r600g: let radeon_winsys maintain the list of relocations

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

7 files changed, 59 insertions, 42 deletions

diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index 043215b3ec7..3ac60bce611 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -239,6 +239,7 @@ struct r600_query {
 
 struct r600_context {
 	struct radeon		*radeon;
+	struct radeon_winsys_cs	*cs;
 	struct r600_range	*range;
 	unsigned		nblocks;
 	struct r600_block	**blocks;
@@ -250,7 +251,7 @@ struct r600_context {
 	unsigned		pm4_dirty_cdwords;
 	unsigned		ctx_pm4_ndwords;
 	unsigned		init_dwords;
-	unsigned		nreloc;
+
 	unsigned		creloc;
 	struct r600_reloc	*reloc;
 	struct radeon_bo	**bo;
diff --git a/src/gallium/winsys/r600/drm/evergreen_hw_context.c b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
index df89047ebc5..3f6f8b5368b 100644
--- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
+++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
@@ -986,14 +986,10 @@ int evergreen_context_init(struct r600_context *ctx, struct radeon *radeon)
 	if (r)
 		goto out_err;
 
+	ctx->cs = radeon->ws->cs_create(radeon->ws);
+
 	/* allocate cs variables */
-	ctx->nreloc = RADEON_CTX_MAX_PM4;
-	ctx->reloc = calloc(ctx->nreloc, sizeof(struct r600_reloc));
-	if (ctx->reloc == NULL) {
-		r = -ENOMEM;
-		goto out_err;
-	}
-	ctx->bo = calloc(ctx->nreloc, sizeof(void *));
+	ctx->bo = calloc(RADEON_CTX_MAX_PM4, sizeof(void *));
 	if (ctx->bo == NULL) {
 		r = -ENOMEM;
 		goto out_err;
@@ -1146,10 +1142,6 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr
 
 	if (draw->indices) {
 		ndwords = 11;
-		/* make sure there is enough relocation space before scheduling draw */
-		if (ctx->creloc >= (ctx->nreloc - 1)) {
-			r600_context_flush(ctx);
-		}
 	}
 
 	/* queries need some special values */
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c
index 4d8bb184cbd..bba55d68267 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -776,9 +776,9 @@ void r600_context_fini(struct r600_context *ctx)
 	r600_free_resource_range(ctx, &ctx->fs_resources, ctx->num_fs_resources);
 	free(ctx->range);
 	free(ctx->blocks);
-	free(ctx->reloc);
 	free(ctx->bo);
 	free(ctx->pm4);
+	ctx->radeon->ws->cs_destroy(ctx->cs);
 
 	memset(ctx, 0, sizeof(struct r600_context));
 }
@@ -912,14 +912,10 @@ int r600_context_init(struct r600_context *ctx, struct radeon *radeon)
 	if (r)
 		goto out_err;
 
+	ctx->cs = radeon->ws->cs_create(radeon->ws);
+
 	/* allocate cs variables */
-	ctx->nreloc = RADEON_CTX_MAX_PM4;
-	ctx->reloc = calloc(ctx->nreloc, sizeof(struct r600_reloc));
-	if (ctx->reloc == NULL) {
-		r = -ENOMEM;
-		goto out_err;
-	}
-	ctx->bo = calloc(ctx->nreloc, sizeof(void *));
+	ctx->bo = calloc(RADEON_CTX_MAX_PM4, sizeof(void *));
 	if (ctx->bo == NULL) {
 		r = -ENOMEM;
 		goto out_err;
@@ -1009,14 +1005,15 @@ void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags,
 void r600_context_get_reloc(struct r600_context *ctx, struct r600_bo *rbo)
 {
 	struct radeon_bo *bo = rbo->bo;
-	bo->reloc = &ctx->reloc[ctx->creloc];
-	bo->reloc_id = ctx->creloc * sizeof(struct r600_reloc) / 4;
-	ctx->reloc[ctx->creloc].handle = bo->handle;
-	ctx->reloc[ctx->creloc].read_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM);
-	ctx->reloc[ctx->creloc].write_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM);
-	ctx->reloc[ctx->creloc].flags = 0;
-	radeon_bo_reference(ctx->radeon, &ctx->bo[ctx->creloc], bo);
-	ctx->creloc++;
+
+	unsigned reloc_index = ctx->radeon->ws->trans_add_reloc(ctx->cs, bo->cs_buf,
+								rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM),
+								rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM),
+								(void**)&ctx->reloc, &ctx->creloc);
+
+	bo->reloc = (void*)1;
+	bo->reloc_id = reloc_index * 4;
+	radeon_bo_reference(ctx->radeon, &ctx->bo[reloc_index], bo);
 }
 
 void r600_context_reg(struct r600_context *ctx,
@@ -1444,10 +1441,6 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
 
 	if (draw->indices) {
 		ndwords = 11;
-		/* make sure there is enough relocation space before scheduling draw */
-		if (ctx->creloc >= (ctx->nreloc - 1)) {
-			r600_context_flush(ctx);
-		}
 	}
 
 	/* queries need some special values */
@@ -1570,6 +1563,7 @@ void r600_context_flush(struct r600_context *ctx)
 	ctx->pm4_dirty_cdwords = 0;
 	ctx->pm4_cdwords = 0;
 	ctx->flags = 0;
+	ctx->radeon->ws->cs_flush(ctx->cs, 0);
 
 	r600_init_cs(ctx);
 
@@ -1601,8 +1595,7 @@ void r600_context_emit_fence(struct r600_context *ctx, struct r600_bo *fence_bo,
 {
 	unsigned ndwords = 10;
 
