diff options
author | Samuel Pitoiset <[email protected]> | 2016-03-29 19:27:49 +0200 |
---|---|---|
committer | Samuel Pitoiset <[email protected]> | 2016-03-29 19:55:51 +0200 |
commit | 9d57c84994abe45133382cf72ae617570bfe89da (patch) | |
tree | 4e7885ed93f1eadf29c40f85e35d3c88d13ab2ef | |
parent | cc68dc2b5e3c603580f70d682b0772d179b28ce3 (diff) |
nvc0/ir: move load/store lowering pass to handleLDST()
Having all this code in a big switch is not really a good pratice.
Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Ilia Mirkin <[email protected]>
-rw-r--r-- | src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 114 | ||||
-rw-r--r-- | src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h | 1 |
2 files changed, 61 insertions, 54 deletions
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp index 68a30ecb8d7..850147b62e9 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp @@ -1692,6 +1692,65 @@ NVC0LoweringPass::handleWRSV(Instruction *i) } void +NVC0LoweringPass::handleLDST(Instruction *i) +{ + if (i->src(0).getFile() == FILE_SHADER_INPUT) { + if (prog->getType() == Program::TYPE_COMPUTE) { + i->getSrc(0)->reg.file = FILE_MEMORY_CONST; + i->getSrc(0)->reg.fileIndex = 0; + } else + if (prog->getType() == Program::TYPE_GEOMETRY && + i->src(0).isIndirect(0)) { + // XXX: this assumes vec4 units + Value *ptr = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getSSA(), + i->getIndirect(0, 0), bld.mkImm(4)); + i->setIndirect(0, 0, ptr); + i->op = OP_VFETCH; + } else { + i->op = OP_VFETCH; + assert(prog->getType() != Program::TYPE_FRAGMENT); // INTERP + } + } else if (i->src(0).getFile() == FILE_MEMORY_CONST) { + if (i->src(0).isIndirect(1)) { + Value *ptr; + if (i->src(0).isIndirect(0)) + ptr = bld.mkOp3v(OP_INSBF, TYPE_U32, bld.getSSA(), + i->getIndirect(0, 1), bld.mkImm(0x1010), + i->getIndirect(0, 0)); + else + ptr = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getSSA(), + i->getIndirect(0, 1), bld.mkImm(16)); + i->setIndirect(0, 1, NULL); + i->setIndirect(0, 0, ptr); + i->subOp = NV50_IR_SUBOP_LDC_IS; + } + } else if (i->src(0).getFile() == FILE_SHADER_OUTPUT) { + assert(prog->getType() == Program::TYPE_TESSELLATION_CONTROL); + i->op = OP_VFETCH; + } else if (i->src(0).getFile() == FILE_MEMORY_GLOBAL) { + Value *ind = i->getIndirect(0, 1); + Value *ptr = loadBufInfo64(ind, i->getSrc(0)->reg.fileIndex * 16); + // XXX come up with a way not to do this for EVERY little access but + // rather to batch these up somehow. Unfortunately we've lost the + // information about the field width by the time we get here. + Value *offset = bld.loadImm(NULL, i->getSrc(0)->reg.data.offset + typeSizeof(i->sType)); + Value *length = loadBufLength32(ind, i->getSrc(0)->reg.fileIndex * 16); + Value *pred = new_LValue(func, FILE_PREDICATE); + if (i->src(0).isIndirect(0)) { + bld.mkOp2(OP_ADD, TYPE_U64, ptr, ptr, i->getIndirect(0, 0)); + bld.mkOp2(OP_ADD, TYPE_U32, offset, offset, i->getIndirect(0, 0)); + } + i->setIndirect(0, 1, NULL); + i->setIndirect(0, 0, ptr); + bld.mkCmp(OP_SET, CC_GT, TYPE_U32, pred, TYPE_U32, offset, length); + i->setPredicate(CC_NOT_P, pred); + if (i->defExists(0)) { + bld.mkMov(i->getDef(0), bld.mkImm(0)); + } + } +} + +void NVC0LoweringPass::readTessCoord(LValue *dst, int c) { Value *laneid = bld.getSSA(); @@ -2016,60 +2075,7 @@ NVC0LoweringPass::visit(Instruction *i) return handleWRSV(i); case OP_STORE: case OP_LOAD: - if (i->src(0).getFile() == FILE_SHADER_INPUT) { - if (prog->getType() == Program::TYPE_COMPUTE) { - i->getSrc(0)->reg.file = FILE_MEMORY_CONST; - i->getSrc(0)->reg.fileIndex = 0; - } else - if (prog->getType() == Program::TYPE_GEOMETRY && - i->src(0).isIndirect(0)) { - // XXX: this assumes vec4 units - Value *ptr = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getSSA(), - i->getIndirect(0, 0), bld.mkImm(4)); - i->setIndirect(0, 0, ptr); - i->op = OP_VFETCH; - } else { - i->op = OP_VFETCH; - assert(prog->getType() != Program::TYPE_FRAGMENT); // INTERP - } - } else if (i->src(0).getFile() == FILE_MEMORY_CONST) { - if (i->src(0).isIndirect(1)) { - Value *ptr; - if (i->src(0).isIndirect(0)) - ptr = bld.mkOp3v(OP_INSBF, TYPE_U32, bld.getSSA(), - i->getIndirect(0, 1), bld.mkImm(0x1010), - i->getIndirect(0, 0)); - else - ptr = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getSSA(), - i->getIndirect(0, 1), bld.mkImm(16)); - i->setIndirect(0, 1, NULL); - i->setIndirect(0, 0, ptr); - i->subOp = NV50_IR_SUBOP_LDC_IS; - } - } else if (i->src(0).getFile() == FILE_SHADER_OUTPUT) { - assert(prog->getType() == Program::TYPE_TESSELLATION_CONTROL); - i->op = OP_VFETCH; - } else if (i->src(0).getFile() == FILE_MEMORY_GLOBAL) { - Value *ind = i->getIndirect(0, 1); - Value *ptr = loadBufInfo64(ind, i->getSrc(0)->reg.fileIndex * 16); - // XXX come up with a way not to do this for EVERY little access but - // rather to batch these up somehow. Unfortunately we've lost the - // information about the field width by the time we get here. - Value *offset = bld.loadImm(NULL, i->getSrc(0)->reg.data.offset + typeSizeof(i->sType)); - Value *length = loadBufLength32(ind, i->getSrc(0)->reg.fileIndex * 16); - Value *pred = new_LValue(func, FILE_PREDICATE); - if (i->src(0).isIndirect(0)) { - bld.mkOp2(OP_ADD, TYPE_U64, ptr, ptr, i->getIndirect(0, 0)); - bld.mkOp2(OP_ADD, TYPE_U32, offset, offset, i->getIndirect(0, 0)); - } - i->setIndirect(0, 1, NULL); - i->setIndirect(0, 0, ptr); - bld.mkCmp(OP_SET, CC_GT, TYPE_U32, pred, TYPE_U32, offset, length); - i->setPredicate(CC_NOT_P, pred); - if (i->defExists(0)) { - bld.mkMov(i->getDef(0), bld.mkImm(0)); - } - } + handleLDST(i); break; case OP_ATOM: { diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h index d2cb23f45d5..be81d29eb0a 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.h @@ -106,6 +106,7 @@ protected: bool handleCasExch(Instruction *, bool needCctl); void handleSurfaceOpNVE4(TexInstruction *); void handleSharedATOM(Instruction *); + void handleLDST(Instruction *); void checkPredicate(Instruction *); |