Merge remote branch 'origin/master' into gallium_draw_llvm
[mesa.git] / src / gallium / drivers / nv50 / nv50_transfer.c
index bb7731855cd4b4af0732f9419dbcf2d19e75bd6d..9eb223eca65583415652eb8e478cd40fcdc72758 100644 (file)
@@ -1,6 +1,8 @@
 
 #include "pipe/p_context.h"
-#include "pipe/p_inlines.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
+#include "util/u_math.h"
 
 #include "nv50_context.h"
 
@@ -12,18 +14,22 @@ struct nv50_transfer {
        int level_pitch;
        int level_width;
        int level_height;
+       int level_depth;
        int level_x;
        int level_y;
+       int level_z;
+       unsigned nblocksx;
+       unsigned nblocksy;
 };
 
 static void
 nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
                        struct nouveau_bo *src_bo, unsigned src_offset,
                        int src_pitch, unsigned src_tile_mode,
-                       int sx, int sy, int sw, int sh,
+                       int sx, int sy, int sz, int sw, int sh, int sd,
                        struct nouveau_bo *dst_bo, unsigned dst_offset,
                        int dst_pitch, unsigned dst_tile_mode,
-                       int dx, int dy, int dw, int dh,
+                       int dx, int dy, int dz, int dw, int dh, int dd,
                        int cpp, int width, int height,
                        unsigned src_reloc, unsigned dst_reloc)
 {
@@ -41,7 +47,7 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
                        NV50_MEMORY_TO_MEMORY_FORMAT_LINEAR_IN, 1);
                OUT_RING  (chan, 1);
                BEGIN_RING(chan, m2mf,
-                       NV50_MEMORY_TO_MEMORY_FORMAT_PITCH_IN, 1);
+                       NV04_MEMORY_TO_MEMORY_FORMAT_PITCH_IN, 1);
                OUT_RING  (chan, src_pitch);
                src_offset += (sy * src_pitch) + (sx * cpp);
        } else {
@@ -51,8 +57,8 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
                OUT_RING  (chan, src_tile_mode << 4);
                OUT_RING  (chan, sw * cpp);
                OUT_RING  (chan, sh);
-               OUT_RING  (chan, 1);
-               OUT_RING  (chan, 0);
+               OUT_RING  (chan, sd);
+               OUT_RING  (chan, sz); /* copying only 1 zslice per call */
        }
 
        if (!dst_bo->tile_flags) {
@@ -60,7 +66,7 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
                        NV50_MEMORY_TO_MEMORY_FORMAT_LINEAR_OUT, 1);
                OUT_RING  (chan, 1);
                BEGIN_RING(chan, m2mf,
-                       NV50_MEMORY_TO_MEMORY_FORMAT_PITCH_OUT, 1);
+                       NV04_MEMORY_TO_MEMORY_FORMAT_PITCH_OUT, 1);
                OUT_RING  (chan, dst_pitch);
                dst_offset += (dy * dst_pitch) + (dx * cpp);
        } else {
@@ -70,20 +76,20 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
                OUT_RING  (chan, dst_tile_mode << 4);
                OUT_RING  (chan, dw * cpp);
                OUT_RING  (chan, dh);
-               OUT_RING  (chan, 1);
-               OUT_RING  (chan, 0);
+               OUT_RING  (chan, dd);
+               OUT_RING  (chan, dz); /* copying only 1 zslice per call */
        }
 
        while (height) {
                int line_count = height > 2047 ? 2047 : height;
 
