r600g,radeonsi: consolidate buffer code, add handling of DISCARD_RANGE for SI

This adds 2 optimizations for radeonsi:
- handling of DISCARD_RANGE
- mapping an uninitialized buffer range is automatically UNSYNCHRONIZED

Reviewed-by: Michel Dänzer <[email protected]>

5 files changed, 5 insertions, 232 deletions

diff --git a/src/gallium/drivers/r600/Makefile.sources b/src/gallium/drivers/r600/Makefile.sources
index 76fd164b6e6..d96d98bc0a1 100644
--- a/src/gallium/drivers/r600/Makefile.sources
+++ b/src/gallium/drivers/r600/Makefile.sources
@@ -1,7 +1,6 @@
 C_SOURCES = \
 	r600_asm.c \
 	r600_blit.c \
-	r600_buffer.c \
 	r600_hw_context.c \
 	r600_isa.c \
 	r600_pipe.c \
diff --git a/src/gallium/drivers/r600/r600_buffer.c b/src/gallium/drivers/r600/r600_buffer.c
deleted file mode 100644
index 969803fe951..00000000000
--- a/src/gallium/drivers/r600/r600_buffer.c
+++ /dev/null
@@ -1,202 +0,0 @@
-/*
- * Copyright 2010 Jerome Glisse <[email protected]>
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * on the rights to use, copy, modify, merge, publish, distribute, sub
- * license, and/or sell copies of the Software, and to permit persons to whom
- * the Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *      Jerome Glisse
- *      Corbin Simpson <[email protected]>
- */
-#include "r600_pipe.h"
-#include "util/u_upload_mgr.h"
-#include "util/u_memory.h"
-#include "util/u_surface.h"
-
-static void r600_buffer_destroy(struct pipe_screen *screen,
-				struct pipe_resource *buf)
-{
-	struct r600_resource *rbuffer = r600_resource(buf);
-
-	util_range_destroy(&rbuffer->valid_buffer_range);
-	pb_reference(&rbuffer->buf, NULL);
-	FREE(rbuffer);
-}
-
-static void *r600_buffer_get_transfer(struct pipe_context *ctx,
-				      struct pipe_resource *resource,
-                                      unsigned level,
-                                      unsigned usage,
-                                      const struct pipe_box *box,
-				      struct pipe_transfer **ptransfer,
-				      void *data, struct r600_resource *staging,
-				      unsigned offset)
-{
-	struct r600_context *rctx = (struct r600_context*)ctx;
-	struct r600_transfer *transfer = util_slab_alloc(&rctx->pool_transfers);
-
-	transfer->transfer.resource = resource;
-	transfer->transfer.level = level;
-	transfer->transfer.usage = usage;
-	transfer->transfer.box = *box;
-	transfer->transfer.stride = 0;
-	transfer->transfer.layer_stride = 0;
-	transfer->offset = offset;
-	transfer->staging = staging;
-	*ptransfer = &transfer->transfer;
-	return data;
-}
-
-static void *r600_buffer_transfer_map(struct pipe_context *ctx,
-					struct pipe_resource *resource,
-					unsigned level,
-					unsigned usage,
-					const struct pipe_box *box,
-					struct pipe_transfer **ptransfer)
-{
-	struct r600_context *rctx = (struct r600_context*)ctx;
-	struct r600_resource *rbuffer = r600_resource(resource);
-	uint8_t *data;
-
-	assert(box->x + box->width <= resource->width0);
-
-	/* See if the buffer range being mapped has never been initialized,
-	 * in which case it can be mapped unsynchronized. */
-	if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
-	    usage & PIPE_TRANSFER_WRITE &&
-	    !util_ranges_intersect(&rbuffer->valid_buffer_range, box->x, box->x + box->width)) {
-		usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
-	}
-
-	if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE &&
-	    !(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
-		assert(usage & PIPE_TRANSFER_WRITE);
-
-		/* Check if mapping this buffer would cause waiting for the GPU. */
-		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)) {
-			rctx->b.invalidate_buffer(&rctx->b.b, &rbuffer->b.b);
-		}
-	}
-	else if ((usage & PIPE_TRANSFER_DISCARD_RANGE) &&
-		 !(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
