fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-24 15:20:10 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

gm107/ir: Emit third src 'bound' and optional predicate output of SHFL

Browse source 
v2: Emit the original hard-coded 0x1c03 when OP_SHFL is used in gm107's
    lowering (Samuel Pitoiset)

Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This commit is contained in:
Boyan Ding 2017-04-10 22:55:57 +08:00 • committed by Ilia Mirkin
parent a981e68c26
commit 854554c314
2 changed files with 29 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gm107.cpp
View file

@ -967,11 +967,26 @@ CodeEmitterGM107::emitSHFL()
break;
}
/*XXX: what is this arg? hardcode immediate for now */
emitField(0x22, 13, 0x1c03);
type |= 2;
switch (insn->src(2).getFile()) {
case FILE_GPR:
emitGPR(0x27, insn->src(2));
break;
case FILE_IMMEDIATE:
emitIMMD(0x22, 13, insn->src(2));
type |= 2;
break;
default:
assert(!"invalid src2 file");
break;
}
if (!insn->defExists(1))
emitPRED(0x30);
else {
assert(insn->def(1).getFile() == FILE_PREDICATE);
emitPRED(0x30, insn->def(1));
}
emitPRED (0x30);
emitField(0x1e, 2, insn->subOp);
emitField(0x1c, 2, type);
emitGPR (0x08, insn->src(0));

15
src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_gm107.cpp
View file

@ -41,6 +41,8 @@ namespace nv50_ir {
((QOP_##q << 6) | (QOP_##r << 4) | \
(QOP_##s << 2) | (QOP_##t << 0))
#define SHFL_BOUND_QUAD 0x1c03
void
GM107LegalizeSSA::handlePFETCH(Instruction *i)
{
@ -120,7 +122,8 @@ GM107LoweringPass::handleManualTXD(TexInstruction *i)
// mov coordinates from lane l to all lanes
bld.mkOp(OP_QUADON, TYPE_NONE, NULL);
for (c = 0; c < dim; ++c) {
bld.mkOp2(OP_SHFL, TYPE_F32, crd[c], i->getSrc(c + array), bld.mkImm(l));
bld.mkOp3(OP_SHFL, TYPE_F32, crd[c], i->getSrc(c + array),
bld.mkImm(l), bld.mkImm(SHFL_BOUND_QUAD));
add = bld.mkOp2(OP_QUADOP, TYPE_F32, crd[c], crd[c], zero);
add->subOp = 0x00;
add->lanes = 1; /* abused for .ndv */
@ -128,7 +131,8 @@ GM107LoweringPass::handleManualTXD(TexInstruction *i)
// add dPdx from lane l to lanes dx
for (c = 0; c < dim; ++c) {
bld.mkOp2(OP_SHFL, TYPE_F32, tmp, i->dPdx[c].get(), bld.mkImm(l));
bld.mkOp3(OP_SHFL, TYPE_F32, tmp, i->dPdx[c].get(), bld.mkImm(l),
bld.mkImm(SHFL_BOUND_QUAD));
add = bld.mkOp2(OP_QUADOP, TYPE_F32, crd[c], tmp, crd[c]);
add->subOp = qOps[l][0];
add->lanes = 1; /* abused for .ndv */
@ -136,7 +140,8 @@ GM107LoweringPass::handleManualTXD(TexInstruction *i)
// add dPdy from lane l to lanes dy
for (c = 0; c < dim; ++c) {
bld.mkOp2(OP_SHFL, TYPE_F32, tmp, i->dPdy[c].get(), bld.mkImm(l));
bld.mkOp3(OP_SHFL, TYPE_F32, tmp, i->dPdy[c].get(), bld.mkImm(l),
bld.mkImm(SHFL_BOUND_QUAD));
add = bld.mkOp2(OP_QUADOP, TYPE_F32, crd[c], tmp, crd[c]);
add->subOp = qOps[l][1];
add->lanes = 1; /* abused for .ndv */
@ -203,8 +208,8 @@ GM107LoweringPass::handleDFDX(Instruction *insn)
break;
}
shfl = bld.mkOp2(OP_SHFL, TYPE_F32, bld.getScratch(),
insn->getSrc(0), bld.mkImm(xid));
shfl = bld.mkOp3(OP_SHFL, TYPE_F32, bld.getScratch(), insn->getSrc(0),
bld.mkImm(xid), bld.mkImm(SHFL_BOUND_QUAD));
shfl->subOp = NV50_IR_SUBOP_SHFL_BFLY;
insn->op = OP_QUADOP;
insn->subOp = qop;