-               WAIT_RING (chan, 15);
+               MARK_RING (chan, 15, 4); /* flush on lack of space or relocs */
                BEGIN_RING(chan, m2mf,
                        NV50_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN_HIGH, 2);
                OUT_RELOCh(chan, src_bo, src_offset, src_reloc);
                OUT_RELOCh(chan, dst_bo, dst_offset, dst_reloc);
                BEGIN_RING(chan, m2mf,
-                       NV50_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 2);
+                       NV04_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 2);
                OUT_RELOCl(chan, src_bo, src_offset, src_reloc);
                OUT_RELOCl(chan, dst_bo, dst_offset, dst_reloc);
                if (src_bo->tile_flags) {
@@ -101,7 +107,7 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
                        dst_offset += (line_count * dst_pitch);
                }
                BEGIN_RING(chan, m2mf,
-                       NV50_MEMORY_TO_MEMORY_FORMAT_LINE_LENGTH_IN, 4);
+                       NV04_MEMORY_TO_MEMORY_FORMAT_LINE_LENGTH_IN, 4);
                OUT_RING  (chan, width * cpp);
                OUT_RING  (chan, line_count);
                OUT_RING  (chan, 0x00000101);
@@ -115,61 +121,64 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen,
 }
 
 static struct pipe_transfer *
-nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
+nv50_transfer_new(struct pipe_context *pcontext, struct pipe_texture *pt,
                  unsigned face, unsigned level, unsigned zslice,
                  enum pipe_transfer_usage usage,
                  unsigned x, unsigned y, unsigned w, unsigned h)
 {
+        struct pipe_screen *pscreen = pcontext->screen;
        struct nouveau_device *dev = nouveau_screen(pscreen)->device;
        struct nv50_miptree *mt = nv50_miptree(pt);
        struct nv50_miptree_level *lvl = &mt->level[level];
        struct nv50_transfer *tx;
-       unsigned image = 0;
+       unsigned nx, ny, image = 0;
        int ret;
 
        if (pt->target == PIPE_TEXTURE_CUBE)
                image = face;
-       else
-       if (pt->target == PIPE_TEXTURE_3D)
-               image = zslice;
 
        tx = CALLOC_STRUCT(nv50_transfer);
        if (!tx)
                return NULL;
 
        pipe_texture_reference(&tx->base.texture, pt);
-       tx->base.format = pt->format;
+       tx->nblocksx = util_format_get_nblocksx(pt->format, u_minify(pt->width0, level));
+       tx->nblocksy = util_format_get_nblocksy(pt->format, u_minify(pt->height0, level));
        tx->base.width = w;
        tx->base.height = h;
-       tx->base.block = pt->block;
-       tx->base.nblocksx = pt->nblocksx[level];
-       tx->base.nblocksy = pt->nblocksy[level];
-       tx->base.stride = (w * pt->block.size);
+       tx->base.stride = tx->nblocksx * util_format_get_blocksize(pt->format);
        tx->base.usage = usage;
 
        tx->level_pitch = lvl->pitch;
-       tx->level_width = mt->base.base.width[level];
-       tx->level_height = mt->base.base.height[level];
+       tx->level_width = u_minify(mt->base.base.width0, level);
+       tx->level_height = u_minify(mt->base.base.height0, level);
+       tx->level_depth = u_minify(mt->base.base.depth0, level);
        tx->level_offset = lvl->image_offset[image];
        tx->level_tiling = lvl->tile_mode;
-       tx->level_x = x;
-       tx->level_y = y;
+       tx->level_z = zslice;
+       tx->level_x = util_format_get_nblocksx(pt->format, x);
+       tx->level_y = util_format_get_nblocksy(pt->format, y);
        ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0,
-                            w * pt->block.size * h, &tx->bo);
+                            tx->nblocksy * tx->base.stride, &tx->bo);
        if (ret) {
                FREE(tx);
                return NULL;
        }
 
