On Sat, Oct 22, 2016 at 10:50:32AM -0700, Jason Ekstrand wrote:
> Signed-off-by: Jason Ekstrand <ja...@jlekstrand.net>
> ---
> src/intel/blorp/blorp_genX_exec.h | 33 ++++----------
> src/intel/isl/isl.c | 19 ++++++++
> src/intel/isl/isl.h | 11 +++++
> src/intel/vulkan/anv_batch_chain.c | 4 +-
> src/intel/vulkan/genX_cmd_buffer.c | 8 +---
> src/mesa/drivers/dri/i965/brw_wm_surface_state.c | 56
> +++++++++++-------------
> 6 files changed, 69 insertions(+), 62 deletions(-)
>
> diff --git a/src/intel/blorp/blorp_genX_exec.h
> b/src/intel/blorp/blorp_genX_exec.h
> index ec0d022..d455714 100644
> --- a/src/intel/blorp/blorp_genX_exec.h
> +++ b/src/intel/blorp/blorp_genX_exec.h
> @@ -995,28 +995,13 @@ blorp_emit_depth_stencil_state(struct blorp_batch
> *batch,
> return offset;
> }
>
> -struct surface_state_info {
> - unsigned num_dwords;
> - unsigned ss_align; /* Required alignment of RENDER_SURFACE_STATE in bytes
> */
> - unsigned reloc_dw;
> - unsigned aux_reloc_dw;
> -};
> -
> -static const struct surface_state_info surface_state_infos[] = {
> - [6] = {6, 32, 1, 0},
> - [7] = {8, 32, 1, 6},
> - [8] = {13, 64, 8, 10},
> - [9] = {16, 64, 8, 10},
> -};
> -
> static void
> blorp_emit_surface_state(struct blorp_batch *batch,
> const struct brw_blorp_surface_info *surface,
> - uint32_t *state, uint32_t state_offset,
> + void *state, uint32_t state_offset,
> bool is_render_target)
> {
> - const struct surface_state_info ss_info = surface_state_infos[GEN_GEN];
> -
> + const struct isl_device *isl_dev = batch->blorp->isl_dev;
> struct isl_surf surf = surface->surf;
>
> if (surf.dim == ISL_SURF_DIM_1D &&
> @@ -1038,7 +1023,7 @@ blorp_emit_surface_state(struct blorp_batch *batch,
> .aux_surf = &surface->aux_surf, .aux_usage =
> aux_usage,
> .mocs = mocs, .clear_color = surface->clear_color);
>
> - blorp_surface_reloc(batch, state_offset + ss_info.reloc_dw * 4,
> + blorp_surface_reloc(batch, state_offset + isl_dev->ss.addr_offset,
> surface->addr, 0);
>
> if (aux_usage != ISL_AUX_USAGE_NONE) {
> @@ -1047,8 +1032,9 @@ blorp_emit_surface_state(struct blorp_batch *batch,
> * surface buffer addresses are always 4K page alinged.
> */
> assert((surface->aux_addr.offset & 0xfff) == 0);
> - blorp_surface_reloc(batch, state_offset + ss_info.aux_reloc_dw * 4,
> - surface->aux_addr, state[ss_info.aux_reloc_dw]);
> + uint32_t *aux_addr = state + isl_dev->ss.aux_addr_offset;
Previously 'state' got indexed in dwords (ss_info.aux_reloc_dw). Now this
uses 'isl_dev->ss.aux_addr_offset' instead (which is in number of bytes).
Aren't we offsetting four times too far?
