From 185b5a54c94ce11487146042c8eec24909187ed6 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Tue, 18 Jun 2013 13:52:03 -0700 Subject: [PATCH] i965: Avoid flushing the batch for every blorp op. This brings over the batch-wrap-prevention and aperture space checking code from the normal brw_draw.c path, so that we don't need to flush the batch every time. There's a risk here if the intel_emit_post_sync_nonzero_flush() call isn't high enough up in the state emit sequences -- before, we implicitly had one at the batch flush before any state was emitted, so Mesa's workaround emits didn't really matter. Since the SNB fixes by Ken, I didn't see any regressions after 3 piglit runs. Improves cairo-gl performance by 13.7733% +/- 1.74876% (n=30/32) Improves minecraft apitrace performance by 1.03183% +/- 0.482297% (n=90). Reduces low-resolution GLB 2.7 performance by 1.17553% +/- 0.432263% (n=88) Reduces Lightsmark performance by 3.70246% +/- 0.322432% (n=126) No statistically significant performance difference on unigine tropics (n=10) No statistically significant performance difference on openarena (n=755) The two apps that are hurt happen to include stalls on busy buffer objects, so I think this is an effect of missing out on an opportune flush. Reviewed-by: Kenneth Graunke Reviewed-by: Paul Berry --- src/mesa/drivers/dri/i965/brw_blorp.cpp | 50 ++++++++++++++++++++++++ src/mesa/drivers/dri/i965/brw_blorp.h | 4 -- src/mesa/drivers/dri/i965/gen6_blorp.cpp | 12 ------ src/mesa/drivers/dri/i965/gen7_blorp.cpp | 1 - 4 files changed, 50 insertions(+), 17 deletions(-) diff --git a/src/mesa/drivers/dri/i965/brw_blorp.cpp b/src/mesa/drivers/dri/i965/brw_blorp.cpp index 91df346b887..4cbcebe5b32 100644 --- a/src/mesa/drivers/dri/i965/brw_blorp.cpp +++ b/src/mesa/drivers/dri/i965/brw_blorp.cpp @@ -21,6 +21,7 @@ * IN THE SOFTWARE. */ +#include #include "intel_batchbuffer.h" #include "intel_fbo.h" @@ -195,6 +196,26 @@ intel_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt, void brw_blorp_exec(struct brw_context *brw, const brw_blorp_params *params) { + struct gl_context *ctx = &brw->ctx; + uint32_t estimated_max_batch_usage = 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. + */ + intel_batchbuffer_emit_mi_flush(brw); + +retry: + intel_batchbuffer_require_space(brw, estimated_max_batch_usage, false); + intel_batchbuffer_save_state(brw); + drm_intel_bo *saved_bo = brw->batch.bo; + uint32_t saved_used = brw->batch.used; + uint32_t saved_state_batch_offset = brw->batch.state_batch_offset; + switch (brw->gen) { case 6: gen6_blorp_exec(brw, params); @@ -208,6 +229,35 @@ brw_blorp_exec(struct brw_context *brw, const brw_blorp_params *params) break; } + /* 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((brw->batch.used - 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); diff --git a/src/mesa/drivers/dri/i965/brw_blorp.h b/src/mesa/drivers/dri/i965/brw_blorp.h index 07ab80521b1..85bf099b5de 100644 --- a/src/mesa/drivers/dri/i965/brw_blorp.h +++ b/src/mesa/drivers/dri/i965/brw_blorp.h @@ -370,10 +370,6 @@ private: void gen6_blorp_init(struct brw_context *brw); -void -gen6_blorp_emit_batch_head(struct brw_context *brw, - const brw_blorp_params *params); - void gen6_blorp_emit_state_base_address(struct brw_context *brw, const brw_blorp_params *params); diff --git a/src/mesa/drivers/dri/i965/gen6_blorp.cpp b/src/mesa/drivers/dri/i965/gen6_blorp.cpp index 8735270172c..8fb8ca14dae 100644 --- a/src/mesa/drivers/dri/i965/gen6_blorp.cpp +++ b/src/mesa/drivers/dri/i965/gen6_blorp.cpp @@ -45,17 +45,6 @@ * sizeof(float)) /** \} */ -void -gen6_blorp_emit_batch_head(struct brw_context *brw, - const brw_blorp_params *params) -{ - /* To ensure that the batch contains only the resolve, flush the batch - * before beginning and after finishing emitting the resolve packets. - */ - intel_batchbuffer_flush(brw); -} - - /** * CMD_STATE_BASE_ADDRESS * @@ -1045,7 +1034,6 @@ gen6_blorp_exec(struct brw_context *brw, uint32_t wm_bind_bo_offset = 0; uint32_t prog_offset = params->get_wm_prog(brw, &prog_data); - gen6_blorp_emit_batch_head(brw, params); gen6_emit_3dstate_multisample(brw, params->num_samples); gen6_emit_3dstate_sample_mask(brw, params->num_samples, 1.0, false, ~0u); gen6_blorp_emit_state_base_address(brw, params); diff --git a/src/mesa/drivers/dri/i965/gen7_blorp.cpp b/src/mesa/drivers/dri/i965/gen7_blorp.cpp index 71f31b71103..540c46dcb40 100644 --- a/src/mesa/drivers/dri/i965/gen7_blorp.cpp +++ b/src/mesa/drivers/dri/i965/gen7_blorp.cpp @@ -844,7 +844,6 @@ gen7_blorp_exec(struct brw_context *brw, uint32_t sampler_offset = 0; uint32_t prog_offset = params->get_wm_prog(brw, &prog_data); - gen6_blorp_emit_batch_head(brw, params); gen6_emit_3dstate_multisample(brw, params->num_samples); gen6_emit_3dstate_sample_mask(brw, params->num_samples, 1.0, false, ~0u); gen6_blorp_emit_state_base_address(brw, params); -- 2.30.2