-	if (((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) ||
-	    (ctx->creloc >= (ctx->nreloc - 1))) {
+	if ((ctx->pm4_dirty_cdwords + ndwords + ctx->pm4_cdwords) > ctx->pm4_ndwords) {
 		/* need to flush */
 		r600_context_flush(ctx);
 	}
diff --git a/src/gallium/winsys/r600/drm/r600_priv.h b/src/gallium/winsys/r600/drm/r600_priv.h
index 49d3060bbf2..baf7c98d578 100644
--- a/src/gallium/winsys/r600/drm/r600_priv.h
+++ b/src/gallium/winsys/r600/drm/r600_priv.h
@@ -65,6 +65,7 @@ struct r600_reg {
 struct radeon_bo {
 	struct pipe_reference		reference;
 	struct pb_buffer		*buf;
+	struct radeon_winsys_cs_handle	*cs_buf;
 	unsigned			handle;
 	unsigned			size;
 	int				map_count;
diff --git a/src/gallium/winsys/r600/drm/radeon_bo.c b/src/gallium/winsys/r600/drm/radeon_bo.c
index 63dc44ddb44..536bbe45bff 100644
--- a/src/gallium/winsys/r600/drm/radeon_bo.c
+++ b/src/gallium/winsys/r600/drm/radeon_bo.c
@@ -94,6 +94,7 @@ struct radeon_bo *radeon_bo(struct radeon *radeon, unsigned handle,
 		FREE(bo);
 		return NULL;
 	}
+	bo->cs_buf = radeon->ws->buffer_get_cs_handle(bo->buf);
 	bo->handle = radeon->ws->trans_get_buffer_handle(bo->buf);
 	bo->size = size;
 	return bo;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index 1b30b95a318..9a1e16957d2 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -219,11 +219,11 @@ int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo)
     return -1;
 }
 
-static void radeon_add_reloc(struct radeon_cs_context *csc,
-                             struct radeon_bo *bo,
-                             enum radeon_bo_domain rd,
-                             enum radeon_bo_domain wd,
-                             enum radeon_bo_domain *added_domains)
+static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
+                                 struct radeon_bo *bo,
+                                 enum radeon_bo_domain rd,
+                                 enum radeon_bo_domain wd,
+                                 enum radeon_bo_domain *added_domains)
 {
     struct drm_radeon_cs_reloc *reloc;
     unsigned i;
@@ -233,7 +233,7 @@ static void radeon_add_reloc(struct radeon_cs_context *csc,
         reloc = csc->relocs_hashlist[hash];
         if (reloc->handle == bo->handle) {
             update_domains(reloc, rd, wd, added_domains);
-            return;
+            return csc->reloc_indices_hashlist[hash];
         }
 
         /* Hash collision, look for the BO in the list of relocs linearly. */
@@ -246,7 +246,7 @@ static void radeon_add_reloc(struct radeon_cs_context *csc,
                 csc->relocs_hashlist[hash] = reloc;
                 csc->reloc_indices_hashlist[hash] = i;
                 /*printf("write_reloc collision, hash: %i, handle: %i\n", hash, bo->handle);*/
-                return;
+                return i;
             }
         }
     }
@@ -280,9 +280,9 @@ static void radeon_add_reloc(struct radeon_cs_context *csc,
     csc->reloc_indices_hashlist[hash] = csc->crelocs;
 
     csc->chunks[1].length_dw += RELOC_DWORDS;
-    csc->crelocs++;
 
     *added_domains = rd | wd;
+    return csc->crelocs++;
 }
 
 static void radeon_drm_cs_add_reloc(struct radeon_winsys_cs *rcs,
@@ -470,6 +470,28 @@ static boolean radeon_bo_is_referenced(struct radeon_winsys_cs *rcs,
     return radeon_bo_is_referenced_by_cs(cs, bo);
 }
 
+static unsigned trans_add_reloc(struct radeon_winsys_cs *rcs,
+				struct radeon_winsys_cs_handle *buf,
+				enum radeon_bo_domain rd,
+                                enum radeon_bo_domain wd,
+				void **reloc_list, unsigned *reloc_count)
+{
+	struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
+        struct radeon_bo *bo = (struct radeon_bo*)buf;
+        enum radeon_bo_domain added_domains;
+
+        unsigned index = radeon_add_reloc(cs->csc, bo, rd, wd, &added_domains);
+
+        if (added_domains & RADEON_DOMAIN_GTT)
+            cs->csc->used_gart += bo->size;
+        if (added_domains & RADEON_DOMAIN_VRAM)
+            cs->csc->used_vram += bo->size;
+
+	*reloc_list = cs->csc->relocs;
+	*reloc_count = cs->csc->crelocs;
+	return index;
+}
+
 void radeon_drm_cs_init_functions(struct radeon_drm_winsys *ws)
 {
     ws->base.cs_create = radeon_drm_cs_create;
@@ -480,4 +502,6 @@ void radeon_drm_cs_init_functions(struct radeon_drm_winsys *ws)
     ws->base.cs_flush = radeon_drm_cs_flush;
     ws->base.cs_set_flush = radeon_drm_cs_set_flush;
     ws->base.cs_is_buffer_referenced = radeon_bo_is_referenced;
+
+    ws->base.trans_add_reloc = trans_add_reloc;
 }
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index 2e5000702e0..8e81fa1e301 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -329,6 +329,11 @@ struct radeon_winsys {
 
     /* Transitional functions for r600g when moving to winsys/radeon */
     unsigned (*trans_get_buffer_handle)(struct pb_buffer *buf);
+    unsigned (*trans_add_reloc)(struct radeon_winsys_cs *cs,
+				struct radeon_winsys_cs_handle *buf,
+				enum radeon_bo_domain rd,
+                                enum radeon_bo_domain wd,
+				void **reloc_list, unsigned *reloc_count);
 };
 
 #endif