[Mesa-dev] [PATCH v2 1/3] i965: add FS_OPCODE_OVERWRITE_DST

[Ian Romanick]

On 10/09/2013 12:06 AM, Chia-I Wu wrote:
> From: Chia-I Wu <olv at lunarg.com>
> 
> FS_OPCODE_OVERWRITE_DST is used to indicate that the destination register is
> (completely) overwritten.  No code is emitted, but the liveness analysis can
> use it as a hint to add the destination register to DEF bitset.  This is
> needed because it is hard to figure out if some partial writes combined
> constitute a complete write during liveness analysis, while it is easier for
> the FS visitor to know if that is the case.

What changed from v1 of the series?  Just reordering (since some other
patches from the original series landed) and rebasing on current master?

> Signed-off-by: Chia-I Wu <olv at lunarg.com>
> ---
>  src/mesa/drivers/dri/i965/brw_defines.h                 | 1 +
>  src/mesa/drivers/dri/i965/brw_fs_generator.cpp          | 4 ++++
>  src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp     | 5 +++--
>  src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp | 3 ++-
>  src/mesa/drivers/dri/i965/brw_shader.cpp                | 3 +++
>  5 files changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
> index c1e7f31..753a9ec 100644
> --- a/src/mesa/drivers/dri/i965/brw_defines.h
> +++ b/src/mesa/drivers/dri/i965/brw_defines.h
> @@ -793,6 +793,7 @@ enum opcode {
>     FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X,
>     FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y,
>     FS_OPCODE_PLACEHOLDER_HALT,
> +   FS_OPCODE_OVERWRITE_DST,
>  
>     VS_OPCODE_URB_WRITE,
>     VS_OPCODE_SCRATCH_READ,
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> index dbfbc11..4b37784 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_generator.cpp
> @@ -1547,6 +1547,10 @@ fs_generator::generate_code(exec_list *instructions)
>           patch_discard_jumps_to_fb_writes();
>           break;
>  
> +      case FS_OPCODE_OVERWRITE_DST:
> +         /* This is to help liveness analysis. */
> +         break;
> +
>        default:
>  	 if (inst->opcode < (int) ARRAY_SIZE(opcode_descs)) {
>  	    _mesa_problem(ctx, "Unsupported opcode `%s' in FS",
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp b/src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp
> index f5daab2..13891f8 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp
> @@ -77,8 +77,9 @@ fs_live_variables::setup_def_use()
>  	  * variable, and thus qualify for being in def[].
>  	  */
>  	 if (inst->dst.file == GRF &&
> -	     inst->regs_written == v->virtual_grf_sizes[inst->dst.reg] &&
> -	     !inst->is_partial_write()) {
> +	     (inst->opcode == FS_OPCODE_OVERWRITE_DST ||
> +	      (inst->regs_written == v->virtual_grf_sizes[inst->dst.reg] &&
> +	       !inst->is_partial_write()))) {
>  	    int reg = inst->dst.reg;
>              if (!BITSET_TEST(bd[b].use, reg))
>                 BITSET_SET(bd[b].def, reg);
> diff --git a/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp b/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> index 5530683..4e59a10 100644
> --- a/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_schedule_instructions.cpp
> @@ -562,7 +562,8 @@ fs_instruction_scheduler::calculate_deps()
>        schedule_node *n = (schedule_node *)node;
>        fs_inst *inst = (fs_inst *)n->inst;
>  
> -      if (inst->opcode == FS_OPCODE_PLACEHOLDER_HALT)
> +      if (inst->opcode == FS_OPCODE_PLACEHOLDER_HALT ||
> +          inst->opcode == FS_OPCODE_OVERWRITE_DST)
>           add_barrier_deps(n);
>  
>        /* read-after-write deps. */
> diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
> index 61c4bf5..e226c94 100644
> --- a/src/mesa/drivers/dri/i965/brw_shader.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
> @@ -485,6 +485,9 @@ brw_instruction_name(enum opcode op)
>     case FS_OPCODE_PLACEHOLDER_HALT:
>        return "placeholder_halt";
>  
> +   case FS_OPCODE_OVERWRITE_DST:
> +      return "overwrite_dst";
> +
>     case VS_OPCODE_URB_WRITE:
>        return "vs_urb_write";
>     case VS_OPCODE_SCRATCH_READ:
>