static void
nouveau_vpe_fini(struct nouveau_decoder *dec) {
struct nouveau_pushbuf *push = dec->push;
if (!dec->cmds)
return;
nouveau_pushbuf_space(push, 16, 2, 0);
nouveau_bufctx_reset(dec->bufctx, NV31_VIDEO_BIND_CMD);
#define BCTX_ARGS dec->bufctx, NV31_VIDEO_BIND_CMD, NOUVEAU_BO_RD
BEGIN_NV04(push, NV31_MPEG(CMD_OFFSET), 2);
PUSH_MTHDl(push, NV31_MPEG(CMD_OFFSET), dec->cmd_bo, 0, BCTX_ARGS);
PUSH_DATA (push, dec->ofs * 4);
BEGIN_NV04(push, NV31_MPEG(DATA_OFFSET), 2);
PUSH_MTHDl(push, NV31_MPEG(DATA_OFFSET), dec->data_bo, 0, BCTX_ARGS);
PUSH_DATA (push, dec->data_pos * 4);
#undef BCTX_ARGS
if (unlikely(nouveau_pushbuf_validate(dec->push)))
return;
BEGIN_NV04(push, NV31_MPEG(EXEC), 1);
PUSH_DATA (push, 1);
nouveau_vpe_synch(dec);
dec->ofs = dec->data_pos = dec->num_surfaces = 0;
dec->cmds = dec->data = NULL;
dec->current = dec->future = dec->past = 8;
}
static void
nv50_compute_upload_input(struct nv50_context *nv50, const uint32_t *input)
{
struct nv50_screen *screen = nv50->screen;
struct nouveau_pushbuf *push = screen->base.pushbuf;
unsigned size = align(nv50->compprog->parm_size, 0x4);
BEGIN_NV04(push, NV50_COMPUTE(USER_PARAM_COUNT), 1);
PUSH_DATA (push, (size / 4) << 8);
if (size) {
struct nouveau_mm_allocation *mm;
struct nouveau_bo *bo = NULL;
unsigned offset;
mm = nouveau_mm_allocate(screen->base.mm_GART, size, &bo, &offset);
assert(mm);
nouveau_bo_map(bo, 0, screen->base.client);
memcpy(bo->map + offset, input, size);
nouveau_bufctx_refn(nv50->bufctx, 0, bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
nouveau_pushbuf_bufctx(push, nv50->bufctx);
nouveau_pushbuf_validate(push);
BEGIN_NV04(push, NV50_COMPUTE(USER_PARAM(0)), size / 4);
nouveau_pushbuf_data(push, bo, offset, size);
nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work, mm);
nouveau_bo_ref(NULL, &bo);
nouveau_bufctx_reset(nv50->bufctx, 0);
}
}
static void
nve4_m2mf_copy_linear(struct nouveau_context *nv,
struct nouveau_bo *dst, unsigned dstoff, unsigned dstdom,
struct nouveau_bo *src, unsigned srcoff, unsigned srcdom,
unsigned size)
{
struct nouveau_pushbuf *push = nv->pushbuf;
struct nouveau_bufctx *bctx = nvc0_context(&nv->pipe)->bufctx;
nouveau_bufctx_refn(bctx, 0, src, srcdom | NOUVEAU_BO_RD);
nouveau_bufctx_refn(bctx, 0, dst, dstdom | NOUVEAU_BO_WR);
nouveau_pushbuf_bufctx(push, bctx);
nouveau_pushbuf_validate(push);
BEGIN_NVC0(push, SUBC_COPY(0x0400), 4);
PUSH_DATAh(push, src->offset + srcoff);
PUSH_DATA (push, src->offset + srcoff);
PUSH_DATAh(push, dst->offset + dstoff);
PUSH_DATA (push, dst->offset + dstoff);
BEGIN_NVC0(push, SUBC_COPY(0x0418), 1);
PUSH_DATA (push, size);
BEGIN_NVC0(push, SUBC_COPY(0x0300), 1);
PUSH_DATA (push, 0x186);
nouveau_bufctx_reset(bctx, 0);
}
static void
nv10_clear(struct gl_context *ctx, GLbitfield buffers)
{
struct nouveau_context *nctx = to_nouveau_context(ctx);
