On Fri, Aug 19, 2016 at 09:55:54AM -0700, Jason Ekstrand wrote:
> ---
> src/mesa/drivers/dri/i965/blorp.c | 66
> -----------------------------
> src/mesa/drivers/dri/i965/genX_blorp_exec.c | 66
> +++++++++++++++++++++++++++++
> src/mesa/drivers/dri/i965/genX_blorp_exec.h | 8 ++--
> 3 files changed, 70 insertions(+), 70 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/blorp.c
> b/src/mesa/drivers/dri/i965/blorp.c
> index dba3441..0688f6b 100644
> --- a/src/mesa/drivers/dri/i965/blorp.c
> +++ b/src/mesa/drivers/dri/i965/blorp.c
> @@ -347,28 +347,6 @@ brw_blorp_compile_nir_shader(struct brw_context *brw,
> struct nir_shader *nir,
> void
> brw_blorp_exec(struct brw_context *brw, const struct brw_blorp_params
> *params)
> {
> - struct gl_context *ctx = &brw->ctx;
> - const uint32_t estimated_max_batch_usage = brw->gen >= 8 ? 1800 : 1500;
> - bool check_aperture_failed_once = false;
> -
> - /* Flush the sampler and render caches. We definitely need to flush the
> - * sampler cache so that we get updated contents from the render cache for
> - * the glBlitFramebuffer() source. Also, we are sometimes warned in the
> - * docs to flush the cache between reinterpretations of the same surface
> - * data with different formats, which blorp does for stencil and depth
> - * data.
> - */
> - brw_emit_mi_flush(brw);
> -
> - brw_select_pipeline(brw, BRW_RENDER_PIPELINE);
> -
> -retry:
> - intel_batchbuffer_require_space(brw, estimated_max_batch_usage,
> RENDER_RING);
> - intel_batchbuffer_save_state(brw);
> - drm_intel_bo *saved_bo = brw->batch.bo;
> - uint32_t saved_used = USED_BATCH(brw->batch);
> - uint32_t saved_state_batch_offset = brw->batch.state_batch_offset;
> -
> switch (brw->gen) {
> case 6:
> gen6_blorp_exec(brw, params);
> @@ -389,50 +367,6 @@ retry:
> /* BLORP is not supported before Gen6. */
> unreachable("not reached");
> }
> -
> - /* Make sure we didn't wrap the batch unintentionally, and make sure we
> - * reserved enough space that a wrap will never happen.
> - */
> - assert(brw->batch.bo == saved_bo);
> - assert((USED_BATCH(brw->batch) - saved_used) * 4 +
> - (saved_state_batch_offset - brw->batch.state_batch_offset) <
> - estimated_max_batch_usage);
> - /* Shut up compiler warnings on release build */
> - (void)saved_bo;
> - (void)saved_used;
> - (void)saved_state_batch_offset;
> -
> - /* Check if the blorp op we just did would make our batch likely to fail
> to
> - * map all the BOs into the GPU at batch exec time later. If so, flush
> the
> - * batch and try again with nothing else in the batch.
> - */
> - if (dri_bufmgr_check_aperture_space(&brw->batch.bo, 1)) {
> - if (!check_aperture_failed_once) {
> - check_aperture_failed_once = true;
> - intel_batchbuffer_reset_to_saved(brw);
> - intel_batchbuffer_flush(brw);
> - goto retry;
> - } else {
> - int ret = intel_batchbuffer_flush(brw);
> - WARN_ONCE(ret == -ENOSPC,
> - "i965: blorp emit exceeded available aperture space\n");
> - }
> - }
> -
> - if (unlikely(brw->always_flush_batch))
> - intel_batchbuffer_flush(brw);
> -
> - /* We've smashed all state compared to what the normal 3D pipeline
> - * rendering tracks for GL.
> - */
> - brw->ctx.NewDriverState |= BRW_NEW_BLORP;
> - brw->no_depth_or_stencil = false;
> - brw->ib.type = -1;
> -
> - /* Flush the sampler cache so any texturing from the destination is
> - * coherent.
> - */
> - brw_emit_mi_flush(brw);
> }
>
> void
> diff --git a/src/mesa/drivers/dri/i965/genX_blorp_exec.c
> b/src/mesa/drivers/dri/i965/genX_blorp_exec.c
> index e07fa0a..9ba1f8a 100644
> --- a/src/mesa/drivers/dri/i965/genX_blorp_exec.c
> +++ b/src/mesa/drivers/dri/i965/genX_blorp_exec.c
> @@ -170,6 +170,28 @@ void
> genX(blorp_exec)(struct brw_context *brw,
> const struct brw_blorp_params *params)
> {
> + struct gl_context *ctx = &brw->ctx;
> + const uint32_t estimated_max_batch_usage = GEN_GEN >= 8 ? 1800 : 1500;
> + bool check_aperture_failed_once = false;
> +
> + /* Flush the sampler and render caches. We definitely need to flush the
> + * sampler cache so that we get updated contents from the render cache for
> + * the glBlitFramebuffer() source. Also, we are sometimes warned in the
> + * docs to flush the cache between reinterpretations of the same surface
> + * data with different formats, which blorp does for stencil and depth
> + * data.