> + blorp_surface_reloc(batch, state_offset + isl_dev->ss.aux_addr_offset,
> + surface->aux_addr, *aux_addr);
> }
> }
>
> @@ -1086,14 +1072,13 @@ static void
> blorp_emit_surface_states(struct blorp_batch *batch,
> const struct blorp_params *params)
> {
> + const struct isl_device *isl_dev = batch->blorp->isl_dev;
> uint32_t bind_offset, surface_offsets[2];
> void *surface_maps[2];
>
> - const unsigned ss_size = GENX(RENDER_SURFACE_STATE_length) * 4;
> - const unsigned ss_align = GENX(RENDER_SURFACE_STATE_length) > 8 ? 64 : 32;
> -
> unsigned num_surfaces = 1 + params->src.enabled;
> - blorp_alloc_binding_table(batch, num_surfaces, ss_size, ss_align,
> + blorp_alloc_binding_table(batch, num_surfaces,
> + isl_dev->ss.size, isl_dev->ss.align,
> &bind_offset, surface_offsets, surface_maps);
>
> if (params->dst.enabled) {
> diff --git a/src/intel/isl/isl.c b/src/intel/isl/isl.c
> index 7831c5e..ec53072 100644
> --- a/src/intel/isl/isl.c
> +++ b/src/intel/isl/isl.c
> @@ -46,6 +46,20 @@ __isl_finishme(const char *file, int line, const char
> *fmt, ...)
> fprintf(stderr, "%s:%d: FINISHME: %s\n", file, line, buf);
> }
>
> +static const struct {
> + uint8_t size;
> + uint8_t align;
> + uint8_t addr_offset;
> + uint8_t aux_addr_offset;
> +} ss_infos[] = {
> + [4] = {24, 32, 4},
> + [5] = {24, 32, 4},
> + [6] = {24, 32, 4},
> + [7] = {32, 32, 4, 24},
> + [8] = {52, 64, 32, 40},
> + [9] = {64, 64, 32, 40},
> +};
> +
> void
> isl_device_init(struct isl_device *dev,
> const struct gen_device_info *info,
> @@ -67,6 +81,11 @@ isl_device_init(struct isl_device *dev,
> assert(info->has_hiz_and_separate_stencil);
> if (info->must_use_separate_stencil)
> assert(ISL_DEV_USE_SEPARATE_STENCIL(dev));
> +
> + dev->ss.size = ss_infos[ISL_DEV_GEN(dev)].size;
> + dev->ss.align = ss_infos[ISL_DEV_GEN(dev)].align;
> + dev->ss.addr_offset = ss_infos[ISL_DEV_GEN(dev)].addr_offset;
> + dev->ss.aux_addr_offset = ss_infos[ISL_DEV_GEN(dev)].aux_addr_offset;
> }
>
> /**
> diff --git a/src/intel/isl/isl.h b/src/intel/isl/isl.h
> index 11ad891..07368f9 100644
> --- a/src/intel/isl/isl.h
> +++ b/src/intel/isl/isl.h
> @@ -671,6 +671,17 @@ struct isl_device {
> const struct gen_device_info *info;
> bool use_separate_stencil;
> bool has_bit6_swizzling;
> +
> + /**
> + * Describes the layout of a RENDER_SURFACE_STATE structure for the
> + * current gen.
> + */
> + struct {
> + uint8_t size;
> + uint8_t align;
> + uint8_t addr_offset;
> + uint8_t aux_addr_offset;
> + } ss;
> };
>
> struct isl_extent2d {
> diff --git a/src/intel/vulkan/anv_batch_chain.c
> b/src/intel/vulkan/anv_batch_chain.c
> index dfa9abf..93425f0 100644
> --- a/src/intel/vulkan/anv_batch_chain.c
> +++ b/src/intel/vulkan/anv_batch_chain.c
> @@ -550,7 +550,9 @@ anv_cmd_buffer_alloc_binding_table(struct anv_cmd_buffer
> *cmd_buffer,
> struct anv_state
> anv_cmd_buffer_alloc_surface_state(struct anv_cmd_buffer *cmd_buffer)
> {
> - return anv_state_stream_alloc(&cmd_buffer->surface_state_stream, 64, 64);
> + struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;
> + return anv_state_stream_alloc(&cmd_buffer->surface_state_stream,
> + isl_dev->ss.size, isl_dev->ss.align);
> }
>
> struct anv_state
> diff --git a/src/intel/vulkan/genX_cmd_buffer.c
> b/src/intel/vulkan/genX_cmd_buffer.c
> index 24e0012..ce53526 100644