-		 !(rctx->screen->b.debug_flags & DBG_NO_DISCARD_RANGE) &&
-		 (rctx->screen->b.has_cp_dma ||
-		  (rctx->screen->b.has_streamout &&
-		   /* The buffer range must be aligned to 4 with streamout. */
-		   box->x % 4 == 0 && box->width % 4 == 0))) {
-		assert(usage & PIPE_TRANSFER_WRITE);
-
-		/* Check if mapping this buffer would cause waiting for the GPU. */
-		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;
-			struct r600_resource *staging = NULL;
-
-			u_upload_alloc(rctx->uploader, 0, box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT),
-				       &offset, (struct pipe_resource**)&staging, (void**)&data);
-
-			if (staging) {
-				data += box->x % R600_MAP_BUFFER_ALIGNMENT;
-				return r600_buffer_get_transfer(ctx, resource, level, usage, box,
-								ptransfer, data, staging, offset);
-			}
-		}
-	}
-
-	/* mmap and synchronize with rings */
-	data = r600_buffer_map_sync_with_rings(&rctx->b, rbuffer, usage);
-	if (!data) {
-		return NULL;
-	}
-	data += box->x;
-
-	return r600_buffer_get_transfer(ctx, resource, level, usage, box,
-					ptransfer, data, NULL, 0);
-}
-
-static void r600_buffer_transfer_unmap(struct pipe_context *pipe,
-					struct pipe_transfer *transfer)
-{
-	struct r600_context *rctx = (struct r600_context*)pipe;
-	struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
-	struct r600_resource *rbuffer = r600_resource(transfer->resource);
-
-	if (rtransfer->staging) {
-		struct pipe_resource *dst, *src;
-		unsigned soffset, doffset, size;
-		struct pipe_box box;
-
-		dst = transfer->resource;
-		src = &rtransfer->staging->b.b;
-		size = transfer->box.width;
-		doffset = transfer->box.x;
-		soffset = rtransfer->offset + transfer->box.x % R600_MAP_BUFFER_ALIGNMENT;
-
-		u_box_1d(soffset, size, &box);
-
-		/* Copy the staging buffer into the original one. */
-		if (!(size % 4) && !(doffset % 4) && !(soffset % 4) &&
-		    rctx->b.dma_copy(pipe, dst, 0, doffset, 0, 0, src, 0, &box)) {
-			/* DONE. */
-		} else {
-			pipe->resource_copy_region(pipe, dst, 0, doffset, 0, 0, src, 0, &box);
-		}
-		pipe_resource_reference((struct pipe_resource**)&rtransfer->staging, NULL);
-	}
-
-	if (transfer->usage & PIPE_TRANSFER_WRITE) {
-		util_range_add(&rbuffer->valid_buffer_range, transfer->box.x,
-			       transfer->box.x + transfer->box.width);
-	}
-	util_slab_free(&rctx->pool_transfers, transfer);
-}
-
-static const struct u_resource_vtbl r600_buffer_vtbl =
-{
-	u_default_resource_get_handle,		/* get_handle */
-	r600_buffer_destroy,			/* resource_destroy */
-	r600_buffer_transfer_map,		/* transfer_map */
-	NULL,					/* transfer_flush_region */
-	r600_buffer_transfer_unmap,		/* transfer_unmap */
-	NULL					/* transfer_inline_write */
-};
-
-struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
-					 const struct pipe_resource *templ,
-					 unsigned alignment)
-{
-	struct r600_screen *rscreen = (struct r600_screen*)screen;
-	struct r600_resource *rbuffer;
-
-	rbuffer = MALLOC_STRUCT(r600_resource);
-
-	rbuffer->b.b = *templ;
-	pipe_reference_init(&rbuffer->b.b.reference, 1);
-	rbuffer->b.b.screen = screen;
-	rbuffer->b.vtbl = &r600_buffer_vtbl;
-	util_range_init(&rbuffer->valid_buffer_range);
-
-	if (!r600_init_resource(&rscreen->b, rbuffer, templ->width0, alignment, TRUE, templ->usage)) {
-		FREE(rbuffer);
-		return NULL;
-	}
-	return &rbuffer->b.b;
-}
diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index 296d4660182..4016bbe1520 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -179,13 +179,9 @@ static void r600_destroy_context(struct pipe_context *context)
 	if (rctx->blitter) {
 		util_blitter_destroy(rctx->blitter);
 	}
-	if (rctx->uploader) {
-		u_upload_destroy(rctx->uploader);
-	}
 	if (rctx->allocator_fetch_shader) {
 		u_suballocator_destroy(rctx->allocator_fetch_shader);
 	}
-	util_slab_destroy(&rctx->pool_transfers);
 
