On 05/19/2014 11:55 AM, Matt Turner wrote:
> ---
> src/mesa/drivers/dri/i965/brw_fs_generator.cpp | 4 ++--
> src/mesa/drivers/dri/i965/gen8_fs_generator.cpp | 4 ++--
> 2 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> index 872b5a4..914fb29 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> @@ -1843,12 +1843,12 @@ fs_generator::generate_assembly(exec_list
> *simd8_instructions,
>
> if (simd16_instructions) {
> /* align to 64 byte boundary. */
> - while ((p->nr_insn * sizeof(struct brw_instruction)) % 64) {
> + while (p->next_insn_offset % 64) {
> brw_NOP(p);
> }
So, I wondered how this worked (before and after your change). In
theory, we could have an odd number of compacted instructions, giving us
a next_insn_offset of (64*n + 2). Emitting uncompacted NOPs adds 4
bytes each time, which would never get us to a multiple of 64.
Apparently, this doesn't happen because sendc cannot be compacted and
must be aligned on a 128-bit/4-byte boundary. So, it works.
Someday we should probably just do:
p->next_insn_offset = ALIGN(p->next_insn_offset, 64);
and memset to 0 between the two programs.
In the meantime, this seems better than the existing code.
Reviewed-by: Kenneth Graunke <kenn...@whitecape.org>
>
> /* Save off the start of this SIMD16 program */
> - prog_data->prog_offset_16 = p->nr_insn * sizeof(struct
> brw_instruction);
> + prog_data->prog_offset_16 = p->next_insn_offset;
>
> brw_set_compression_control(p, BRW_COMPRESSION_COMPRESSED);
>
> diff --git a/src/mesa/drivers/dri/i965/gen8_fs_generator.cpp
> b/src/mesa/drivers/dri/i965/gen8_fs_generator.cpp
> index 9011bff..272f668 100644
> --- a/src/mesa/drivers/dri/i965/gen8_fs_generator.cpp
> +++ b/src/mesa/drivers/dri/i965/gen8_fs_generator.cpp
> @@ -1330,11 +1330,11 @@ gen8_fs_generator::generate_assembly(exec_list
> *simd8_instructions,
>
> if (simd16_instructions) {
> /* Align to a 64-byte boundary. */
> - while ((nr_inst * sizeof(gen8_instruction)) % 64)
> + while (next_inst_offset % 64)
> NOP();
>
> /* Save off the start of this SIMD16 program */
> - prog_data->prog_offset_16 = nr_inst * sizeof(gen8_instruction);
> + prog_data->prog_offset_16 = next_inst_offset;
>
> struct annotation *annotation;
> int num_annotations;
>
signature.asc
Description: OpenPGP digital signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
