diff options
| -rw-r--r-- | lib/mesa/src/gallium/drivers/v3d/v3d_tiling.c | 125 | ||||
| -rw-r--r-- | lib/mesa/src/gallium/drivers/v3d/v3d_tiling.h | 2 |
2 files changed, 12 insertions, 115 deletions
diff --git a/lib/mesa/src/gallium/drivers/v3d/v3d_tiling.c b/lib/mesa/src/gallium/drivers/v3d/v3d_tiling.c index 95cd0dff6..bc227be23 100644 --- a/lib/mesa/src/gallium/drivers/v3d/v3d_tiling.c +++ b/lib/mesa/src/gallium/drivers/v3d/v3d_tiling.c @@ -31,7 +31,6 @@ #include "v3d_screen.h" #include "v3d_context.h" #include "v3d_tiling.h" -#include "broadcom/common/v3d_cpu_tiling.h" /** Return the width in pixels of a 64-byte microtile. */ uint32_t @@ -79,8 +78,9 @@ static inline uint32_t v3d_get_utile_pixel_offset(uint32_t cpp, uint32_t x, uint32_t y) { uint32_t utile_w = v3d_utile_width(cpp); + uint32_t utile_h = v3d_utile_height(cpp); - assert(x < utile_w && y < v3d_utile_height(cpp)); + assert(x < utile_w && y < utile_h); return x * cpp + y * utile_w * cpp; } @@ -211,19 +211,15 @@ v3d_get_uif_no_xor_pixel_offset(uint32_t cpp, uint32_t image_h, return v3d_get_uif_pixel_offset(cpp, image_h, x, y, false); } -/* Loads/stores non-utile-aligned boxes by walking over the destination - * rectangle, computing the address on the GPU, and storing/loading a pixel at - * a time. - */ static inline void -v3d_move_pixels_unaligned(void *gpu, uint32_t gpu_stride, - void *cpu, uint32_t cpu_stride, - int cpp, uint32_t image_h, - const struct pipe_box *box, - uint32_t (*get_pixel_offset)(uint32_t cpp, - uint32_t image_h, - uint32_t x, uint32_t y), - bool is_load) +v3d_move_pixels_general_percpp(void *gpu, uint32_t gpu_stride, + void *cpu, uint32_t cpu_stride, + int cpp, uint32_t image_h, + const struct pipe_box *box, + uint32_t (*get_pixel_offset)(uint32_t cpp, + uint32_t image_h, + uint32_t x, uint32_t y), + bool is_load) { for (uint32_t y = 0; y < box->height; y++) { void *cpu_row = cpu + y * cpu_stride; @@ -252,107 +248,6 @@ v3d_move_pixels_unaligned(void *gpu, uint32_t gpu_stride, } } -/* Breaks the image down into utiles and calls either the fast whole-utile - * load/store functions, or the unaligned fallback case. - */ -static inline void -v3d_move_pixels_general_percpp(void *gpu, uint32_t gpu_stride, - void *cpu, uint32_t cpu_stride, - int cpp, uint32_t image_h, - const struct pipe_box *box, - uint32_t (*get_pixel_offset)(uint32_t cpp, - uint32_t image_h, - uint32_t x, uint32_t y), - bool is_load) -{ - uint32_t utile_w = v3d_utile_width(cpp); - uint32_t utile_h = v3d_utile_height(cpp); - uint32_t utile_gpu_stride = utile_w * cpp; - uint32_t x1 = box->x; - uint32_t y1 = box->y; - uint32_t x2 = box->x + box->width; - uint32_t y2 = box->y + box->height; - uint32_t align_x1 = align(x1, utile_w); - uint32_t align_y1 = align(y1, utile_h); - uint32_t align_x2 = x2 & ~(utile_w - 1); - uint32_t align_y2 = y2 & ~(utile_h - 1); - - /* Load/store all the whole utiles first. */ - for (uint32_t y = align_y1; y < align_y2; y += utile_h) { - void *cpu_row = cpu + (y - box->y) * cpu_stride; - - for (uint32_t x = align_x1; x < align_x2; x += utile_w) { - void *utile_gpu = (gpu + - get_pixel_offset(cpp, image_h, x, y)); - void *utile_cpu = cpu_row + (x - box->x) * cpp; - - if (is_load) { - v3d_load_utile(utile_cpu, cpu_stride, - utile_gpu, utile_gpu_stride); - } else { - v3d_store_utile(utile_gpu, utile_gpu_stride, - utile_cpu, cpu_stride); - } - } - } - - /* If there were no aligned utiles in the middle, load/store the whole - * thing unaligned. - */ - if (align_y2 <= align_y1 || - align_x2 <= align_x1) { - v3d_move_pixels_unaligned(gpu, gpu_stride, - cpu, cpu_stride, - cpp, image_h, - box, - get_pixel_offset, is_load); - return; - } - - /* Load/store the partial utiles. */ - struct pipe_box partial_boxes[4] = { - /* Top */ - { - .x = x1, - .width = x2 - x1, - .y = y1, - .height = align_y1 - y1, - }, - /* Bottom */ - { - .x = x1, - .width = x2 - x1, - .y = align_y2, - .height = y2 - align_y2, - }, - /* Left */ - { - .x = x1, - .width = align_x1 - x1, - .y = align_y1, - .height = align_y2 - align_y1, - }, - /* Right */ - { - .x = align_x2, - .width = x2 - align_x2, - .y = align_y1, - .height = align_y2 - align_y1, - }, - }; - for (int i = 0; i < ARRAY_SIZE(partial_boxes); i++) { - void *partial_cpu = (cpu + - (partial_boxes[i].y - y1) * cpu_stride + - (partial_boxes[i].x - x1) * cpp); - - v3d_move_pixels_unaligned(gpu, gpu_stride, - partial_cpu, cpu_stride, - cpp, image_h, - &partial_boxes[i], - get_pixel_offset, is_load); - } -} - static inline void v3d_move_pixels_general(void *gpu, uint32_t gpu_stride, void *cpu, uint32_t cpu_stride, diff --git a/lib/mesa/src/gallium/drivers/v3d/v3d_tiling.h b/lib/mesa/src/gallium/drivers/v3d/v3d_tiling.h index d04349108..79577445e 100644 --- a/lib/mesa/src/gallium/drivers/v3d/v3d_tiling.h +++ b/lib/mesa/src/gallium/drivers/v3d/v3d_tiling.h @@ -27,6 +27,8 @@ uint32_t v3d_utile_width(int cpp) ATTRIBUTE_CONST; uint32_t v3d_utile_height(int cpp) ATTRIBUTE_CONST; bool v3d_size_is_lt(uint32_t width, uint32_t height, int cpp) ATTRIBUTE_CONST; +void v3d_load_utile(void *dst, void *src, uint32_t dst_stride, uint32_t cpp); +void v3d_store_utile(void *dst, void *src, uint32_t src_stride, uint32_t cpp); void v3d_load_tiled_image(void *dst, uint32_t dst_stride, void *src, uint32_t src_stride, enum v3d_tiling_mode tiling_format, int cpp, |