> + */
> + brw_emit_mi_flush(brw);
> +
> + brw_select_pipeline(brw, BRW_RENDER_PIPELINE);
> +
> +retry:
> + intel_batchbuffer_require_space(brw, estimated_max_batch_usage,
> RENDER_RING);
> + intel_batchbuffer_save_state(brw);
> + drm_intel_bo *saved_bo = brw->batch.bo;
> + uint32_t saved_used = USED_BATCH(brw->batch);
> + uint32_t saved_state_batch_offset = brw->batch.state_batch_offset;
> +
> #if GEN_GEN == 6
> /* Emit workaround flushes when we switch from drawing to blorping. */
> brw_emit_post_sync_nonzero_flush(brw);
> @@ -187,4 +209,48 @@ genX(blorp_exec)(struct brw_context *brw,
> brw_emit_depth_stall_flushes(brw);
>
> blorp_exec(&brw->blorp, brw, params);
> +
> + /* Make sure we didn't wrap the batch unintentionally, and make sure we
> + * reserved enough space that a wrap will never happen.
> + */
> + assert(brw->batch.bo == saved_bo);
> + assert((USED_BATCH(brw->batch) - saved_used) * 4 +
> + (saved_state_batch_offset - brw->batch.state_batch_offset) <
> + estimated_max_batch_usage);
> + /* Shut up compiler warnings on release build */
> + (void)saved_bo;
> + (void)saved_used;
> + (void)saved_state_batch_offset;
> +
> + /* Check if the blorp op we just did would make our batch likely to fail
> to
> + * map all the BOs into the GPU at batch exec time later. If so, flush
> the
> + * batch and try again with nothing else in the batch.
> + */
> + if (dri_bufmgr_check_aperture_space(&brw->batch.bo, 1)) {
> + if (!check_aperture_failed_once) {
> + check_aperture_failed_once = true;
> + intel_batchbuffer_reset_to_saved(brw);
> + intel_batchbuffer_flush(brw);
> + goto retry;
> + } else {
> + int ret = intel_batchbuffer_flush(brw);
> + WARN_ONCE(ret == -ENOSPC,
> + "i965: blorp emit exceeded available aperture space\n");
> + }
> + }
> +
> + if (unlikely(brw->always_flush_batch))
> + intel_batchbuffer_flush(brw);
> +
> + /* We've smashed all state compared to what the normal 3D pipeline
> + * rendering tracks for GL.
> + */
> + brw->ctx.NewDriverState |= BRW_NEW_BLORP;
> + brw->no_depth_or_stencil = false;
> + brw->ib.type = -1;
> +
> + /* Flush the sampler cache so any texturing from the destination is
> + * coherent.
> + */
> + brw_emit_mi_flush(brw);
> }
> diff --git a/src/mesa/drivers/dri/i965/genX_blorp_exec.h
> b/src/mesa/drivers/dri/i965/genX_blorp_exec.h
> index 02a0397..f7fbf04 100644
> --- a/src/mesa/drivers/dri/i965/genX_blorp_exec.h
> +++ b/src/mesa/drivers/dri/i965/genX_blorp_exec.h
> @@ -204,7 +204,7 @@ blorp_emit_input_varying_data(struct blorp_batch batch,
> for (unsigned i = 0; i < max_num_varyings; i++) {
> const gl_varying_slot attr = VARYING_SLOT_VAR0 + i;
>
> - if (!(params->wm_prog_data->inputs_read & BITFIELD64_BIT(attr)))
> + if (!(params->wm_prog_data->inputs_read & (1ull << attr)))
This looks to belong to the previous patch (type of inputs_read gets changed
already there).
> continue;
>
> memcpy(inputs, inputs_src + i * 4, vec4_size_in_bytes);
> @@ -391,7 +391,7 @@ blorp_emit_sf_config(struct blorp_batch batch,
> }
>
> blorp_emit(batch, GENX(3DSTATE_SBE), sbe) {
> - sbe.VertexURBEntryReadOffset = BRW_SF_URB_ENTRY_READ_OFFSET;
> + sbe.VertexURBEntryReadOffset = 1;
Dropping BRW_SF_URB_ENTRY_READ_OFFSET here and below don't seem to be directly
related to this patch either.
Otherwise this patch is:
Reviewed-by: Topi Pohjolainen <topi.pohjolai...@intel.com>
> sbe.NumberofSFOutputAttributes = prog_data->num_varying_inputs;
> sbe.VertexURBEntryReadLength = brw_blorp_get_urb_length(prog_data);
> sbe.ForceVertexURBEntryReadLength = true;
> @@ -419,7 +419,7 @@ blorp_emit_sf_config(struct blorp_batch batch,
> }
>
> blorp_emit(batch, GENX(3DSTATE_SBE), sbe) {
> - sbe.VertexURBEntryReadOffset = BRW_SF_URB_ENTRY_READ_OFFSET;
> + sbe.VertexURBEntryReadOffset = 1;
> if (prog_data) {
> sbe.NumberofSFOutputAttributes = prog_data->num_varying_inputs;
> sbe.VertexURBEntryReadLength = brw_blorp_get_urb_length(prog_data);
> @@ -439,7 +439,7 @@ blorp_emit_sf_config(struct blorp_batch batch,
> sf.MultisampleRasterizationMode = params->dst.surf.samples > 1 ?
> MSRASTMODE_ON_PATTERN : MSRASTMODE_OFF_PIXEL;
>
> - sf.VertexURBEntryReadOffset = BRW_SF_URB_ENTRY_READ_OFFSET;
> + sf.VertexURBEntryReadOffset = 1;
> if (prog_data) {
> sf.NumberofSFOutputAttributes = prog_data->num_varying_inputs;
> sf.VertexURBEntryReadLength = brw_blorp_get_urb_length(prog_data);
> --
> 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