-       if (usage != PIPE_TRANSFER_WRITE) {
+       if (usage & PIPE_TRANSFER_READ) {
+               nx = util_format_get_nblocksx(pt->format, tx->base.width);
+               ny = util_format_get_nblocksy(pt->format, tx->base.height);
+
                nv50_transfer_rect_m2mf(pscreen, mt->base.bo, tx->level_offset,
                                        tx->level_pitch, tx->level_tiling,
-                                       x, y,
-                                       tx->level_width, tx->level_height,
-                                       tx->bo, 0, tx->base.stride,
-                                       tx->bo->tile_mode, 0, 0,
-                                       tx->base.width, tx->base.height,
-                                       tx->base.block.size, w, h,
+                                       x, y, zslice,
+                                       tx->nblocksx, tx->nblocksy,
+                                       tx->level_depth,
+                                       tx->bo, 0,
+                                       tx->base.stride, tx->bo->tile_mode,
+                                       0, 0, 0,
+                                       tx->nblocksx, tx->nblocksy, 1,
+                                       util_format_get_blocksize(pt->format), nx, ny,
                                        NOUVEAU_BO_VRAM | NOUVEAU_BO_GART,
                                        NOUVEAU_BO_GART);
        }
@@ -178,22 +187,28 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 }
 
 static void
-nv50_transfer_del(struct pipe_transfer *ptx)
+nv50_transfer_del(struct pipe_context *pcontext, struct pipe_transfer *ptx)
 {
        struct nv50_transfer *tx = (struct nv50_transfer *)ptx;
        struct nv50_miptree *mt = nv50_miptree(ptx->texture);
+       struct pipe_texture *pt = ptx->texture;
 
-       if (ptx->usage != PIPE_TRANSFER_READ) {
-               struct pipe_screen *pscreen = ptx->texture->screen;
-               nv50_transfer_rect_m2mf(pscreen, tx->bo, 0, tx->base.stride,
-                                       tx->bo->tile_mode, 0, 0,
-                                       tx->base.width, tx->base.height,
+       unsigned nx = util_format_get_nblocksx(pt->format, tx->base.width);
+       unsigned ny = util_format_get_nblocksy(pt->format, tx->base.height);
+
+       if (ptx->usage & PIPE_TRANSFER_WRITE) {
+               struct pipe_screen *pscreen = pcontext->screen;
+
+               nv50_transfer_rect_m2mf(pscreen, tx->bo, 0,
+                                       tx->base.stride, tx->bo->tile_mode,
+                                       0, 0, 0,
+                                       tx->nblocksx, tx->nblocksy, 1,
                                        mt->base.bo, tx->level_offset,
                                        tx->level_pitch, tx->level_tiling,
-                                       tx->level_x, tx->level_y,
-                                       tx->level_width, tx->level_height,
-                                       tx->base.block.size, tx->base.width,
-                                       tx->base.height,
+                                       tx->level_x, tx->level_y, tx->level_z,
+                                       tx->nblocksx, tx->nblocksy,
+                                       tx->level_depth,
+                                       util_format_get_blocksize(pt->format), nx, ny,
                                        NOUVEAU_BO_GART, NOUVEAU_BO_VRAM |
                                        NOUVEAU_BO_GART);
        }
@@ -204,7 +219,7 @@ nv50_transfer_del(struct pipe_transfer *ptx)
 }
 
 static void *
-nv50_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+nv50_transfer_map(struct pipe_context *pcontext, struct pipe_transfer *ptx)
 {
        struct nv50_transfer *tx = (struct nv50_transfer *)ptx;
        unsigned flags = 0;
@@ -222,7 +237,7 @@ nv50_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
 }
 
 static void
-nv50_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+nv50_transfer_unmap(struct pipe_context *pcontext, struct pipe_transfer *ptx)
 {
        struct nv50_transfer *tx = (struct nv50_transfer *)ptx;
 
@@ -230,10 +245,96 @@ nv50_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
 }
 
