diff options
author | Ilia Mirkin <[email protected]> | 2015-04-18 15:00:45 -0400 |
---|---|---|
committer | Ilia Mirkin <[email protected]> | 2015-11-22 23:27:22 -0500 |
commit | 4deb118d06e96731f3481daa72c201d7258bfbbb (patch) | |
tree | 3883913b64bc66fc09f91f38664afc4e58d3ec34 /src/gallium/drivers/nouveau | |
parent | 6463d36394bf95f73cfe3ba6bdf900da431e4e55 (diff) |
nv50/ir: fix (un)spilling of 3-wide results
There is no 96-bit load/store operations, so we have to split it up
into a 32-bit parts, with a split/merge around it.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90348
Signed-off-by: Ilia Mirkin <[email protected]>
Cc: "11.0 11.1" <[email protected]>
Diffstat (limited to 'src/gallium/drivers/nouveau')
-rw-r--r-- | src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp | 46 |
1 files changed, 42 insertions, 4 deletions
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp index 7859c8e79bd..41d2cc9167c 100644 --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp @@ -1573,10 +1573,28 @@ SpillCodeInserter::spill(Instruction *defi, Value *slot, LValue *lval) Instruction *st; if (slot->reg.file == FILE_MEMORY_LOCAL) { - st = new_Instruction(func, OP_STORE, ty); - st->setSrc(0, slot); - st->setSrc(1, lval); lval->noSpill = 1; + if (ty != TYPE_B96) { + st = new_Instruction(func, OP_STORE, ty); + st->setSrc(0, slot); + st->setSrc(1, lval); + } else { + st = new_Instruction(func, OP_SPLIT, ty); + st->setSrc(0, lval); + for (int d = 0; d < lval->reg.size / 4; ++d) + st->setDef(d, new_LValue(func, FILE_GPR)); + + for (int d = lval->reg.size / 4 - 1; d >= 0; --d) { + Value *tmp = cloneShallow(func, slot); + tmp->reg.size = 4; + tmp->reg.data.offset += 4 * d; + + Instruction *s = new_Instruction(func, OP_STORE, TYPE_U32); + s->setSrc(0, tmp); + s->setSrc(1, st->getDef(d)); + defi->bb->insertAfter(defi, s); + } + } } else { st = new_Instruction(func, OP_CVT, ty); st->setDef(0, slot); @@ -1596,7 +1614,27 @@ SpillCodeInserter::unspill(Instruction *usei, LValue *lval, Value *slot) Instruction *ld; if (slot->reg.file == FILE_MEMORY_LOCAL) { lval->noSpill = 1; - ld = new_Instruction(func, OP_LOAD, ty); + if (ty != TYPE_B96) { + ld = new_Instruction(func, OP_LOAD, ty); + } else { + ld = new_Instruction(func, OP_MERGE, ty); + for (int d = 0; d < lval->reg.size / 4; ++d) { + Value *tmp = cloneShallow(func, slot); + LValue *val; + tmp->reg.size = 4; + tmp->reg.data.offset += 4 * d; + + Instruction *l = new_Instruction(func, OP_LOAD, TYPE_U32); + l->setDef(0, (val = new_LValue(func, FILE_GPR))); + l->setSrc(0, tmp); + usei->bb->insertBefore(usei, l); + ld->setSrc(d, val); + val->noSpill = 1; + } + ld->setDef(0, lval); + usei->bb->insertBefore(usei, ld); + return lval; + } } else { ld = new_Instruction(func, OP_CVT, ty); } |