drm/vc4: Add exec flags to allow forcing a specific X/Y tile walk order.

[uclinux-h8/linux.git] / drivers / gpu / drm / vc4 / vc4_render_cl.c

diff --git a/drivers/gpu/drm/vc4/vc4_render_cl.c b/drivers/gpu/drm/vc4/vc4_render_cl.c
index 5dc1942..273984f 100644 (file)
--- a/drivers/gpu/drm/vc4/vc4_render_cl.c
+++ b/drivers/gpu/drm/vc4/vc4_render_cl.c
@@ -261,8 +261,17 @@ static int vc4_create_rcl_bo(struct drm_device *dev, struct vc4_exec_info *exec,
        uint8_t max_y_tile = args->max_y_tile;
        uint8_t xtiles = max_x_tile - min_x_tile + 1;
        uint8_t ytiles = max_y_tile - min_y_tile + 1;
-       uint8_t x, y;
+       uint8_t xi, yi;
        uint32_t size, loop_body_size;
+       bool positive_x = true;
+       bool positive_y = true;
+
+       if (args->flags & VC4_SUBMIT_CL_FIXED_RCL_ORDER) {
+               if (!(args->flags & VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X))
+                       positive_x = false;
+               if (!(args->flags & VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y))
+                       positive_y = false;
+       }
 
        size = VC4_PACKET_TILE_RENDERING_MODE_CONFIG_SIZE;
        loop_body_size = VC4_PACKET_TILE_COORDINATES_SIZE;
@@ -320,7 +329,7 @@ static int vc4_create_rcl_bo(struct drm_device *dev, struct vc4_exec_info *exec,
 
        size += xtiles * ytiles * loop_body_size;
 
-       setup->rcl = &vc4_bo_create(dev, size, true)->base;
+       setup->rcl = &vc4_bo_create(dev, size, true, VC4_BO_TYPE_RCL)->base;
        if (IS_ERR(setup->rcl))
                return PTR_ERR(setup->rcl);
        list_add_tail(&to_vc4_bo(&setup->rcl->base)->unref_head,
@@ -354,10 +363,12 @@ static int vc4_create_rcl_bo(struct drm_device *dev, struct vc4_exec_info *exec,
        rcl_u16(setup, args->height);
        rcl_u16(setup, args->color_write.bits);
 
-       for (y = min_y_tile; y <= max_y_tile; y++) {
-               for (x = min_x_tile; x <= max_x_tile; x++) {
-                       bool first = (x == min_x_tile && y == min_y_tile);
-                       bool last = (x == max_x_tile && y == max_y_tile);
+       for (yi = 0; yi < ytiles; yi++) {
+               int y = positive_y ? min_y_tile + yi : max_y_tile - yi;
+               for (xi = 0; xi < xtiles; xi++) {
+                       int x = positive_x ? min_x_tile + xi : max_x_tile - xi;
+                       bool first = (xi == 0 && yi == 0);
+                       bool last = (xi == xtiles - 1 && yi == ytiles - 1);
 
                        emit_tile(exec, setup, x, y, first, last);
                }
@@ -378,14 +389,14 @@ static int vc4_full_res_bounds_check(struct vc4_exec_info *exec,
        u32 render_tiles_stride = DIV_ROUND_UP(exec->args->width, 32);
 
        if (surf->offset > obj->base.size) {
-               DRM_ERROR("surface offset %d > BO size %zd\n",
+               DRM_DEBUG("surface offset %d > BO size %zd\n",
                          surf->offset, obj->base.size);
                return -EINVAL;
        }
 
        if ((obj->base.size - surf->offset) / VC4_TILE_BUFFER_SIZE <
            render_tiles_stride * args->max_y_tile + args->max_x_tile) {
-               DRM_ERROR("MSAA tile %d, %d out of bounds "
+               DRM_DEBUG("MSAA tile %d, %d out of bounds "
                          "(bo size %zd, offset %d).\n",
                          args->max_x_tile, args->max_y_tile,
                          obj->base.size,
@@ -401,7 +412,7 @@ static int vc4_rcl_msaa_surface_setup(struct vc4_exec_info *exec,
                                      struct drm_vc4_submit_rcl_surface *surf)
 {
        if (surf->flags != 0 || surf->bits != 0) {
-               DRM_ERROR("MSAA surface had nonzero flags/bits\n");
+               DRM_DEBUG("MSAA surface had nonzero flags/bits\n");
                return -EINVAL;
        }
 
@@ -415,7 +426,7 @@ static int vc4_rcl_msaa_surface_setup(struct vc4_exec_info *exec,
        exec->rcl_write_bo[exec->rcl_write_bo_count++] = *obj;
 
        if (surf->offset & 0xf) {
-               DRM_ERROR("MSAA write must be 16b aligned.\n");
+               DRM_DEBUG("MSAA write must be 16b aligned.\n");
                return -EINVAL;
        }
 
@@ -437,7 +448,7 @@ static int vc4_rcl_surface_setup(struct vc4_exec_info *exec,
        int ret;
 
        if (surf->flags & ~VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
-               DRM_ERROR("Extra flags set\n");
+               DRM_DEBUG("Extra flags set\n");
                return -EINVAL;
        }
 