struct nouveau_pushbuf *push = context_push(ctx);
nouveau_validate_framebuffer(ctx);
nouveau_pushbuf_bufctx(push, nctx->hw.bufctx);
if (nouveau_pushbuf_validate(push)) {
nouveau_pushbuf_bufctx(push, NULL);
return;
}
if ((buffers & BUFFER_BIT_DEPTH) && ctx->Depth.Mask) {
if (context_chipset(ctx) >= 0x17)
nv17_zclear(ctx, &buffers);
else
nv10_zclear(ctx, &buffers);
/* Emit the zclear state if it's dirty */
_mesa_update_state(ctx);
}
nouveau_pushbuf_bufctx(push, NULL);
nouveau_clear(ctx, buffers);
}
static void
nv20_clear(struct gl_context *ctx, GLbitfield buffers)
{
struct nouveau_context *nctx = to_nouveau_context(ctx);
struct nouveau_pushbuf *push = context_push(ctx);
struct gl_framebuffer *fb = ctx->DrawBuffer;
uint32_t clear = 0;
nouveau_validate_framebuffer(ctx);
nouveau_pushbuf_bufctx(push, nctx->hw.bufctx);
if (nouveau_pushbuf_validate(push)) {
nouveau_pushbuf_bufctx(push, NULL);
return;
}
if (buffers & BUFFER_BITS_COLOR) {
struct nouveau_surface *s = &to_nouveau_renderbuffer(
fb->_ColorDrawBuffers[0])->surface;
if (ctx->Color.ColorMask[0][RCOMP])
clear |= NV20_3D_CLEAR_BUFFERS_COLOR_R;
if (ctx->Color.ColorMask[0][GCOMP])
clear |= NV20_3D_CLEAR_BUFFERS_COLOR_G;
if (ctx->Color.ColorMask[0][BCOMP])
clear |= NV20_3D_CLEAR_BUFFERS_COLOR_B;
if (ctx->Color.ColorMask[0][ACOMP])
clear |= NV20_3D_CLEAR_BUFFERS_COLOR_A;
BEGIN_NV04(push, NV20_3D(CLEAR_VALUE), 1);
PUSH_DATA (push, pack_rgba_clamp_f(s->format, ctx->Color.ClearColor.f));
buffers &= ~BUFFER_BITS_COLOR;
}
if (buffers & (BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL)) {
struct nouveau_surface *s = &to_nouveau_renderbuffer(
fb->Attachment[BUFFER_DEPTH].Renderbuffer)->surface;
if (buffers & BUFFER_BIT_DEPTH && ctx->Depth.Mask)
clear |= NV20_3D_CLEAR_BUFFERS_DEPTH;
if (buffers & BUFFER_BIT_STENCIL && ctx->Stencil.WriteMask[0])
clear |= NV20_3D_CLEAR_BUFFERS_STENCIL;
BEGIN_NV04(push, NV20_3D(CLEAR_DEPTH_VALUE), 1);
PUSH_DATA (push, pack_zs_f(s->format, ctx->Depth.Clear,
ctx->Stencil.Clear));
buffers &= ~(BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL);
}
BEGIN_NV04(push, NV20_3D(CLEAR_BUFFERS), 1);
PUSH_DATA (push, clear);
nouveau_pushbuf_bufctx(push, NULL);
nouveau_clear(ctx, buffers);
}
Bool
NV10EXAPrepareComposite(int op,
PicturePtr pict_src,
PicturePtr pict_mask,
PicturePtr pict_dst,
PixmapPtr src,
PixmapPtr mask,
PixmapPtr dst)
{
ScrnInfoPtr pScrn = xf86ScreenToScrn(dst->drawable.pScreen);
NVPtr pNv = NVPTR(pScrn);
struct nouveau_pushbuf *push = pNv->pushbuf;
uint32_t sc, sa, mc, ma;
if (!PUSH_SPACE(push, 128))
return FALSE;
PUSH_RESET(push);
/* setup render target and blending */
if (!setup_render_target(pNv, pict_dst, dst))
return FALSE;
setup_blend_function(pNv, pict_dst, op);
/* select picture sources */
if (!setup_picture(pNv, pict_src, src, 0, &sc, &sa))
return FALSE;
