From ef6680d3eee621bbb207ca0eda2e142bcc099ed3 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Marek=20Ol=C5=A1=C3=A1k?= Date: Mon, 23 Sep 2013 02:37:05 +0200 Subject: [PATCH] r600g: move the low-level buffer functions for multiple rings to drivers/radeon Also slightly optimize r600_buffer_map_sync_with_rings. --- src/gallium/drivers/r600/r600_asm.c | 2 +- src/gallium/drivers/r600/r600_blit.c | 2 +- src/gallium/drivers/r600/r600_buffer.c | 6 +- src/gallium/drivers/r600/r600_hw_context.c | 4 +- src/gallium/drivers/r600/r600_pipe.c | 69 +------------------ src/gallium/drivers/r600/r600_pipe.h | 6 -- src/gallium/drivers/r600/r600_query.c | 8 +-- src/gallium/drivers/r600/r600_shader.c | 2 +- src/gallium/drivers/r600/r600_texture.c | 4 +- src/gallium/drivers/radeon/r600_pipe_common.c | 66 ++++++++++++++++++ src/gallium/drivers/radeon/r600_pipe_common.h | 6 ++ 11 files changed, 87 insertions(+), 88 deletions(-) diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c index 581fd0e5e87..3cd14fc8291 100644 --- a/src/gallium/drivers/r600/r600_asm.c +++ b/src/gallium/drivers/r600/r600_asm.c @@ -2415,7 +2415,7 @@ void *r600_create_vertex_fetch_shader(struct pipe_context *ctx, return NULL; } - bytecode = r600_buffer_mmap_sync_with_rings(rctx, shader->buffer, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_UNSYNCHRONIZED); + bytecode = r600_buffer_map_sync_with_rings(&rctx->b, shader->buffer, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_UNSYNCHRONIZED); bytecode += shader->offset / 4; if (R600_BIG_ENDIAN) { diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c index d240c292f12..20faabd4b4e 100644 --- a/src/gallium/drivers/r600/r600_blit.c +++ b/src/gallium/drivers/r600/r600_blit.c @@ -649,7 +649,7 @@ static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *ds /* Flush again in case the 3D engine has been prefetching the resource. */ r600_flag_resource_cache_flush(rctx, dst); } else { - uint32_t *map = r600_buffer_mmap_sync_with_rings(rctx, r600_resource(dst), + uint32_t *map = r600_buffer_map_sync_with_rings(&rctx->b, r600_resource(dst), PIPE_TRANSFER_WRITE); size /= 4; for (unsigned i = 0; i < size; i++) diff --git a/src/gallium/drivers/r600/r600_buffer.c b/src/gallium/drivers/r600/r600_buffer.c index 7fdd2ad9a55..8fe78c3f521 100644 --- a/src/gallium/drivers/r600/r600_buffer.c +++ b/src/gallium/drivers/r600/r600_buffer.c @@ -112,7 +112,7 @@ static void *r600_buffer_transfer_map(struct pipe_context *ctx, assert(usage & PIPE_TRANSFER_WRITE); /* Check if mapping this buffer would cause waiting for the GPU. */ - if (r600_rings_is_buffer_referenced(rctx, rbuffer->cs_buf, RADEON_USAGE_READWRITE) || + if (r600_rings_is_buffer_referenced(&rctx->b, rbuffer->cs_buf, RADEON_USAGE_READWRITE) || rctx->b.ws->buffer_is_busy(rbuffer->buf, RADEON_USAGE_READWRITE)) { unsigned i, mask; @@ -158,7 +158,7 @@ static void *r600_buffer_transfer_map(struct pipe_context *ctx, assert(usage & PIPE_TRANSFER_WRITE); /* Check if mapping this buffer would cause waiting for the GPU. */ - if (r600_rings_is_buffer_referenced(rctx, rbuffer->cs_buf, RADEON_USAGE_READWRITE) || + if (r600_rings_is_buffer_referenced(&rctx->b, rbuffer->cs_buf, RADEON_USAGE_READWRITE) || rctx->b.ws->buffer_is_busy(rbuffer->buf, RADEON_USAGE_READWRITE)) { /* Do a wait-free write-only transfer using a temporary buffer. */ unsigned offset; @@ -176,7 +176,7 @@ static void *r600_buffer_transfer_map(struct pipe_context *ctx, } /* mmap and synchronize with rings */ - data = r600_buffer_mmap_sync_with_rings(rctx, rbuffer, usage); + data = r600_buffer_map_sync_with_rings(&rctx->b, rbuffer, usage); if (!data) { return NULL; } diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c