@@ -453,12 +464,12 @@ static int vc4_rcl_surface_setup(struct vc4_exec_info *exec,
 
        if (surf->flags & VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
                if (surf == &exec->args->zs_write) {
-                       DRM_ERROR("general zs write may not be a full-res.\n");
+                       DRM_DEBUG("general zs write may not be a full-res.\n");
                        return -EINVAL;
                }
 
                if (surf->bits != 0) {
-                       DRM_ERROR("load/store general bits set with "
+                       DRM_DEBUG("load/store general bits set with "
                                  "full res load/store.\n");
                        return -EINVAL;
                }
@@ -473,19 +484,19 @@ static int vc4_rcl_surface_setup(struct vc4_exec_info *exec,
        if (surf->bits & ~(VC4_LOADSTORE_TILE_BUFFER_TILING_MASK |
                           VC4_LOADSTORE_TILE_BUFFER_BUFFER_MASK |
                           VC4_LOADSTORE_TILE_BUFFER_FORMAT_MASK)) {
-               DRM_ERROR("Unknown bits in load/store: 0x%04x\n",
+               DRM_DEBUG("Unknown bits in load/store: 0x%04x\n",
                          surf->bits);
                return -EINVAL;
        }
 
        if (tiling > VC4_TILING_FORMAT_LT) {
-               DRM_ERROR("Bad tiling format\n");
+               DRM_DEBUG("Bad tiling format\n");
                return -EINVAL;
        }
 
        if (buffer == VC4_LOADSTORE_TILE_BUFFER_ZS) {
                if (format != 0) {
-                       DRM_ERROR("No color format should be set for ZS\n");
+                       DRM_DEBUG("No color format should be set for ZS\n");
                        return -EINVAL;
                }
                cpp = 4;
@@ -499,16 +510,16 @@ static int vc4_rcl_surface_setup(struct vc4_exec_info *exec,
                        cpp = 4;
                        break;
                default:
-                       DRM_ERROR("Bad tile buffer format\n");
+                       DRM_DEBUG("Bad tile buffer format\n");
                        return -EINVAL;
                }
        } else {
-               DRM_ERROR("Bad load/store buffer %d.\n", buffer);
+               DRM_DEBUG("Bad load/store buffer %d.\n", buffer);
                return -EINVAL;
        }
 
        if (surf->offset & 0xf) {
-               DRM_ERROR("load/store buffer must be 16b aligned.\n");
+               DRM_DEBUG("load/store buffer must be 16b aligned.\n");
                return -EINVAL;
        }
 
@@ -533,7 +544,7 @@ vc4_rcl_render_config_surface_setup(struct vc4_exec_info *exec,
        int cpp;
 
        if (surf->flags != 0) {
-               DRM_ERROR("No flags supported on render config.\n");
+               DRM_DEBUG("No flags supported on render config.\n");
                return -EINVAL;
        }
 
@@ -541,7 +552,7 @@ vc4_rcl_render_config_surface_setup(struct vc4_exec_info *exec,
                           VC4_RENDER_CONFIG_FORMAT_MASK |
                           VC4_RENDER_CONFIG_MS_MODE_4X |
                           VC4_RENDER_CONFIG_DECIMATE_MODE_4X)) {
-               DRM_ERROR("Unknown bits in render config: 0x%04x\n",
+               DRM_DEBUG("Unknown bits in render config: 0x%04x\n",
                          surf->bits);
                return -EINVAL;
        }
@@ -556,7 +567,7 @@ vc4_rcl_render_config_surface_setup(struct vc4_exec_info *exec,
        exec->rcl_write_bo[exec->rcl_write_bo_count++] = *obj;
 
        if (tiling > VC4_TILING_FORMAT_LT) {
-               DRM_ERROR("Bad tiling format\n");
+               DRM_DEBUG("Bad tiling format\n");
                return -EINVAL;
        }
 
@@ -569,7 +580,7 @@ vc4_rcl_render_config_surface_setup(struct vc4_exec_info *exec,
                cpp = 4;
                break;
        default:
-               DRM_ERROR("Bad tile buffer format\n");
+               DRM_DEBUG("Bad tile buffer format\n");
                return -EINVAL;
        }
 
@@ -590,7 +601,7 @@ int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec)
 
        if (args->min_x_tile > args->max_x_tile ||
            args->min_y_tile > args->max_y_tile) {
-               DRM_ERROR("Bad render tile set (%d,%d)-(%d,%d)\n",
+               DRM_DEBUG("Bad render tile set (%d,%d)-(%d,%d)\n",
                          args->min_x_tile, args->min_y_tile,
                          args->max_x_tile, args->max_y_tile);
                return -EINVAL;
@@ -599,7 +610,7 @@ int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec)
        if (has_bin &&
            (args->max_x_tile > exec->bin_tiles_x ||
             args->max_y_tile > exec->bin_tiles_y)) {
-               DRM_ERROR("Render tiles (%d,%d) outside of bin config "
+               DRM_DEBUG("Render tiles (%d,%d) outside of bin config "
                          "(%d,%d)\n",
                          args->max_x_tile, args->max_y_tile,
                          exec->bin_tiles_x, exec->bin_tiles_y);
@@ -642,7 +653,7 @@ int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec)
         */
        if (!setup.color_write && !setup.zs_write &&
            !setup.msaa_color_write && !setup.msaa_zs_write) {
-               DRM_ERROR("RCL requires color or Z/S write\n");
+               DRM_DEBUG("RCL requires color or Z/S write\n");
                return -EINVAL;
        }