if (!setup_picture(pNv, pict_mask, mask, 1, &mc, &ma))
return FALSE;
/* configure register combiners */
BEGIN_NV04(push, NV10_3D(RC_IN_ALPHA(0)), 1);
PUSH_DATA (push, sa | ma);
BEGIN_NV04(push, NV10_3D(RC_IN_RGB(0)), 1);
if (effective_component_alpha(pict_mask)) {
if (needs_src_alpha(op))
PUSH_DATA(push, sa | mc);
else
PUSH_DATA(push, sc | mc);
} else {
PUSH_DATA(push, sc | ma);
}
nouveau_pushbuf_bufctx(push, pNv->bufctx);
if (nouveau_pushbuf_validate(push)) {
nouveau_pushbuf_bufctx(push, NULL);
return FALSE;
}
pNv->pspict = pict_src;
pNv->pmpict = pict_mask;
return TRUE;
}
void
nvc0_m2mf_push_linear(struct nouveau_context *nv,
struct nouveau_bo *dst, unsigned offset, unsigned domain,
unsigned size, const void *data)
{
struct nvc0_context *nvc0 = nvc0_context(&nv->pipe);
struct nouveau_pushbuf *push = nv->pushbuf;
uint32_t *src = (uint32_t *)data;
unsigned count = (size + 3) / 4;
nouveau_bufctx_refn(nvc0->bufctx, 0, dst, domain | NOUVEAU_BO_WR);
nouveau_pushbuf_bufctx(push, nvc0->bufctx);
nouveau_pushbuf_validate(push);
while (count) {
unsigned nr;
if (!PUSH_SPACE(push, 16))
break;
nr = PUSH_AVAIL(push);
assert(nr >= 16);
nr = MIN2(count, nr - 9);
nr = MIN2(nr, NV04_PFIFO_MAX_PACKET_LEN);
BEGIN_NVC0(push, NVC0_M2MF(OFFSET_OUT_HIGH), 2);
PUSH_DATAh(push, dst->offset + offset);
PUSH_DATA (push, dst->offset + offset);
BEGIN_NVC0(push, NVC0_M2MF(LINE_LENGTH_IN), 2);
PUSH_DATA (push, MIN2(size, nr * 4));
PUSH_DATA (push, 1);
BEGIN_NVC0(push, NVC0_M2MF(EXEC), 1);
PUSH_DATA (push, 0x100111);
/* must not be interrupted (trap on QUERY fence, 0x50 works however) */
BEGIN_NIC0(push, NVC0_M2MF(DATA), nr);
PUSH_DATAp(push, src, nr);
count -= nr;
src += nr;
offset += nr * 4;
size -= nr * 4;
}
nouveau_bufctx_reset(nvc0->bufctx, 0);
}
static boolean
nvc0_compute_state_validate(struct nvc0_context *nvc0)
{
if (!nvc0_compute_validate_program(nvc0))
return FALSE;
/* TODO: textures, samplers, surfaces, global memory buffers */
nvc0_bufctx_fence(nvc0, nvc0->bufctx_cp, FALSE);
nouveau_pushbuf_bufctx(nvc0->base.pushbuf, nvc0->bufctx_cp);
if (unlikely(nouveau_pushbuf_validate(nvc0->base.pushbuf)))
return FALSE;
if (unlikely(nvc0->state.flushed))
nvc0_bufctx_fence(nvc0, nvc0->bufctx_cp, TRUE);
return TRUE;
}
static bool
nv50_compute_state_validate(struct nv50_context *nv50)
{
if (!nv50_compute_validate_program(nv50))
return false;
if (nv50->dirty_cp & NV50_NEW_CP_GLOBALS)
nv50_compute_validate_globals(nv50);
/* TODO: validate textures, samplers, surfaces */
nv50_bufctx_fence(nv50->bufctx_cp, false);
nouveau_pushbuf_bufctx(nv50->base.pushbuf, nv50->bufctx_cp);
if (unlikely(nouveau_pushbuf_validate(nv50->base.pushbuf)))
return false;
if (unlikely(nv50->state.flushed))
nv50_bufctx_fence(nv50->bufctx_cp, true);
return true;
}
void
nvc0_cb_push(struct nouveau_context *nv,
struct nouveau_bo *bo, unsigned domain,