> --- a/src/intel/vulkan/genX_cmd_buffer.c
> +++ b/src/intel/vulkan/genX_cmd_buffer.c
> @@ -633,14 +633,10 @@ add_surface_state_reloc(struct anv_cmd_buffer
> *cmd_buffer,
> struct anv_state state, struct anv_bo *bo,
> uint32_t offset)
> {
> - /* The address goes in SURFACE_STATE dword 1 for gens < 8 and dwords 8 and
> - * 9 for gen8+. We only write the first dword for gen8+ here and rely on
> - * the initial state to set the high bits to 0. */
> -
> - const uint32_t dword = GEN_GEN < 8 ? 1 : 8;
> + const struct isl_device *isl_dev = &cmd_buffer->device->isl_dev;
>
> anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
> - state.offset + dword * 4, bo, offset);
> + state.offset + isl_dev->ss.addr_offset, bo, offset);
> }
>
> static struct anv_state
> diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> index b774294..df59541 100644
> --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> @@ -60,22 +60,16 @@ enum {
> INTEL_AUX_BUFFER_DISABLED = 1 << 1,
> };
>
> -struct surface_state_info {
> - unsigned num_dwords;
> - unsigned ss_align; /* Required alignment of RENDER_SURFACE_STATE in bytes
> */
> - unsigned reloc_dw;
> - unsigned aux_reloc_dw;
> - unsigned tex_mocs;
> - unsigned rb_mocs;
> +uint32_t tex_mocs[] = {
> + [7] = GEN7_MOCS_L3,
> + [8] = BDW_MOCS_WB,
> + [9] = SKL_MOCS_WB,
> };
>
> -static const struct surface_state_info surface_state_infos[] = {
> - [4] = {6, 32, 1, 0},
> - [5] = {6, 32, 1, 0},
> - [6] = {6, 32, 1, 0},
> - [7] = {8, 32, 1, 6, GEN7_MOCS_L3, GEN7_MOCS_L3},
> - [8] = {13, 64, 8, 10, BDW_MOCS_WB, BDW_MOCS_PTE},
> - [9] = {16, 64, 8, 10, SKL_MOCS_WB, SKL_MOCS_PTE},
> +uint32_t rb_mocs[] = {
> + [7] = GEN7_MOCS_L3,
> + [8] = BDW_MOCS_PTE,
> + [9] = SKL_MOCS_PTE,
> };
>
> static void
> @@ -85,7 +79,6 @@ brw_emit_surface_state(struct brw_context *brw,
> uint32_t mocs, uint32_t *surf_offset, int surf_index,
> unsigned read_domains, unsigned write_domains)
> {
> - const struct surface_state_info ss_info = surface_state_infos[brw->gen];
> uint32_t tile_x = mt->level[0].slice[0].x_offset;
> uint32_t tile_y = mt->level[0].slice[0].y_offset;
> uint32_t offset = mt->offset;
> @@ -155,11 +148,12 @@ brw_emit_surface_state(struct brw_context *brw,
> clear_color = intel_miptree_get_isl_clear_color(brw, mt);
> }
>
> - uint32_t *dw = __brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
> - ss_info.num_dwords * 4, ss_info.ss_align,
> - surf_index, surf_offset);
> + void *state = __brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
> + brw->isl_dev.ss.size,
> + brw->isl_dev.ss.align,
> + surf_index, surf_offset);
>
> - isl_surf_fill_state(&brw->isl_dev, dw, .surf = &surf, .view = &view,
> + isl_surf_fill_state(&brw->isl_dev, state, .surf = &surf, .view = &view,
> .address = mt->bo->offset64 + offset,
> .aux_surf = aux_surf, .aux_usage = aux_usage,
> .aux_address = aux_offset,
> @@ -167,7 +161,7 @@ brw_emit_surface_state(struct brw_context *brw,
> .x_offset_sa = tile_x, .y_offset_sa = tile_y);
>
> drm_intel_bo_emit_reloc(brw->batch.bo,
> - *surf_offset + 4 * ss_info.reloc_dw,
> + *surf_offset + brw->isl_dev.ss.addr_offset,
> mt->bo, offset,
> read_domains, write_domains);
>
> @@ -179,9 +173,10 @@ brw_emit_surface_state(struct brw_context *brw,
> * an ordinary reloc to do the necessary address translation.
