diff options
| author | Rhys Perry <pendingchaos02@gmail.com> | 2022-12-08 19:32:25 +0000 |
|---|---|---|
| committer | Eric Engestrom <eric@engestrom.ch> | 2022-12-14 20:47:01 +0000 |
| commit | 2c738559376950bb040c69cd869dcf2f81daa71f (patch) | |
| tree | 6a4db9555372b9f407e9b0c90b1c4dd1957cb0e1 | |
| parent | 6922882d8ff0b75d99a1458e2951e0b72ee3001a (diff) | |
| download | mesa-2c738559376950bb040c69cd869dcf2f81daa71f.tar.gz | |
aco/ra: don't swap p_create_vector operand with definition blocker for scc
SCC is 1-bit, and we can't copy a 32-bit value into it.
Fixes dEQP-VK.spirv_assembly.type.scalar.i32.iequal_tesse with
ACO_DEBUG=noopt.
No fossil-db changes.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: 9476986e6f6 ("aco/ra: special-case get_reg_for_create_vector_copy()")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20240>
(cherry picked from commit a05dd58309caf7e3857eb7f02084d8c9667c9efc)
| -rw-r--r-- | .pick_status.json | 2 | ||||
| -rw-r--r-- | src/amd/compiler/aco_register_allocation.cpp | 2 |
2 files changed, 2 insertions, 2 deletions
diff --git a/.pick_status.json b/.pick_status.json index 2cb6b52a1c8..152c704a1fe 100644 --- a/.pick_status.json +++ b/.pick_status.json @@ -1831,7 +1831,7 @@ "description": "aco/ra: don't swap p_create_vector operand with definition blocker for scc", "nominated": true, "nomination_type": 1, - "resolution": 0, + "resolution": 1, "main_sha": null, "because_sha": "9476986e6f6fe49ef8bc511f4ebeca4232263888" }, diff --git a/src/amd/compiler/aco_register_allocation.cpp b/src/amd/compiler/aco_register_allocation.cpp index eef4ff06fbb..c59c6353815 100644 --- a/src/amd/compiler/aco_register_allocation.cpp +++ b/src/amd/compiler/aco_register_allocation.cpp @@ -1091,7 +1091,7 @@ get_reg_for_create_vector_copy(ra_ctx& ctx, RegisterFile& reg_file, assignment& op = ctx.assignments[instr->operands[i].tempId()]; /* if everything matches, create parallelcopy for the killed operand */ if (!intersects(def_reg, PhysRegInterval{op.reg, op.rc.size()}) && - reg_file.get_id(op.reg) == instr->operands[i].tempId()) { + op.reg != scc && reg_file.get_id(op.reg) == instr->operands[i].tempId()) { Definition pc_def = Definition(reg, info.rc); parallelcopies.emplace_back(instr->operands[i], pc_def); return {op.reg, true}; |