unsigned base, unsigned size,
unsigned offset, unsigned words, const uint32_t *data)
{
struct nouveau_bufctx *bctx = nvc0_context(&nv->pipe)->bufctx;
struct nouveau_pushbuf *push = nv->pushbuf;
NOUVEAU_DRV_STAT(nv->screen, constbuf_upload_count, 1);
NOUVEAU_DRV_STAT(nv->screen, constbuf_upload_bytes, words * 4);
assert(!(offset & 3));
size = align(size, 0x100);
nouveau_bufctx_refn(bctx, 0, bo, NOUVEAU_BO_WR | domain);
nouveau_pushbuf_bufctx(push, bctx);
nouveau_pushbuf_validate(push);
BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
PUSH_DATA (push, size);
PUSH_DATAh(push, bo->offset + base);
PUSH_DATA (push, bo->offset + base);
while (words) {
unsigned nr = PUSH_AVAIL(push);
nr = MIN2(nr, words);
nr = MIN2(nr, NV04_PFIFO_MAX_PACKET_LEN - 1);
BEGIN_1IC0(push, NVC0_3D(CB_POS), nr + 1);
PUSH_DATA (push, offset);
PUSH_DATAp(push, data, nr);
words -= nr;
data += nr;
offset += nr * 4;
}
nouveau_bufctx_reset(bctx, 0);
}
void
nve4_p2mf_push_linear(struct nouveau_context *nv,
struct nouveau_bo *dst, unsigned offset, unsigned domain,
unsigned size, const void *data)
{
struct nvc0_context *nvc0 = nvc0_context(&nv->pipe);
struct nouveau_pushbuf *push = nv->pushbuf;
uint32_t *src = (uint32_t *)data;
unsigned count = (size + 3) / 4;
nouveau_bufctx_refn(nvc0->bufctx, 0, dst, domain | NOUVEAU_BO_WR);
nouveau_pushbuf_bufctx(push, nvc0->bufctx);
nouveau_pushbuf_validate(push);
while (count) {
unsigned nr = MIN2(count, (NV04_PFIFO_MAX_PACKET_LEN - 1));
if (!PUSH_SPACE(push, nr + 10))
break;
BEGIN_NVC0(push, NVE4_P2MF(UPLOAD_DST_ADDRESS_HIGH), 2);
PUSH_DATAh(push, dst->offset + offset);
PUSH_DATA (push, dst->offset + offset);
BEGIN_NVC0(push, NVE4_P2MF(UPLOAD_LINE_LENGTH_IN), 2);
PUSH_DATA (push, MIN2(size, nr * 4));
PUSH_DATA (push, 1);
/* must not be interrupted (trap on QUERY fence, 0x50 works however) */
BEGIN_1IC0(push, NVE4_P2MF(UPLOAD_EXEC), nr + 1);
PUSH_DATA (push, 0x1001);
PUSH_DATAp(push, src, nr);
count -= nr;
src += nr;
offset += nr * 4;
size -= nr * 4;
}
nouveau_bufctx_reset(nvc0->bufctx, 0);
}
static INLINE void *
nvc0_push_setup_vertex_array(struct nvc0_context *nvc0, const unsigned count)
{
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nouveau_bo *bo;
uint64_t va;
const unsigned size = count * nvc0->vertex->size;
void *const dest = nouveau_scratch_get(&nvc0->base, size, &va, &bo);
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_START_HIGH(0)), 2);
PUSH_DATAh(push, va);
PUSH_DATA (push, va);
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(0)), 2);
PUSH_DATAh(push, va + size - 1);
PUSH_DATA (push, va + size - 1);
BCTX_REFN_bo(nvc0->bufctx_3d, VTX_TMP, NOUVEAU_BO_GART | NOUVEAU_BO_RD,
bo);
nouveau_pushbuf_validate(push);
return dest;
}
static void
nvc0_m2mf_copy_linear(struct nouveau_context *nv,
struct nouveau_bo *dst, unsigned dstoff, unsigned dstdom,