> */
> assert((aux_offset & 0xfff) == 0);
> + uint32_t *aux_addr = state + brw->isl_dev.ss.aux_addr_offset;
Same question here.
> drm_intel_bo_emit_reloc(brw->batch.bo,
> - *surf_offset + 4 * ss_info.aux_reloc_dw,
> - mt->mcs_mt->bo, dw[ss_info.aux_reloc_dw] &
> 0xfff,
> + *surf_offset + brw->isl_dev.ss.aux_addr_offset,
> + mt->mcs_mt->bo, *aux_addr & 0xfff,
> read_domains, write_domains);
> }
> }
> @@ -226,7 +221,7 @@ brw_update_renderbuffer_surface(struct brw_context *brw,
>
> uint32_t offset;
> brw_emit_surface_state(brw, mt, flags, mt->target, view,
> - surface_state_infos[brw->gen].rb_mocs,
> + rb_mocs[brw->gen],
> &offset, surf_index,
> I915_GEM_DOMAIN_RENDER,
> I915_GEM_DOMAIN_RENDER);
> @@ -627,7 +622,7 @@ brw_update_texture_surface(struct gl_context *ctx,
> const int flags =
> brw_disable_aux_surface(brw, mt) ? INTEL_AUX_BUFFER_DISABLED : 0;
> brw_emit_surface_state(brw, mt, flags, mt->target, view,
> - surface_state_infos[brw->gen].tex_mocs,
> + tex_mocs[brw->gen],
> surf_offset, surf_index,
> I915_GEM_DOMAIN_SAMPLER, 0);
> }
> @@ -643,10 +638,9 @@ brw_emit_buffer_surface_state(struct brw_context *brw,
> unsigned pitch,
> bool rw)
> {
> - const struct surface_state_info ss_info = surface_state_infos[brw->gen];
> -
> uint32_t *dw = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
> - ss_info.num_dwords * 4, ss_info.ss_align,
> + brw->isl_dev.ss.size,
> + brw->isl_dev.ss.align,
> out_offset);
>
> isl_buffer_fill_state(&brw->isl_dev, dw,
> @@ -654,11 +648,11 @@ brw_emit_buffer_surface_state(struct brw_context *brw,
> .size = buffer_size,
> .format = surface_format,
> .stride = pitch,
> - .mocs = ss_info.tex_mocs);
> + .mocs = tex_mocs[brw->gen]);
>
> if (bo) {
> drm_intel_bo_emit_reloc(brw->batch.bo,
> - *out_offset + 4 * ss_info.reloc_dw,
> + *out_offset + brw->isl_dev.ss.addr_offset,
> bo, buffer_offset,
> I915_GEM_DOMAIN_SAMPLER,
> (rw ? I915_GEM_DOMAIN_SAMPLER : 0));
> @@ -1202,7 +1196,7 @@ update_renderbuffer_read_surfaces(struct brw_context
> *brw)
> const int flags = brw->draw_aux_buffer_disabled[i] ?
> INTEL_AUX_BUFFER_DISABLED : 0;
> brw_emit_surface_state(brw, irb->mt, flags, target, view,
> - surface_state_infos[brw->gen].tex_mocs,
> + tex_mocs[brw->gen],
> surf_offset, surf_index,
> I915_GEM_DOMAIN_SAMPLER, 0);
>
> @@ -1759,7 +1753,7 @@ update_image_surface(struct brw_context *brw,
> mt->fast_clear_state == INTEL_FAST_CLEAR_STATE_RESOLVED ?
> INTEL_AUX_BUFFER_DISABLED : 0;
> brw_emit_surface_state(brw, mt, flags, mt->target, view,
> - surface_state_infos[brw->gen].tex_mocs,
> + tex_mocs[brw->gen],
> surf_offset, surf_index,
> I915_GEM_DOMAIN_SAMPLER,
> access == GL_READ_ONLY ? 0 :
> --
> 2.5.0.400.gff86faf
>
> _______________________________________________
> 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