index 3714cd7be37..d8ae30cbdde 100644 --- a/src/gallium/drivers/r600/r600_hw_context.c +++ b/src/gallium/drivers/r600/r600_hw_context.c @@ -75,7 +75,7 @@ void r600_get_backend_mask(struct r600_context *ctx) va = r600_resource_va(&ctx->screen->b.b, (void*)buffer); /* initialize buffer with zeroes */ - results = r600_buffer_mmap_sync_with_rings(ctx, buffer, PIPE_TRANSFER_WRITE); + results = r600_buffer_map_sync_with_rings(&ctx->b, buffer, PIPE_TRANSFER_WRITE); if (results) { memset(results, 0, ctx->max_db * 4 * 4); ctx->b.ws->buffer_unmap(buffer->cs_buf); @@ -90,7 +90,7 @@ void r600_get_backend_mask(struct r600_context *ctx) cs->buf[cs->cdw++] = r600_context_bo_reloc(&ctx->b, &ctx->b.rings.gfx, buffer, RADEON_USAGE_WRITE); /* analyze results */ - results = r600_buffer_mmap_sync_with_rings(ctx, buffer, PIPE_TRANSFER_READ); + results = r600_buffer_map_sync_with_rings(&ctx->b, buffer, PIPE_TRANSFER_READ); if (results) { for(i = 0; i < ctx->max_db; i++) { /* at least highest bit will be set if backend is used */ diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c index 87581b42e24..8ee9487c585 100644 --- a/src/gallium/drivers/r600/r600_pipe.c +++ b/src/gallium/drivers/r600/r600_pipe.c @@ -84,7 +84,7 @@ static struct r600_fence *r600_create_fence(struct r600_context *rctx) R600_ERR("r600: failed to create bo for fence objects\n"); goto out; } - rscreen->fences.data = r600_buffer_mmap_sync_with_rings(rctx, rscreen->fences.bo, PIPE_TRANSFER_READ_WRITE); + rscreen->fences.data = r600_buffer_map_sync_with_rings(&rctx->b, rscreen->fences.bo, PIPE_TRANSFER_READ_WRITE); } if (!LIST_IS_EMPTY(&rscreen->fences.pool)) { @@ -213,73 +213,6 @@ static void r600_flush_dma_ring(void *ctx, unsigned flags) rctx->b.rings.dma.flushing = false; } -boolean r600_rings_is_buffer_referenced(struct r600_context *ctx, - struct radeon_winsys_cs_handle *buf, - enum radeon_bo_usage usage) -{ - if (ctx->b.ws->cs_is_buffer_referenced(ctx->b.rings.gfx.cs, buf, usage)) { - return TRUE; - } - if (ctx->b.rings.dma.cs) { - if (ctx->b.ws->cs_is_buffer_referenced(ctx->b.rings.dma.cs, buf, usage)) { - return TRUE; - } - } - return FALSE; -} - -void *r600_buffer_mmap_sync_with_rings(struct r600_context *ctx, - struct r600_resource *resource, - unsigned usage) -{ - enum radeon_bo_usage rusage = RADEON_USAGE_READWRITE; - unsigned flags = 0; - bool sync_flush = TRUE; - - if (usage & PIPE_TRANSFER_UNSYNCHRONIZED) { - return ctx->b.ws->buffer_map(resource->cs_buf, NULL, usage); - } - - if (!(usage & PIPE_TRANSFER_WRITE)) { - /* have to wait for pending read */ - rusage = RADEON_USAGE_WRITE; - } - if (usage & PIPE_TRANSFER_DONTBLOCK) { - flags |= RADEON_FLUSH_ASYNC; - } - - if (ctx->b.ws->cs_is_buffer_referenced(ctx->b.rings.gfx.cs, resource->cs_buf, rusage) && ctx->b.rings.gfx.cs->cdw) { - ctx->b.rings.gfx.flush(ctx, flags); - if (usage & PIPE_TRANSFER_DONTBLOCK) { - return NULL; - } - } - if (ctx->b.rings.dma.cs) { - if (ctx->b.ws->cs_is_buffer_referenced(ctx->b.rings.dma.cs, resource->cs_buf, rusage) && ctx->b.rings.dma.cs->cdw) { - ctx->b.rings.dma.flush(ctx, flags); - if (usage & PIPE_TRANSFER_DONTBLOCK) { - return NULL; - } - } - } - - if (usage & PIPE_TRANSFER_DONTBLOCK) { - if (ctx->b.ws->buffer_is_busy(resource->buf, rusage)) { - return NULL; - } - } - if (sync_flush) { - /* Try to avoid busy-waiting in radeon_bo_wait. */ - ctx->b.ws->cs_sync_flush(ctx->b.rings.gfx.cs); - if (ctx->b.rings.dma.cs) { - ctx->b.ws->cs_sync_flush(ctx->b.rings.dma.cs); - } - } - - /* at this point everything is synchronized */ - return ctx->b.ws->buffer_map(resource->cs_buf, NULL, usage); -} - static void r600_flush_from_winsys(void *ctx, unsigned flags) { struct r600_context *rctx = (struct r600_context *)ctx; diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h index 80a96ca71d4..dc8274bf54d 100644 --- a/src/gallium/drivers/r600/r600_pipe.h +++ b/src/gallium/drivers/r600/r600_pipe.h @@ -637,12 +637,6 @@ struct pipe_resource *r600_buffer_create(struct pipe_screen *screen, unsigned alignment); /* r600_pipe.c */ -boolean r600_rings_is_buffer_referenced(struct r600_context *ctx, - struct radeon_winsys_cs_handle *buf, - enum radeon_bo_usage usage); -void *r600_buffer_mmap_sync_with_rings(struct r600_context *ctx, - struct r600_resource *resource, - unsigned usage); const char * r600_llvm_gpu_string(enum radeon_family family); diff --git a/src/gallium/drivers/r600/r600_query.c b/src/gallium/drivers/r600/r600_query.c index 457c9ad3903..b9ff6a62155 100644 --- a/src/gallium/drivers/r600/r600_query.c +++ b/src/gallium/drivers/r600/r600_query.c @@ -62,7 +62,7 @@ static struct r600_resource *r600_new_query_buffer(struct r600_context *ctx, uns switch (type) { case PIPE_QUERY_OCCLUSION_COUNTER: case PIPE_QUERY_OCCLUSION_PREDICATE: - results = r600_buffer_mmap_sync_with_rings(ctx, buf, PIPE_TRANSFER_WRITE); + results = r600_buffer_map_sync_with_rings(&ctx->b, buf, PIPE_TRANSFER_WRITE); memset(results, 0, buf_size); /* Set top bits for unused backends. */ @@ -86,7 +86,7 @@ static struct r600_resource *r600_new_query_buffer(struct r600_context *ctx, uns case PIPE_QUERY_SO_STATISTICS: case PIPE_QUERY_SO_OVERFLOW_PREDICATE: case PIPE_QUERY_PIPELINE_STATISTICS: - results = r600_buffer_mmap_sync_with_rings(ctx, buf, PIPE_TRANSFER_WRITE); + results = r600_buffer_map_sync_with_rings(&ctx->b, buf, PIPE_TRANSFER_WRITE); memset(results, 0, buf_size); ctx->b.ws->buffer_unmap(buf->cs_buf); break; @@ -415,7 +415,7 @@ static void r600_begin_query(struct pipe_context *ctx, struct pipe_query *query) } /* Obtain a new buffer if the current one can't be mapped without a stall. */ - if (r600_rings_is_buffer_referenced(rctx, rquery->buffer.buf->cs_buf, RADEON_USAGE_READWRITE) || + if (r600_rings_is_buffer_referenced(&rctx->b, rquery->buffer.buf->cs_buf, RADEON_USAGE_READWRITE) || rctx->b.ws->buffer_is_busy(rquery->buffer.buf->buf, RADEON_USAGE_READWRITE)) { pipe_resource_reference((struct pipe_resource**)&rquery->buffer.buf, NULL); rquery->buffer.buf = r600_new_query_buffer(rctx, rquery->type); @@ -496,7 +496,7 @@ static boolean r600_get_query_buffer_result(struct r600_context *ctx, return TRUE; } - map = r600_buffer_mmap_sync_with_rings(ctx, qbuf->buf, + map = r600_buffer_map_sync_with_rings(&ctx->b, qbuf->buf, PIPE_TRANSFER_READ | (wait ? 0 : PIPE_TRANSFER_DONTBLOCK)); if (!map) diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c index 71818c7f0dc..80cdcd500f6 100644 --- a/src/gallium/drivers/r600/r600_shader.c +++ b/src/gallium/drivers/r600/r600_shader.c @@ -164,7 +164,7 @@ int r600_pipe_shader_create(struct pipe_context *ctx, if (shader->bo == NULL) { return -ENOMEM; } - ptr = r600_buffer_mmap_sync_with_rings(rctx, shader->bo, PIPE_TRANSFER_WRITE); + ptr = r600_buffer_map_sync_with_rings(&rctx->b, shader->bo, PIPE_TRANSFER_WRITE); if (R600_BIG_ENDIAN) { for (i = 0; i < shader->shader.bc.ndw; ++i) { ptr[i] = util_bswap32(shader->shader.bc.bytecode[i]); diff --git a/src/gallium/drivers/r600/r600_texture.c b/src/gallium/drivers/r600/r600_texture.c index 481c6543923..d505d6b6f82 100644 --- a/src/gallium/drivers/r600/r600_texture.c +++ b/src/gallium/drivers/r600/r600_texture.c @@ -791,7 +791,7 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx, /* Use a staging texture for uploads if the underlying BO is busy. */ if (!