struct nouveau_bo *src, unsigned srcoff, unsigned srcdom,
unsigned size)
{
struct nouveau_pushbuf *push = nv->pushbuf;
struct nouveau_bufctx *bctx = nvc0_context(&nv->pipe)->bufctx;
nouveau_bufctx_refn(bctx, 0, src, srcdom | NOUVEAU_BO_RD);
nouveau_bufctx_refn(bctx, 0, dst, dstdom | NOUVEAU_BO_WR);
nouveau_pushbuf_bufctx(push, bctx);
nouveau_pushbuf_validate(push);
while (size) {
unsigned bytes = MIN2(size, 1 << 17);
BEGIN_NVC0(push, NVC0_M2MF(OFFSET_OUT_HIGH), 2);
PUSH_DATAh(push, dst->offset + dstoff);
PUSH_DATA (push, dst->offset + dstoff);
BEGIN_NVC0(push, NVC0_M2MF(OFFSET_IN_HIGH), 2);
PUSH_DATAh(push, src->offset + srcoff);
PUSH_DATA (push, src->offset + srcoff);
BEGIN_NVC0(push, NVC0_M2MF(LINE_LENGTH_IN), 2);
PUSH_DATA (push, bytes);
PUSH_DATA (push, 1);
BEGIN_NVC0(push, NVC0_M2MF(EXEC), 1);
PUSH_DATA (push, (1 << NVC0_M2MF_EXEC_INC__SHIFT) |
NVC0_M2MF_EXEC_LINEAR_IN | NVC0_M2MF_EXEC_LINEAR_OUT);
srcoff += bytes;
dstoff += bytes;
size -= bytes;
}
nouveau_bufctx_reset(bctx, 0);
}
static void
nvc0_m2mf_transfer_rect(struct nvc0_context *nvc0,
const struct nv50_m2mf_rect *dst,
const struct nv50_m2mf_rect *src,
uint32_t nblocksx, uint32_t nblocksy)
{
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nouveau_bufctx *bctx = nvc0->bufctx;
const int cpp = dst->cpp;
uint32_t src_ofst = src->base;
uint32_t dst_ofst = dst->base;
uint32_t height = nblocksy;
uint32_t sy = src->y;
uint32_t dy = dst->y;
uint32_t exec = (1 << 20);
assert(dst->cpp == src->cpp);
nouveau_bufctx_refn(bctx, 0, src->bo, src->domain | NOUVEAU_BO_RD);
nouveau_bufctx_refn(bctx, 0, dst->bo, dst->domain | NOUVEAU_BO_WR);
nouveau_pushbuf_bufctx(push, bctx);
nouveau_pushbuf_validate(push);
if (nouveau_bo_memtype(src->bo)) {
BEGIN_NVC0(push, NVC0_M2MF(TILING_MODE_IN), 5);
PUSH_DATA (push, src->tile_mode);
PUSH_DATA (push, src->width * cpp);
PUSH_DATA (push, src->height);
PUSH_DATA (push, src->depth);
PUSH_DATA (push, src->z);
} else {
src_ofst += src->y * src->pitch + src->x * cpp;
BEGIN_NVC0(push, NVC0_M2MF(PITCH_IN), 1);
PUSH_DATA (push, src->width * cpp);
exec |= NVC0_M2MF_EXEC_LINEAR_IN;
}
if (nouveau_bo_memtype(dst->bo)) {
BEGIN_NVC0(push, NVC0_M2MF(TILING_MODE_OUT), 5);
PUSH_DATA (push, dst->tile_mode);
PUSH_DATA (push, dst->width * cpp);
PUSH_DATA (push, dst->height);
PUSH_DATA (push, dst->depth);
PUSH_DATA (push, dst->z);
} else {
dst_ofst += dst->y * dst->pitch + dst->x * cpp;
BEGIN_NVC0(push, NVC0_M2MF(PITCH_OUT), 1);
PUSH_DATA (push, dst->width * cpp);
exec |= NVC0_M2MF_EXEC_LINEAR_OUT;
}
while (height) {
int line_count = height > 2047 ? 2047 : height;
BEGIN_NVC0(push, NVC0_M2MF(OFFSET_IN_HIGH), 2);
PUSH_DATAh(push, src->bo->offset + src_ofst);
PUSH_DATA (push, src->bo->offset + src_ofst);
BEGIN_NVC0(push, NVC0_M2MF(OFFSET_OUT_HIGH), 2);
PUSH_DATAh(push, dst->bo->offset + dst_ofst);
