On Sun, May 29, 2016 at 3:38 PM, Jordan Justen <jordan.l.jus...@intel.com>
wrote:
> We need information about push constants in a few places for the GL
> driver, and another couple places for the vulkan driver.
>
> When we add support for uploading both a common (cross-thread) set of
> push constants, combined with the previous per-thread push constant
> data, things are going to get even more complicated. To simplify
> things, we add push constant info into the cs prog_data struct.
>
> The cross-thread constant support is added as of Haswell. To support
> it we need to make sure all push constants with uniform values are
> added to earlier registers. The register that varies per thread and
> holds the thread invocation's unique local ID needs to be added last.
>
> Signed-off-by: Jordan Justen <jordan.l.jus...@intel.com>
> ---
> src/mesa/drivers/dri/i965/brw_compiler.h | 12 +++++++
> src/mesa/drivers/dri/i965/brw_fs.cpp | 58
> ++++++++++++++++++++++++++++++++
> 2 files changed, 70 insertions(+)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_compiler.h
> b/src/mesa/drivers/dri/i965/brw_compiler.h
> index f1f9e56..dda6297 100644
> --- a/src/mesa/drivers/dri/i965/brw_compiler.h
> +++ b/src/mesa/drivers/dri/i965/brw_compiler.h
> @@ -424,6 +424,12 @@ struct brw_wm_prog_data {
> int urb_setup[VARYING_SLOT_MAX];
> };
>
> +struct brw_push_const_block {
> + unsigned dwords; /* Dword count, not reg aligned */
> + unsigned regs;
> + unsigned size; /* Bytes, register aligned */
> +};
> +
> struct brw_cs_prog_data {
> struct brw_stage_prog_data base;
>
> @@ -437,6 +443,12 @@ struct brw_cs_prog_data {
> int thread_local_id_index;
>
> struct {
> + struct brw_push_const_block cross_thread;
> + struct brw_push_const_block per_thread;
> + struct brw_push_const_block total;
> + } push;
> +
> + struct {
> /** @{
> * surface indices the CS-specific surfaces
> */
> diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp
> b/src/mesa/drivers/dri/i965/brw_fs.cpp
> index 836ade0..bd37fbd 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
> @@ -6479,6 +6479,61 @@ fs_visitor::emit_cs_work_group_id_setup()
> }
>
> static void
> +fill_push_const_block_info(struct brw_push_const_block *block, unsigned
> dwords)
> +{
> + block->dwords = dwords;
> + block->regs = DIV_ROUND_UP(dwords, 8);
> + block->size = block->regs * 32;
> +}
> +
> +static void
> +cs_fill_push_const_info(const struct brw_device_info *devinfo,
> + struct brw_cs_prog_data *cs_prog_data)
> +{
> + const struct brw_stage_prog_data *prog_data =
> + (struct brw_stage_prog_data*) cs_prog_data;
> + bool fill_thread_id =
> + cs_prog_data->thread_local_id_index >= 0 &&
> + cs_prog_data->thread_local_id_index < (int)prog_data->nr_params;
> + bool cross_thread_supported = devinfo->gen > 7 || devinfo->is_haswell;
> +
> + /* The thread ID should be stored in the last param dword */
> + assert(prog_data->nr_params > 0 || !fill_thread_id);
> + assert(!fill_thread_id ||
> + cs_prog_data->thread_local_id_index ==
> + (int)prog_data->nr_params - 1);
> +
> + unsigned cross_thread_dwords, per_thread_dwords;
> + if (cross_thread_supported && fill_thread_id) {
> + /* Fill all but the last register with cross-thread payload */
> + cross_thread_dwords = 8 * (cs_prog_data->thread_local_id_index / 8);
> + per_thread_dwords = prog_data->nr_params - cross_thread_dwords;
> + assert(per_thread_dwords > 0 && per_thread_dwords <= 8);
>
If I understand you correctly here, you're putting the bottom registers
(aligned down to 8) into the cross-thread space and putting whatever is
left including the thread_local_id into the per-thread space. Seems
reasonable. I probably would have been more lazy and burned a whole
register on the local_id. :-)
> + } else if (cross_thread_supported && !fill_thread_id) {
> + /* Fill all data using cross-thread payload */
> + cross_thread_dwords = prog_data->nr_params;
> + per_thread_dwords = 0u;
> + } else {
> + cross_thread_dwords = 0u;
> + per_thread_dwords = prog_data->nr_params;
>
Mind putting the !cross_thread_supported case first?
> + }
> +
> + fill_push_const_block_info(&cs_prog_data->push.cross_thread,
> cross_thread_dwords);
> + fill_push_const_block_info(&cs_prog_data->push.per_thread,
> per_thread_dwords);
> +
> + unsigned total_dwords =
> + (cs_prog_data->push.per_thread.size * cs_prog_data->threads +
> + cs_prog_data->push.cross_thread.size) / 4;
> + fill_push_const_block_info(&cs_prog_data->push.total, total_dwords);
> +
> + assert(cs_prog_data->push.cross_thread.dwords % 8 == 0 ||
> + cs_prog_data->push.per_thread.size == 0);
> + assert(cs_prog_data->push.cross_thread.dwords +
> + cs_prog_data->push.per_thread.dwords ==
> + prog_data->nr_params);
> +}
> +
> +static void
> cs_set_simd_size(struct brw_cs_prog_data *cs_prog_data, unsigned size)
> {
> cs_prog_data->simd_size = size;
> @@ -6536,6 +6591,7 @@ brw_compile_cs(const struct brw_compiler *compiler,
> void *log_data,
> } else {
> cfg = v8.cfg;
> cs_set_simd_size(prog_data, 8);
> + cs_fill_push_const_info(compiler->devinfo, prog_data);
> prog_data->base.dispatch_grf_start_reg = v8.payload.num_regs;
> }
> }
> @@ -6561,6 +6617,7 @@ brw_compile_cs(const struct brw_compiler *compiler,
> void *log_data,
> } else {
> cfg = v16.cfg;
> cs_set_simd_size(prog_data, 16);
> + cs_fill_push_const_info(compiler->devinfo, prog_data);
> prog_data->dispatch_grf_start_reg_16 = v16.payload.num_regs;
> }
> }
> @@ -6588,6 +6645,7 @@ brw_compile_cs(const struct brw_compiler *compiler,
> void *log_data,
> } else {
> cfg = v32.cfg;
> cs_set_simd_size(prog_data, 32);
> + cs_fill_push_const_info(compiler->devinfo, prog_data);
> }
> }
>
> --
> 2.8.1
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
