winsys/amdgpu: add support for 64-bit buffer sizes

v2: fail in radeon_winsys_bo_create if size > 32 bits

Reviewed-by: Nicolai Hähnle <[email protected]>

5 files changed, 30 insertions, 19 deletions

diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
index e92f83a8109..b4ac0db3c50 100644
--- a/src/gallium/auxiliary/util/u_math.h
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -792,6 +792,12 @@ align(int value, int alignment)
    return (value + alignment - 1) & ~(alignment - 1);
 }
 
+static inline uint64_t
+align64(uint64_t value, unsigned alignment)
+{
+   return (value + alignment - 1) & ~(alignment - 1);
+}
+
 /**
  * Works like align but on npot alignments.
  */
diff --git a/src/gallium/drivers/radeon/radeon_winsys.h b/src/gallium/drivers/radeon/radeon_winsys.h
index baecca72383..743dbd1fb46 100644
--- a/src/gallium/drivers/radeon/radeon_winsys.h
+++ b/src/gallium/drivers/radeon/radeon_winsys.h
@@ -449,7 +449,7 @@ struct radeon_winsys {
      * \return          The created buffer object.
      */
     struct pb_buffer *(*buffer_create)(struct radeon_winsys *ws,
-                                       unsigned size,
+                                       uint64_t size,
                                        unsigned alignment,
                                        boolean use_reusable_pool,
                                        enum radeon_bo_domain domain,
@@ -528,7 +528,7 @@ struct radeon_winsys {
      * \param Size      Size in bytes for the new buffer.
      */
     struct pb_buffer *(*buffer_from_ptr)(struct radeon_winsys *ws,
-                                         void *pointer, unsigned size);
+                                         void *pointer, uint64_t size);
 
     /**
      * Whether the buffer was created from a user pointer.
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c b/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
index c79bed45753..04ef17da7bf 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
@@ -36,6 +36,7 @@
 #include <amdgpu_drm.h>
 #include <xf86drm.h>
 #include <stdio.h>
+#include <inttypes.h>
 
 static inline struct amdgpu_winsys_bo *amdgpu_winsys_bo(struct pb_buffer *bo)
 {
@@ -141,9 +142,9 @@ void amdgpu_bo_destroy(struct pb_buffer *_buf)
       amdgpu_fence_reference(&bo->fence[i], NULL);
 
    if (bo->initial_domain & RADEON_DOMAIN_VRAM)
-      bo->ws->allocated_vram -= align(bo->base.size, bo->ws->gart_page_size);
+      bo->ws->allocated_vram -= align64(bo->base.size, bo->ws->gart_page_size);
    else if (bo->initial_domain & RADEON_DOMAIN_GTT)
-      bo->ws->allocated_gtt -= align(bo->base.size, bo->ws->gart_page_size);
+      bo->ws->allocated_gtt -= align64(bo->base.size, bo->ws->gart_page_size);
    FREE(bo);
 }
 
@@ -265,7 +266,7 @@ static void amdgpu_add_buffer_to_global_list(struct amdgpu_winsys_bo *bo)
 }
 
 static struct amdgpu_winsys_bo *amdgpu_create_bo(struct amdgpu_winsys *ws,
-                                                 unsigned size,
+                                                 uint64_t size,
                                                  unsigned alignment,
                                                  unsigned usage,
                                                  enum radeon_bo_domain initial_domain,
@@ -303,9 +304,9 @@ static struct amdgpu_winsys_bo *amdgpu_create_bo(struct amdgpu_winsys *ws,
    r = amdgpu_bo_alloc(ws->dev, &request, &buf_handle);
    if (r) {
       fprintf(stderr, "amdgpu: Failed to allocate a buffer:\n");
-      fprintf(stderr, "amdgpu:    size      : %d bytes\n", size);
-      fprintf(stderr, "amdgpu:    alignment : %d bytes\n", alignment);
-      fprintf(stderr, "amdgpu:    domains   : %d\n", initial_domain);
+      fprintf(stderr, "amdgpu:    size      : %"PRIu64" bytes\n", size);
+      fprintf(stderr, "amdgpu:    alignment : %u bytes\n", alignment);
+      fprintf(stderr, "amdgpu:    domains   : %u\n", initial_domain);
       goto error_bo_alloc;
    }
 
@@ -331,9 +332,9 @@ static struct amdgpu_winsys_bo *amdgpu_create_bo(struct amdgpu_winsys *ws,
    bo->unique_id = __sync_fetch_and_add(&ws->next_bo_unique_id, 1);
 
    if (initial_domain & RADEON_DOMAIN_VRAM)
-      ws->allocated_vram += align(size, ws->gart_page_size);
+      ws->allocated_vram += align64(size, ws->gart_page_size);
    else if (initial_domain & RADEON_DOMAIN_GTT)
-      ws->allocated_gtt += align(size, ws->gart_page_size);
+      ws->allocated_gtt += align64(size, ws->gart_page_size);
 
    amdgpu_add_buffer_to_global_list(bo);
 
