Hi all,
On 31 August 2017 at 23:48, Jason Ekstrand <ja...@jlekstrand.net> wrote:
> We are looking up the execution type prior to checking how many sources
> we have. This leads to looking for a type for src1 on MOV instructions
> which is bogus. On BDW+, the src1 register type overlaps with the
> 64-bit immediate and causes us problems.
> ---
> src/intel/compiler/brw_eu_validate.c | 6 ++----
> 1 file changed, 2 insertions(+), 4 deletions(-)
>
> diff --git a/src/intel/compiler/brw_eu_validate.c
> b/src/intel/compiler/brw_eu_validate.c
> index 249342f..d72c0a0 100644
> --- a/src/intel/compiler/brw_eu_validate.c
> +++ b/src/intel/compiler/brw_eu_validate.c
> @@ -300,15 +300,13 @@ execution_type(const struct gen_device_info *devinfo,
> const brw_inst *inst)
> {
> unsigned num_sources = num_sources_from_inst(devinfo, inst);
> enum brw_reg_type src0_exec_type, src1_exec_type;
> - enum brw_reg_type src0_type = brw_inst_src0_type(devinfo, inst);
> - enum brw_reg_type src1_type = brw_inst_src1_type(devinfo, inst);
>
> /* Execution data type is independent of destination data type, except in
> * mixed F/HF instructions on CHV and SKL+.
> */
> enum brw_reg_type dst_exec_type = brw_inst_dst_type(devinfo, inst);
>
> - src0_exec_type = execution_type_for_type(src0_type);
> + src0_exec_type = execution_type_for_type(brw_inst_src0_type(devinfo,
> inst));
> if (num_sources == 1) {
> if ((devinfo->gen >= 9 || devinfo->is_cherryview) &&
> src0_exec_type == BRW_REGISTER_TYPE_HF) {
> @@ -317,7 +315,7 @@ execution_type(const struct gen_device_info *devinfo,
> const brw_inst *inst)
> return src0_exec_type;
> }
>
> - src1_exec_type = execution_type_for_type(src1_type);
> + src1_exec_type = execution_type_for_type(brw_inst_src1_type(devinfo,
> inst));
As Andres pointed out this commit depends on 4fab67a4415 et al, which
seems to be missing in 17.1 and 17.2.
I've ported this to 17.2 by moving the brw_inst_src[01]_reg_type()
calls as in the original patch.
The brw_inst_src[01]_reg_file() ones are left as-is.
Let me know if extra porting is required.
Thanks
Emil
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