 void
-nv50_transfer_init_screen_functions(struct pipe_screen *pscreen)
+nv50_init_transfer_functions(struct nv50_context *nv50)
 {
-       pscreen->get_tex_transfer = nv50_transfer_new;
-       pscreen->tex_transfer_destroy = nv50_transfer_del;
-       pscreen->transfer_map = nv50_transfer_map;
-       pscreen->transfer_unmap = nv50_transfer_unmap;
+       nv50->pipe.get_tex_transfer = nv50_transfer_new;
+       nv50->pipe.tex_transfer_destroy = nv50_transfer_del;
+       nv50->pipe.transfer_map = nv50_transfer_map;
+       nv50->pipe.transfer_unmap = nv50_transfer_unmap;
+}
+
+void
+nv50_upload_sifc(struct nv50_context *nv50,
+                struct nouveau_bo *bo, unsigned dst_offset, unsigned reloc,
+                unsigned dst_format, int dst_w, int dst_h, int dst_pitch,
+                void *src, unsigned src_format, int src_pitch,
+                int x, int y, int w, int h, int cpp)
+{
+       struct nouveau_channel *chan = nv50->screen->base.channel;
+       struct nouveau_grobj *eng2d = nv50->screen->eng2d;
+       struct nouveau_grobj *tesla = nv50->screen->tesla;
+       unsigned line_dwords = (w * cpp + 3) / 4;
+
+       reloc |= NOUVEAU_BO_WR;
+
+       MARK_RING (chan, 32, 2); /* flush on lack of space or relocs */
+
+       if (bo->tile_flags) {
+               BEGIN_RING(chan, eng2d, NV50_2D_DST_FORMAT, 5);
+               OUT_RING  (chan, dst_format);
+               OUT_RING  (chan, 0);
+               OUT_RING  (chan, bo->tile_mode << 4);
+               OUT_RING  (chan, 1);
+               OUT_RING  (chan, 0);
+       } else {
+               BEGIN_RING(chan, eng2d, NV50_2D_DST_FORMAT, 2);
+               OUT_RING  (chan, dst_format);
+               OUT_RING  (chan, 1);
+               BEGIN_RING(chan, eng2d, NV50_2D_DST_PITCH, 1);
+               OUT_RING  (chan, dst_pitch);
+       }
+
+       BEGIN_RING(chan, eng2d, NV50_2D_DST_WIDTH, 4);
+       OUT_RING  (chan, dst_w);
+       OUT_RING  (chan, dst_h);
+       OUT_RELOCh(chan, bo, dst_offset, reloc);
+       OUT_RELOCl(chan, bo, dst_offset, reloc);
+
+       /* NV50_2D_OPERATION_SRCCOPY assumed already set */
+
+       BEGIN_RING(chan, eng2d, NV50_2D_SIFC_BITMAP_ENABLE, 2);
+       OUT_RING  (chan, 0);
+       OUT_RING  (chan, src_format);
+       BEGIN_RING(chan, eng2d, NV50_2D_SIFC_WIDTH, 10);
+       OUT_RING  (chan, w);
+       OUT_RING  (chan, h);
+       OUT_RING  (chan, 0);
+       OUT_RING  (chan, 1);
+       OUT_RING  (chan, 0);
+       OUT_RING  (chan, 1);
+       OUT_RING  (chan, 0);
+       OUT_RING  (chan, x);
+       OUT_RING  (chan, 0);
+       OUT_RING  (chan, y);
+
+       while (h--) {
+               const uint32_t *p = src;
+               unsigned count = line_dwords;
+
+               while (count) {
+                       unsigned nr = MIN2(count, 1792);
+
+                       if (AVAIL_RING(chan) <= nr) {
+                               FIRE_RING (chan);
+
+                               BEGIN_RING(chan, eng2d,
+                                          NV50_2D_DST_ADDRESS_HIGH, 2);
+                               OUT_RELOCh(chan, bo, dst_offset, reloc);
+                               OUT_RELOCl(chan, bo, dst_offset, reloc);
+                       }
+                       assert(AVAIL_RING(chan) > nr);
+
+                       BEGIN_RING(chan, eng2d,
+                                  NV50_2D_SIFC_DATA | (2 << 29), nr);
+                       OUT_RINGp (chan, p, nr);
+
+                       p += nr;
+                       count -= nr;
+               }
+
+               src += src_pitch;
+       }
+
+       BEGIN_RING(chan, tesla, NV50TCL_CODE_CB_FLUSH, 1);
+       OUT_RING  (chan, 0);
 }