTType rpot(&F);
> + Function::BasicBlockListType &bbList = F.getBasicBlockList();
> +
> + for (RPOTType::rpo_iterator bbI = rpot.begin(), bbE = rpot.end(); bbI != bbE; ++bbI) {
> + (*bbI)->removeFromParent();
> + }
> + for (RPOTType::rpo_iterator bbI = rpot.begin(), bbE = rpot.end(); bbI != bbE; ++bbI) {
> + bbList.push_back(*bbI);
> + }
> + }
> void GenWriter::emitFunction(Function &F)
> {
> switch (F.getCallingConv()) {
> @@ -1915,6 +1964,8 @@ namespace gbe
> this->emitFunctionPrototype(F);
>
> this->allocateGlobalVariableRegister(F);
> +
> + sortBasicBlock(F);
> // Visit all the instructions and emit the IR registers or the value to
> // value mapping when a new register is not needed
> pass = PASS_EMIT_REGISTERS;
> --
> 1.7.10.4
>
> _______________________________________________
> Beignet mailing list
> Beignet at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/beignet
More information about the Beignet
mailing list