 	r600_release_command_buffer(&rctx->start_cs_cmd);
 
@@ -208,10 +204,6 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 	if (rctx == NULL)
 		return NULL;
 
-	util_slab_create(&rctx->pool_transfers,
-			 sizeof(struct r600_transfer), 64,
-			 UTIL_SLAB_SINGLETHREADED);
-
 	rctx->b.b.screen = screen;
 	rctx->b.b.priv = priv;
 	rctx->b.b.destroy = r600_destroy_context;
@@ -295,12 +287,6 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 		rctx->b.rings.dma.flushing = false;
 	}
 
-	rctx->uploader = u_upload_create(&rctx->b.b, 1024 * 1024, 256,
-					PIPE_BIND_INDEX_BUFFER |
-					PIPE_BIND_CONSTANT_BUFFER);
-	if (!rctx->uploader)
-		goto fail;
-
 	rctx->allocator_fetch_shader = u_suballocator_create(&rctx->b.b, 64 * 1024, 256,
 							     0, PIPE_USAGE_STATIC, FALSE);
 	if (!rctx->allocator_fetch_shader)
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 15e89a0e263..735047920ff 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -34,7 +34,6 @@
 #include "r600_resource.h"
 
 #include "util/u_blitter.h"
-#include "util/u_slab.h"
 #include "util/u_suballoc.h"
 #include "util/u_double_list.h"
 #include "util/u_transfer.h"
@@ -63,8 +62,6 @@
 #define R600_BIG_ENDIAN 0
 #endif
 
-#define R600_MAP_BUFFER_ALIGNMENT 64
-
 #define R600_QUERY_DRAW_CALLS		(PIPE_QUERY_DRIVER_SPECIFIC + 0)
 #define R600_QUERY_REQUESTED_VRAM	(PIPE_QUERY_DRIVER_SPECIFIC + 1)
 #define R600_QUERY_REQUESTED_GTT	(PIPE_QUERY_DRIVER_SPECIFIC + 2)
@@ -399,9 +396,7 @@ struct r600_context {
 	struct r600_common_context	b;
 	struct r600_screen		*screen;
 	struct blitter_context		*blitter;
-	struct u_upload_mgr		*uploader;
 	struct u_suballocator		*allocator_fetch_shader;
-	struct util_slab_mempool	pool_transfers;
 	unsigned			initial_gfx_cs_size;
 
 	/* Hardware info. */
@@ -601,11 +596,6 @@ void r600_decompress_depth_textures(struct r600_context *rctx,
 void r600_decompress_color_textures(struct r600_context *rctx,
 				    struct r600_samplerview_state *textures);
 
-/* r600_buffer.c */
-struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
-					 const struct pipe_resource *templ,
-					 unsigned alignment);
-
 /* r600_pipe.c */
 const char * r600_llvm_gpu_string(enum radeon_family family);
 
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 3c7bfe94745..3dc79910247 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -953,10 +953,10 @@ static void r600_set_constant_buffer(struct pipe_context *ctx, uint shader, uint
 				tmpPtr[i] = util_bswap32(((uint32_t *)ptr)[i]);
 			}
 
-			u_upload_data(rctx->uploader, 0, size, tmpPtr, &cb->buffer_offset, &cb->buffer);
+			u_upload_data(rctx->b.uploader, 0, size, tmpPtr, &cb->buffer_offset, &cb->buffer);
 			free(tmpPtr);
 		} else {
-			u_upload_data(rctx->uploader, 0, input->buffer_size, ptr, &cb->buffer_offset, &cb->buffer);
+			u_upload_data(rctx->b.uploader, 0, input->buffer_size, ptr, &cb->buffer_offset, &cb->buffer);
 		}
 		/* account it in gtt */
 		rctx->b.gtt += input->buffer_size;
@@ -1257,7 +1257,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 			unsigned out_offset;
 			void *ptr;
 
-			u_upload_alloc(rctx->uploader, 0, info.count * 2,
+			u_upload_alloc(rctx->b.uploader, 0, info.count * 2,
 				       &out_offset, &out_buffer, &ptr);
 
 			util_shorten_ubyte_elts_to_userptr(
@@ -1276,7 +1276,7 @@ static void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info
 		 * Note: Instanced rendering in combination with immediate indices hangs. */
 		if (ib.user_buffer && (R600_BIG_ENDIAN || info.instance_count > 1 ||
 				       info.count*ib.index_size > 20)) {
-			u_upload_data(rctx->uploader, 0, info.count * ib.index_size,
+			u_upload_data(rctx->b.uploader, 0, info.count * ib.index_size,
 				      ib.user_buffer, &ib.offset, &ib.buffer);
 			ib.user_buffer = NULL;
 		}
@@ -1452,7 +1452,7 @@ void r600_draw_rectangle(struct blitter_context *blitter,
 	/* Upload vertices. The hw rectangle has only 3 vertices,
 	 * I guess the 4th one is derived from the first 3.
 	 * The vertex specification should match u_blitter's vertex element state. */
-	u_upload_alloc(rctx->uploader, 0, sizeof(float) * 24, &offset, &buf, (void**)&vb);
+	u_upload_alloc(rctx->b.uploader, 0, sizeof(float) * 24, &offset, &buf, (void**)&vb);
 	vb[0] = x1;
 	vb[1] = y1;
 	vb[2] = depth;