From 7b9475fa652b9df6d599edbea8fa5049fdd995e1 Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Fri, 9 May 2014 15:56:03 +1000 Subject: [PATCH] nvc0: maxwell isa has no per-instruction join modifier Signed-off-by: Ben Skeggs Reviewed-by: Ilia Mirkin --- .../drivers/nouveau/codegen/nv50_ir_peephole.cpp | 34 ++++++++++++---------- .../drivers/nouveau/codegen/nv50_ir_target.h | 3 +- .../nouveau/codegen/nv50_ir_target_nv50.cpp | 2 +- .../nouveau/codegen/nv50_ir_target_nvc0.cpp | 3 +- 4 files changed, 23 insertions(+), 19 deletions(-) diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp index cdae3c8..3005922 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp @@ -2067,22 +2067,24 @@ FlatteningPass::visit(BasicBlock *bb) return true; // try to attach join to previous instruction - Instruction *insn = bb->getExit(); - if (insn && insn->op == OP_JOIN && !insn->getPredicate()) { - insn = insn->prev; - if (insn && !insn->getPredicate() && - !insn->asFlow() && - insn->op != OP_TEXBAR && - !isTextureOp(insn->op) && // probably just nve4 - !isSurfaceOp(insn->op) && // not confirmed - insn->op != OP_LINTERP && // probably just nve4 - insn->op != OP_PINTERP && // probably just nve4 - ((insn->op != OP_LOAD && insn->op != OP_STORE) || - typeSizeof(insn->dType) <= 4) && - !insn->isNop()) { - insn->join = 1; - bb->remove(bb->getExit()); - return true; + if (prog->getTarget()->hasJoin) { + Instruction *insn = bb->getExit(); + if (insn && insn->op == OP_JOIN && !insn->getPredicate()) { + insn = insn->prev; + if (insn && !insn->getPredicate() && + !insn->asFlow() && + insn->op != OP_TEXBAR && + !isTextureOp(insn->op) && // probably just nve4 + !isSurfaceOp(insn->op) && // not confirmed + insn->op != OP_LINTERP && // probably just nve4 + insn->op != OP_PINTERP && // probably just nve4 + ((insn->op != OP_LOAD && insn->op != OP_STORE) || + typeSizeof(insn->dType) <= 4) && + !insn->isNop()) { + insn->join = 1; + bb->remove(bb->getExit()); + return true; + } } } diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target.h b/src/gallium/drivers/nouveau/codegen/nv50_ir_target.h index 9eb0a1f..711056e 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target.h +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target.h @@ -120,7 +120,7 @@ enum OpClass class Target { public: - Target(bool j, bool s) : joinAnterior(j), hasSWSched(s) { } + Target(bool m, bool j, bool s) : hasJoin(m), joinAnterior(j), hasSWSched(s) { } virtual ~Target() { } static Target *create(uint32_t chipset); @@ -192,6 +192,7 @@ public: virtual uint32_t getSVAddress(DataFile, const Symbol *) const = 0; public: + const bool hasJoin; // true if instructions have a join modifier const bool joinAnterior; // true if join is executed before the op const bool hasSWSched; // true if code should provide scheduling data diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp index 32ba5c8..799ac2f 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp @@ -29,7 +29,7 @@ Target *getTargetNV50(unsigned int chipset) return new TargetNV50(chipset); } -TargetNV50::TargetNV50(unsigned int card) : Target(true, false) +TargetNV50::TargetNV50(unsigned int card) : Target(true, true, false) { chipset = card; diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp index adf2df8..064e7a2 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp @@ -29,7 +29,8 @@ Target *getTargetNVC0(unsigned int chipset) return new TargetNVC0(chipset); } -TargetNVC0::TargetNVC0(unsigned int card) : Target(false, card >= 0xe4) +TargetNVC0::TargetNVC0(unsigned int card) : + Target(card < 0x110, false, card >= 0xe4) { chipset = card; initOpInfo(); -- 2.7.4