PUSH_DATA (push, dst->bo->offset + dst_ofst);
if (!(exec & NVC0_M2MF_EXEC_LINEAR_IN)) {
BEGIN_NVC0(push, NVC0_M2MF(TILING_POSITION_IN_X), 2);
PUSH_DATA (push, src->x * cpp);
PUSH_DATA (push, sy);
} else {
src_ofst += line_count * src->pitch;
}
if (!(exec & NVC0_M2MF_EXEC_LINEAR_OUT)) {
BEGIN_NVC0(push, NVC0_M2MF(TILING_POSITION_OUT_X), 2);
PUSH_DATA (push, dst->x * cpp);
PUSH_DATA (push, dy);
} else {
dst_ofst += line_count * dst->pitch;
}
BEGIN_NVC0(push, NVC0_M2MF(LINE_LENGTH_IN), 2);
PUSH_DATA (push, nblocksx * cpp);
PUSH_DATA (push, line_count);
BEGIN_NVC0(push, NVC0_M2MF(EXEC), 1);
PUSH_DATA (push, exec);
height -= line_count;
sy += line_count;
dy += line_count;
}
nouveau_bufctx_reset(bctx, 0);
}
static void
nvc0_resource_copy_region(struct pipe_context *pipe,
struct pipe_resource *dst, unsigned dst_level,
unsigned dstx, unsigned dsty, unsigned dstz,
struct pipe_resource *src, unsigned src_level,
const struct pipe_box *src_box)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
int ret;
boolean m2mf;
unsigned dst_layer = dstz, src_layer = src_box->z;
if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
nouveau_copy_buffer(&nvc0->base,
nv04_resource(dst), dstx,
nv04_resource(src), src_box->x, src_box->width);
NOUVEAU_DRV_STAT(&nvc0->screen->base, buf_copy_bytes, src_box->width);
return;
}
NOUVEAU_DRV_STAT(&nvc0->screen->base, tex_copy_count, 1);
/* 0 and 1 are equal, only supporting 0/1, 2, 4 and 8 */
assert((src->nr_samples | 1) == (dst->nr_samples | 1));
m2mf = (src->format == dst->format) ||
(util_format_get_blocksizebits(src->format) ==
util_format_get_blocksizebits(dst->format));
nv04_resource(dst)->status |= NOUVEAU_BUFFER_STATUS_GPU_WRITING;
if (m2mf) {
struct nv50_m2mf_rect drect, srect;
unsigned i;
unsigned nx = util_format_get_nblocksx(src->format, src_box->width);
unsigned ny = util_format_get_nblocksy(src->format, src_box->height);
nv50_m2mf_rect_setup(&drect, dst, dst_level, dstx, dsty, dstz);
nv50_m2mf_rect_setup(&srect, src, src_level,
src_box->x, src_box->y, src_box->z);
for (i = 0; i < src_box->depth; ++i) {
nvc0->m2mf_copy_rect(nvc0, &drect, &srect, nx, ny);
if (nv50_miptree(dst)->layout_3d)
drect.z++;
else
drect.base += nv50_miptree(dst)->layer_stride;
if (nv50_miptree(src)->layout_3d)
srect.z++;
else
srect.base += nv50_miptree(src)->layer_stride;
}
return;
}
assert(nv50_2d_dst_format_faithful(dst->format));
assert(nv50_2d_src_format_faithful(src->format));
BCTX_REFN(nvc0->bufctx, 2D, nv04_resource(src), RD);
BCTX_REFN(nvc0->bufctx, 2D, nv04_resource(dst), WR);
nouveau_pushbuf_bufctx(nvc0->base.pushbuf, nvc0->bufctx);
nouveau_pushbuf_validate(nvc0->base.pushbuf);
for (; dst_layer < dstz + src_box->depth; ++dst_layer, ++src_layer) {
ret = nvc0_2d_texture_do_copy(nvc0->base.pushbuf,
nv50_miptree(dst), dst_level,
dstx, dsty, dst_layer,
nv50_miptree(src), src_level,