(usage & PIPE_TRANSFER_READ) && - (r600_rings_is_buffer_referenced(rctx, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) || + (r600_rings_is_buffer_referenced(&rctx->b, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) || rctx->b.ws->buffer_is_busy(rtex->resource.buf, RADEON_USAGE_READWRITE))) { use_staging_texture = TRUE; } @@ -898,7 +898,7 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx, buf = &rtex->resource; } - if (!(map = r600_buffer_mmap_sync_with_rings(rctx, buf, usage))) { + if (!(map = r600_buffer_map_sync_with_rings(&rctx->b, buf, usage))) { pipe_resource_reference((struct pipe_resource**)&trans->staging, NULL); FREE(trans); return NULL; diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c b/src/gallium/drivers/radeon/r600_pipe_common.c index 06484553b02..8f0798a03ec 100644 --- a/src/gallium/drivers/radeon/r600_pipe_common.c +++ b/src/gallium/drivers/radeon/r600_pipe_common.c @@ -271,3 +271,69 @@ void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_re rscreen->aux_context->flush(rscreen->aux_context, NULL, 0); pipe_mutex_unlock(rscreen->aux_context_lock); } + +boolean r600_rings_is_buffer_referenced(struct r600_common_context *ctx, + struct radeon_winsys_cs_handle *buf, + enum radeon_bo_usage usage) +{ + if (ctx->ws->cs_is_buffer_referenced(ctx->rings.gfx.cs, buf, usage)) { + return TRUE; + } + if (ctx->rings.dma.cs && + ctx->ws->cs_is_buffer_referenced(ctx->rings.dma.cs, buf, usage)) { + return TRUE; + } + return FALSE; +} + +void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx, + struct r600_resource *resource, + unsigned usage) +{ + enum radeon_bo_usage rusage = RADEON_USAGE_READWRITE; + + if (usage & PIPE_TRANSFER_UNSYNCHRONIZED) { + return ctx->ws->buffer_map(resource->cs_buf, NULL, usage); + } + + if (!(usage & PIPE_TRANSFER_WRITE)) { + /* have to wait for the last write */ + rusage = RADEON_USAGE_WRITE; + } + + if (ctx->rings.gfx.cs->cdw && + ctx->ws->cs_is_buffer_referenced(ctx->rings.gfx.cs, + resource->cs_buf, rusage)) { + if (usage & PIPE_TRANSFER_DONTBLOCK) { + ctx->rings.gfx.flush(ctx, RADEON_FLUSH_ASYNC); + return NULL; + } else { + ctx->rings.gfx.flush(ctx, 0); + } + } + if (ctx->rings.dma.cs && + ctx->rings.dma.cs->cdw && + ctx->ws->cs_is_buffer_referenced(ctx->rings.dma.cs, + resource->cs_buf, rusage)) { + if (usage & PIPE_TRANSFER_DONTBLOCK) { + ctx->rings.dma.flush(ctx, RADEON_FLUSH_ASYNC); + return NULL; + } else { + ctx->rings.dma.flush(ctx, 0); + } + } + + if (ctx->ws->buffer_is_busy(resource->buf, rusage)) { + if (usage & PIPE_TRANSFER_DONTBLOCK) { + return NULL; + } else { + /* We will be wait for the GPU. Wait for any offloaded + * CS flush to complete to avoid busy-waiting in the winsys. */ + ctx->ws->cs_sync_flush(ctx->rings.gfx.cs); + if (ctx->rings.dma.cs) + ctx->ws->cs_sync_flush(ctx->rings.dma.cs); + } + } + + return ctx->ws->buffer_map(resource->cs_buf, NULL, usage); +} diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h index e5e79b08d61..84fdff1847e 100644 --- a/src/gallium/drivers/radeon/r600_pipe_common.h +++ b/src/gallium/drivers/radeon/r600_pipe_common.h @@ -257,6 +257,12 @@ bool r600_can_dump_shader(struct r600_common_screen *rscreen, const struct tgsi_token *tokens); void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst, unsigned offset, unsigned size, unsigned value); +boolean r600_rings_is_buffer_referenced(struct r600_common_context *ctx, + struct radeon_winsys_cs_handle *buf, + enum radeon_bo_usage usage); +void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx, + struct r600_resource *resource, + unsigned usage); /* r600_streamout.c */ void r600_streamout_buffers_dirty(struct r600_common_context *rctx); -- 2.30.2