A better fix would be to do something along the lines of the FS
back-end spilling code and emit a scratch read before any instruction
that overwrites the register to spill partially due to a non-zero
sub-register offset. In the meantime mark registers used with a
non-zero sub-register offset as no-spill to prevent the spilling code
from miscompiling the program.
---
src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp | 5 +++--
src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp | 2 ++
2 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
b/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
index 947bb49..228e04c 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_reg_allocate.cpp
@@ -392,7 +392,8 @@ vec4_visitor::evaluate_spill_costs(float *spill_costs, bool
*no_spill)
*/
if (!can_use_scratch_for_source(inst, i, inst->src[i].nr)) {
spill_costs[inst->src[i].nr] += loop_scale;
- if (inst->src[i].reladdr)
+ if (inst->src[i].reladdr ||
+ inst->src[i].offset % REG_SIZE != 0)
no_spill[inst->src[i].nr] = true;
}
}
@@ -400,7 +401,7 @@ vec4_visitor::evaluate_spill_costs(float *spill_costs, bool
*no_spill)
if (inst->dst.file == VGRF) {
spill_costs[inst->dst.nr] += loop_scale;
- if (inst->dst.reladdr)
+ if (inst->dst.reladdr || inst->dst.offset % REG_SIZE != 0)
no_spill[inst->dst.nr] = true;
}
diff --git a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
index a4e4f40..c88d7b6 100644
--- a/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp
@@ -1481,6 +1481,7 @@ vec4_visitor::emit_scratch_read(bblock_t *block,
vec4_instruction *inst,
dst_reg temp, src_reg orig_src,
int base_offset)
{
+ assert(orig_src.offset % REG_SIZE == 0);
int reg_offset = base_offset + orig_src.offset / REG_SIZE;
src_reg index = get_scratch_offset(block, inst, orig_src.reladdr,
reg_offset);
@@ -1498,6 +1499,7 @@ void
vec4_visitor::emit_scratch_write(bblock_t *block, vec4_instruction *inst,
int base_offset)
{
+ assert(inst->dst.offset % REG_SIZE == 0);
int reg_offset = base_offset + inst->dst.offset / REG_SIZE;
src_reg index = get_scratch_offset(block, inst, inst->dst.reladdr,
reg_offset);
--
2.9.0
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