@@ -458,7 +459,7 @@ static void amdgpu_buffer_set_metadata(struct pb_buffer *_buf,
 
 static struct pb_buffer *
 amdgpu_bo_create(struct radeon_winsys *rws,
-                 unsigned size,
+                 uint64_t size,
                  unsigned alignment,
                  boolean use_reusable_pool,
                  enum radeon_bo_domain domain,
@@ -482,7 +483,7 @@ amdgpu_bo_create(struct radeon_winsys *rws,
     * BOs. Aligning this here helps the cached bufmgr. Especially small BOs,
     * like constant/uniform buffers, can benefit from better and more reuse.
     */
-   size = align(size, ws->gart_page_size);
+   size = align64(size, ws->gart_page_size);
 
    /* Only set one usage bit each for domains and flags, or the cache manager
     * might consider different sets of domains / flags compatible
@@ -592,9 +593,9 @@ static struct pb_buffer *amdgpu_bo_from_handle(struct radeon_winsys *rws,
       *offset = whandle->offset;
 
    if (bo->initial_domain & RADEON_DOMAIN_VRAM)
-      ws->allocated_vram += align(bo->base.size, ws->gart_page_size);
+      ws->allocated_vram += align64(bo->base.size, ws->gart_page_size);
    else if (bo->initial_domain & RADEON_DOMAIN_GTT)
-      ws->allocated_gtt += align(bo->base.size, ws->gart_page_size);
+      ws->allocated_gtt += align64(bo->base.size, ws->gart_page_size);
 
    amdgpu_add_buffer_to_global_list(bo);
 
@@ -648,7 +649,7 @@ static boolean amdgpu_bo_get_handle(struct pb_buffer *buffer,
 }
 
 static struct pb_buffer *amdgpu_bo_from_ptr(struct radeon_winsys *rws,
-					    void *pointer, unsigned size)
+					    void *pointer, uint64_t size)
 {
     struct amdgpu_winsys *ws = amdgpu_winsys(rws);
     amdgpu_bo_handle buf_handle;
@@ -684,7 +685,7 @@ static struct pb_buffer *amdgpu_bo_from_ptr(struct radeon_winsys *rws,
     bo->initial_domain = RADEON_DOMAIN_GTT;
     bo->unique_id = __sync_fetch_and_add(&ws->next_bo_unique_id, 1);
 
-    ws->allocated_gtt += align(bo->base.size, ws->gart_page_size);
+    ws->allocated_gtt += align64(bo->base.size, ws->gart_page_size);
 
     amdgpu_add_buffer_to_global_list(bo);
 
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_surface.c b/src/gallium/winsys/amdgpu/drm/amdgpu_surface.c
index 4c837a8e20f..1164a3058c5 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_surface.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_surface.c
@@ -212,7 +212,7 @@ static int compute_level(struct amdgpu_winsys *ws,
    }
 
    surf_level = is_stencil ? &surf->stencil_level[level] : &surf->level[level];
-   surf_level->offset = align(surf->bo_size, AddrSurfInfoOut->baseAlign);
+   surf_level->offset = align64(surf->bo_size, AddrSurfInfoOut->baseAlign);
    surf_level->slice_size = AddrSurfInfoOut->sliceSize;
    surf_level->pitch_bytes = AddrSurfInfoOut->pitch * (is_stencil ? 1 : surf->bpe);
    surf_level->npix_x = u_minify(surf->npix_x, level);
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
index c92a66ec5b1..9cc3d0393e3 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
@@ -718,7 +718,7 @@ static void radeon_bo_set_metadata(struct pb_buffer *_buf,
 
 static struct pb_buffer *
 radeon_winsys_bo_create(struct radeon_winsys *rws,
-                        unsigned size,
+                        uint64_t size,
                         unsigned alignment,
                         boolean use_reusable_pool,
                         enum radeon_bo_domain domain,
@@ -728,6 +728,10 @@ radeon_winsys_bo_create(struct radeon_winsys *rws,
     struct radeon_bo *bo;
     unsigned usage = 0;
 
+    /* Only 32-bit sizes are supported. */
+    if (size > UINT_MAX)
+        return NULL;
+
     /* Align size to page size. This is the minimum alignment for normal
      * BOs. Aligning this here helps the cached bufmgr. Especially small BOs,
      * like constant/uniform buffers, can benefit from better and more reuse.
@@ -769,7 +773,7 @@ radeon_winsys_bo_create(struct radeon_winsys *rws,
 }
 
 static struct pb_buffer *radeon_winsys_bo_from_ptr(struct radeon_winsys *rws,
-                                                   void *pointer, unsigned size)
+                                                   void *pointer, uint64_t size)
 {
     struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
     struct drm_radeon_gem_userptr args;