From: Ilia Mirkin Date: Mon, 7 Jul 2014 05:53:52 +0000 (-0400) Subject: nvc0/ir: remove merge/split pairs to allow normal propagation to occur X-Git-Tag: upstream/17.1.0~20572 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=5491458843998e8083baf9b62c14895946de1a3f;p=platform%2Fupstream%2Fmesa.git nvc0/ir: remove merge/split pairs to allow normal propagation to occur Because the TGSI interface creates merges for each instruction source and then splits them back out, there are a lot of unnecessary merge/split pairs which do essentially nothing. The various modifier/etc propagation doesn't know how to walk though those, so just remove them when they're unnecessary. Signed-off-by: Ilia Mirkin --- diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp index 62d2ef7..6a4ea4e 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp @@ -118,6 +118,35 @@ CopyPropagation::visit(BasicBlock *bb) // ============================================================================= +class MergeSplits : public Pass +{ +private: + virtual bool visit(BasicBlock *); +}; + +// For SPLIT / MERGE pairs that operate on the same registers, replace the +// post-merge def with the SPLIT's source. +bool +MergeSplits::visit(BasicBlock *bb) +{ + Instruction *i, *next, *si; + + for (i = bb->getEntry(); i; i = next) { + next = i->next; + if (i->op != OP_MERGE || typeSizeof(i->dType) != 8) + continue; + si = i->getSrc(0)->getInsn(); + if (si->op != OP_SPLIT || si != i->getSrc(1)->getInsn()) + continue; + i->def(0).replace(si->getSrc(0), false); + delete_Instruction(prog, i); + } + + return true; +} + +// ============================================================================= + class LoadPropagation : public Pass { private: @@ -2662,6 +2691,7 @@ Program::optimizeSSA(int level) { RUN_PASS(1, DeadCodeElim, buryAll); RUN_PASS(1, CopyPropagation, run); + RUN_PASS(1, MergeSplits, run); RUN_PASS(2, GlobalCSE, run); RUN_PASS(1, LocalCSE, run); RUN_PASS(2, AlgebraicOpt, run);