KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

aco: fix underestimated pressure in spiller when a phi has a killed def 

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5805>
This commit is contained in:
Rhys Perry 2020-07-07 18:16:47 +01:00 committed by Marge Bot
parent 293f251871
commit 5a941f4d6d
1 changed files with 11 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
src/amd/compiler/aco_spill.cpp
View File

@ -604,6 +604,17 @@ RegisterDemand init_live_in_vars(spill_ctx& ctx, Block* block, unsigned block_id
}
}
} else {
for (unsigned i = 0; i < idx; i++) {
aco_ptr<Instruction>& instr = block->instructions[i];
assert(is_phi(instr));
/* Killed phi definitions increase pressure in the predecessor but not
* the block they're in. Since the loops below are both to control
* pressure of the start of this block and the ends of it's
* predecessors, we need to count killed unspilled phi definitions here. */
if (instr->definitions[0].isKill() &&
!ctx.spills_entry[block_idx].count(instr->definitions[0].getTemp()))
reg_pressure += instr->definitions[0].getTemp();
}
idx--;
}
reg_pressure += ctx.register_demand[block_idx][idx] - spilled_registers;