driver: align mipmaps to 64 bytes

This fixes automatic mipmap generation with 16 bit textures

4 files changed, 29 insertions, 10 deletions

diff --git a/native/driver/etna_internal.h b/native/driver/etna_internal.h
index 7e534a1..d9eaaaf 100644
--- a/native/driver/etna_internal.h
+++ b/native/driver/etna_internal.h
@@ -39,6 +39,8 @@
 #define ETNA_RS_WIDTH_MASK (16-1)
 /* RS tiled operations must have height%4 = 0 */
 #define ETNA_RS_HEIGHT_MASK (3)
+/* PE render targets must be aligned to 64 bytes */
+#define ETNA_PE_ALIGNMENT (64)
 
 /* GPU chip 3D specs */
 struct etna_pipe_specs
diff --git a/native/driver/etna_pipe.c b/native/driver/etna_pipe.c
index fd7e4c3..713e282 100644
--- a/native/driver/etna_pipe.c
+++ b/native/driver/etna_pipe.c
@@ -939,7 +939,6 @@ static void etna_pipe_set_framebuffer_state(struct pipe_context *pipe,
     int nr_samples_color = -1;
     int nr_samples_depth = -1;
 
-
     /* Set up TS as well. Warning: this state is used by both the RS and PE */
     uint32_t ts_mem_config = 0;
     if(sv->nr_cbufs > 0) /* at least one color buffer? */
@@ -953,6 +952,7 @@ static void etna_pipe_set_framebuffer_state(struct pipe_context *pipe,
                 (color_supertiled ? VIVS_PE_COLOR_FORMAT_SUPER_TILED : 0);
                 /* XXX VIVS_PE_COLOR_FORMAT_OVERWRITE and the rest comes from blend_state / depth_stencil_alpha */
                 /* merged with depth_stencil_alpha */
+
         if (priv->ctx->conn->chip.pixel_pipes == 1)
         {
             cs->PE_COLOR_ADDR = cbuf->surf.address;
diff --git a/native/driver/etna_resource.c b/native/driver/etna_resource.c
index bb9a143..3a2272d 100644
--- a/native/driver/etna_resource.c
+++ b/native/driver/etna_resource.c
@@ -203,7 +203,7 @@ static struct pipe_resource * etna_screen_resource_create(struct pipe_screen *sc
         mip->layer_stride = align(mip->padded_width, divSizeX)/divSizeX *
                       align(mip->padded_height, divSizeY)/divSizeY * element_size;
         mip->size = templat->array_size * mip->layer_stride;
-        offset += mip->size;
+        offset += align(mip->size, ETNA_PE_ALIGNMENT); /* align mipmaps to 64 bytes to be able to render to them */
         if(ix == max_mip_level || (x == 1 && y == 1))
             break; // stop at last level
         x = (x+1)>>1;
diff --git a/native/fb/downsample_test.c b/native/fb/downsample_test.c
index e9154fa..de37bb4 100644
--- a/native/fb/downsample_test.c
+++ b/native/fb/downsample_test.c
@@ -38,6 +38,7 @@
 
 #include "etna_pipe.h"
 #include "util/u_inlines.h"
+#include "util/u_format.h"
 #include "util/u_gen_mipmap.h"
 #include "cso_cache/cso_context.h"
 #include "write_bmp.h"
@@ -53,15 +54,28 @@ int main(int argc, char **argv)
     struct fbdemos_scaffold *fbs = 0;
     fbdemo_init(&fbs);
     struct pipe_context *pipe = fbs->pipe;
+    unsigned format = PIPE_FORMAT_B5G6R5_UNORM; /* texture format */
+    //unsigned format = PIPE_FORMAT_B8G8R8X8_UNORM;
+    const struct util_format_description *format_desc = util_format_description(format);
+    unsigned bs = util_format_get_blocksize(format);
 
     /* Convert and upload embedded texture */
-    struct pipe_resource *tex_resource = fbdemo_create_2d(fbs->screen, 
-            PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET, PIPE_FORMAT_B8G8R8X8_UNORM, 
+    struct pipe_resource *tex_resource = fbdemo_create_2d(fbs->screen,
+            PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET, format,
             COMPANION_TEXTURE_WIDTH, COMPANION_TEXTURE_HEIGHT, 1000);
-    void *temp = malloc(COMPANION_TEXTURE_WIDTH * COMPANION_TEXTURE_HEIGHT * 4); 
-    etna_convert_r8g8b8_to_b8g8r8x8(temp, (const uint8_t*)companion_texture, COMPANION_TEXTURE_WIDTH * COMPANION_TEXTURE_HEIGHT);
-    etna_pipe_inline_write(pipe, tex_resource, 0, 0, temp, COMPANION_TEXTURE_WIDTH * COMPANION_TEXTURE_HEIGHT * 4);
-    free(temp);
+
+    /* First convert from cumbersome RGB to RGBX */
+    void *temp_rgbx8888 = malloc(COMPANION_TEXTURE_WIDTH * COMPANION_TEXTURE_HEIGHT * 4);
+    etna_convert_r8g8b8_to_b8g8r8x8(temp_rgbx8888, (const uint8_t*)companion_texture, COMPANION_TEXTURE_WIDTH * COMPANION_TEXTURE_HEIGHT);
+
+    /* Then convert to destination format */
+    void *temp_fmt = malloc(COMPANION_TEXTURE_WIDTH * COMPANION_TEXTURE_HEIGHT * bs);
+    format_desc->pack_rgba_8unorm(temp_fmt, COMPANION_TEXTURE_WIDTH*bs, temp_rgbx8888, COMPANION_TEXTURE_WIDTH*4,
+            COMPANION_TEXTURE_WIDTH, COMPANION_TEXTURE_HEIGHT);
+    etna_pipe_inline_write(pipe, tex_resource, 0, 0, temp_fmt, COMPANION_TEXTURE_WIDTH * COMPANION_TEXTURE_HEIGHT * bs);
+
+    free(temp_rgbx8888);
+    free(temp_fmt);
     /* 0 512x512
      * 1 256x256
      * 2 128x128
@@ -87,7 +101,6 @@ int main(int argc, char **argv)
     assert(cso);
     struct gen_mipmap_state *gen_mipmap = util_create_gen_mipmap(pipe, cso);
     assert(gen_mipmap);
-
     for(int frame=0; frame<1; ++frame)
     {
         if(frame%50 == 0)
@@ -119,8 +132,12 @@ int main(int argc, char **argv)
         void *data = pipe->transfer_map(pipe, tex_resource, level, PIPE_TRANSFER_READ, 
                 &box, &transfer);
 
+        void *temp = malloc(lwidth * lheight * 4);
+        printf("%i: Transfer stride is %i\n", level, transfer->stride);
+        format_desc->unpack_rgba_8unorm(temp, lwidth*4, data, transfer->stride, lwidth, lheight);
         snprintf(filename, sizeof(filename), "mip%i.bmp", level);
-        bmp_dump32_ex(data, lwidth, lheight, /*flip*/ false, /*bgra*/true, /*alpha*/false, filename);
+        bmp_dump32_ex(temp, lwidth, lheight, /*flip*/ false, /*bgra*/true, /*alpha*/false, filename);
+        free(temp);
 
         pipe->transfer_unmap(pipe, transfer);