* [PATCH 0/3] nv50/ir: Preapre for running Opts inside a loop @ 2017-04-03 15:55 Karol Herbst 2017-04-03 15:55 ` [PATCH 1/3] nv50/ir: fix AlgebraicOpt for slcts with mods Karol Herbst [not found] ` <20170403155553.996-1-karolherbst-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> 0 siblings, 2 replies; 4+ messages in thread From: Karol Herbst @ 2017-04-03 15:55 UTC (permalink / raw) To: mesa-dev-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Cc: nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Slowly we are getting to the point, that we miss enough optimization opportunities as the result of our own passes. For this we need to fix AlgebraicOpt to be able to handle mods on sources without creating new issues. The last patch enables looping opts. Karol Herbst (3): nv50/ir: fix AlgebraicOpt for slcts with mods nv50/ir: handle logops with NOT in AlgebraicOpt nv50/ir: run some passes multiple times .../drivers/nouveau/codegen/nv50_ir_peephole.cpp | 29 +++++++++++++++------- 1 file changed, 20 insertions(+), 9 deletions(-) -- 2.12.2 _______________________________________________ Nouveau mailing list Nouveau@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/nouveau ^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH 1/3] nv50/ir: fix AlgebraicOpt for slcts with mods 2017-04-03 15:55 [PATCH 0/3] nv50/ir: Preapre for running Opts inside a loop Karol Herbst @ 2017-04-03 15:55 ` Karol Herbst [not found] ` <20170403155553.996-1-karolherbst-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> 1 sibling, 0 replies; 4+ messages in thread From: Karol Herbst @ 2017-04-03 15:55 UTC (permalink / raw) To: mesa-dev; +Cc: nouveau, Karol Herbst From: Karol Herbst <nouveau@karolherbst.de> Signed-off-by: Karol Herbst <karolherbst@gmail.com> --- src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp index 4c92a1efb5..bd60a84998 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp @@ -1797,10 +1797,10 @@ AlgebraicOpt::handleSLCT(Instruction *slct) if (slct->getSrc(2)->asImm()->compare(slct->asCmp()->setCond, 0.0f)) slct->setSrc(0, slct->getSrc(1)); } else - if (slct->getSrc(0) != slct->getSrc(1)) { + if (slct->getSrc(0) != slct->getSrc(1) || slct->src(0).mod != slct->src(1).mod) return; - } - slct->op = OP_MOV; + slct->op = slct->src(0).mod.getOp(); + slct->src(0).mod = slct->src(0).mod ^ Modifier(slct->op); slct->setSrc(1, NULL); slct->setSrc(2, NULL); } -- 2.12.2 _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev ^ permalink raw reply related [flat|nested] 4+ messages in thread
[parent not found: <20170403155553.996-1-karolherbst-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]
* [PATCH 2/3] nv50/ir: handle logops with NOT in AlgebraicOpt [not found] ` <20170403155553.996-1-karolherbst-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> @ 2017-04-03 15:55 ` Karol Herbst 2017-04-03 15:55 ` [PATCH 3/3] nv50/ir: run some passes multiple times Karol Herbst 1 sibling, 0 replies; 4+ messages in thread From: Karol Herbst @ 2017-04-03 15:55 UTC (permalink / raw) To: mesa-dev-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Cc: nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Signed-off-by: Karol Herbst <karolherbst@gmail.com> --- src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp index bd60a84998..0de84fe9fc 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp @@ -1856,6 +1856,12 @@ AlgebraicOpt::handleLOGOP(Instruction *logop) set0 = cloneForward(func, set0); set1 = cloneShallow(func, set1); + + if (logop->src(0).mod == Modifier(NV50_IR_MOD_NOT)) + set0->asCmp()->setCond = inverseCondCode(set0->asCmp()->setCond); + if (logop->src(1).mod == Modifier(NV50_IR_MOD_NOT)) + set1->asCmp()->setCond = inverseCondCode(set1->asCmp()->setCond); + logop->bb->insertAfter(logop, set1); logop->bb->insertAfter(logop, set0); -- 2.12.2 _______________________________________________ Nouveau mailing list Nouveau@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/nouveau ^ permalink raw reply related [flat|nested] 4+ messages in thread
* [PATCH 3/3] nv50/ir: run some passes multiple times [not found] ` <20170403155553.996-1-karolherbst-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> 2017-04-03 15:55 ` [PATCH 2/3] nv50/ir: handle logops with NOT in AlgebraicOpt Karol Herbst @ 2017-04-03 15:55 ` Karol Herbst 1 sibling, 0 replies; 4+ messages in thread From: Karol Herbst @ 2017-04-03 15:55 UTC (permalink / raw) To: mesa-dev-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW Cc: nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Karol Herbst From: Karol Herbst <nouveau@karolherbst.de> With the shader cache, compilation time matters less. As a side effect we can write more optimizations to produce better optimized code. total instructions in shared programs : 3931743 -> 3917512 (-0.36%) total gprs used in shared programs : 481460 -> 481680 (0.05%) total local used in shared programs : 27481 -> 26761 (-2.62%) total bytes used in shared programs : 36032672 -> 35902648 (-0.36%) local gpr inst bytes helped 48 133 3843 3843 hurt 1 295 75 75 Signed-off-by: Karol Herbst <karolherbst@gmail.com> --- .../drivers/nouveau/codegen/nv50_ir_peephole.cpp | 17 +++++++++++------ 1 file changed, 11 insertions(+), 6 deletions(-) diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp index 0de84fe9fc..505de08573 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp @@ -3729,12 +3729,17 @@ Program::optimizeSSA(int level) 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); - RUN_PASS(2, ModifierFolding, run); // before load propagation -> less checks - RUN_PASS(1, ConstantFolding, foldAll); - RUN_PASS(2, LateAlgebraicOpt, run); - RUN_PASS(1, Split64BitOpPreRA, run); + for (int i = 0; i < 2; ++i) { + RUN_PASS(1, LocalCSE, run); + RUN_PASS(2, AlgebraicOpt, run); + RUN_PASS(2, ModifierFolding, run); // before load propagation -> less checks + RUN_PASS(1, ConstantFolding, foldAll); + RUN_PASS(2, LateAlgebraicOpt, run); + // only once + if (i == 0) + RUN_PASS(1, Split64BitOpPreRA, run); + RUN_PASS(1, DeadCodeElim, buryAll); + } RUN_PASS(1, LoadPropagation, run); RUN_PASS(1, IndirectPropagation, run); RUN_PASS(2, MemoryOpt, run); -- 2.12.2 _______________________________________________ Nouveau mailing list Nouveau@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/nouveau ^ permalink raw reply related [flat|nested] 4+ messages in thread
end of thread, other threads:[~2017-04-03 15:55 UTC | newest] Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2017-04-03 15:55 [PATCH 0/3] nv50/ir: Preapre for running Opts inside a loop Karol Herbst 2017-04-03 15:55 ` [PATCH 1/3] nv50/ir: fix AlgebraicOpt for slcts with mods Karol Herbst [not found] ` <20170403155553.996-1-karolherbst-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> 2017-04-03 15:55 ` [PATCH 2/3] nv50/ir: handle logops with NOT in AlgebraicOpt Karol Herbst 2017-04-03 15:55 ` [PATCH 3/3] nv50/ir: run some passes multiple times Karol Herbst
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.