src_box->x, src_box->y, src_layer,
src_box->width, src_box->height);
if (ret)
break;
}
nouveau_bufctx_reset(nvc0->bufctx, 0);
}
Bool
NV30EXAPrepareComposite(int op, PicturePtr psPict,
PicturePtr pmPict,
PicturePtr pdPict,
PixmapPtr psPix,
PixmapPtr pmPix,
PixmapPtr pdPix)
{
ScrnInfoPtr pScrn = xf86ScreenToScrn(pdPix->drawable.pScreen);
NVPtr pNv = NVPTR(pScrn);
nv_pict_op_t *blend = NV30_GetPictOpRec(op);
struct nouveau_pushbuf *push = pNv->pushbuf;
uint32_t sc, sa, mc, ma, solid[2];
if (!PUSH_SPACE(push, 128))
return FALSE;
PUSH_RESET(push);
/* setup render target and blending */
if (!NV30_SetupSurface(pScrn, pdPix, pdPict))
return FALSE;
NV30_SetupBlend(pScrn, blend, pdPict->format,
(pmPict && pmPict->componentAlpha &&
PICT_FORMAT_RGB(pmPict->format)));
/* select picture sources */
if (!NV30EXAPicture(pScrn, psPix, psPict, 0, &sc, &sa, &solid[0]))
return FALSE;
if (!NV30EXAPicture(pScrn, pmPix, pmPict, 1, &mc, &ma, &solid[1]))
return FALSE;
/* configure register combiners */
BEGIN_NV04(push, NV30_3D(RC_IN_ALPHA(0)), 6);
PUSH_DATA (push, sa | ma);
if (pmPict &&
pmPict->componentAlpha && PICT_FORMAT_RGB(pmPict->format)) {
if (blend->src_alpha)
PUSH_DATA(push, sa | mc);
else
PUSH_DATA(push, sc | mc);
} else {
PUSH_DATA(push, sc | ma);
}
PUSH_DATA (push, solid[0]);
PUSH_DATA (push, solid[1]);
PUSH_DATA (push, 0x00000c00);
PUSH_DATA (push, 0x00000c00);
BEGIN_NV04(push, NV30_3D(RC_FINAL0), 3);
if (pdPict->format != PICT_a8)
PUSH_DATA (push, 0x0000000c);
else
PUSH_DATA (push, 0x0000001c);
PUSH_DATA (push, 0x00001c00);
PUSH_DATA (push, 0x01000101);
/* select fragprog which just sources textures for combiners */
BEGIN_NV04(push, NV30_3D(FP_ACTIVE_PROGRAM), 1);
PUSH_MTHD (push, NV30_3D(FP_ACTIVE_PROGRAM), pNv->scratch, PFP_PASS,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW |
NOUVEAU_BO_OR,
NV30_3D_FP_ACTIVE_PROGRAM_DMA0,
NV30_3D_FP_ACTIVE_PROGRAM_DMA1);
BEGIN_NV04(push, NV30_3D(FP_REG_CONTROL), 1);
PUSH_DATA (push, 0x0001000f);
BEGIN_NV04(push, NV30_3D(FP_CONTROL), 1);
PUSH_DATA (push, 0x00000000);
BEGIN_NV04(push, NV30_3D(TEX_UNITS_ENABLE), 1);
PUSH_DATA (push, 3);
nouveau_pushbuf_bufctx(push, pNv->bufctx);
if (nouveau_pushbuf_validate(push)) {
nouveau_pushbuf_bufctx(push, NULL);
return FALSE;
}
return TRUE;
}
Bool
NV40EXAPrepareComposite(int op, PicturePtr psPict,
PicturePtr pmPict,
PicturePtr pdPict,
PixmapPtr psPix,
PixmapPtr pmPix,
PixmapPtr pdPix)
{
ScrnInfoPtr pScrn = xf86ScreenToScrn(pdPix->drawable.pScreen);
NVPtr pNv = NVPTR(pScrn);
nv_pict_op_t *blend = NV40_GetPictOpRec(op);
struct nouveau_pushbuf *push = pNv->pushbuf;
uint32_t fragprog;
if (!PUSH_SPACE(push, 128))
NOUVEAU_FALLBACK("space\n");
PUSH_RESET(push);
NV40_SetupBlend(pScrn, blend, pdPict->format,
(pmPict && pmPict->componentAlpha &&
PICT_FORMAT_RGB(pmPict->format)));
if (!NV40_SetupSurface(pScrn, pdPix, pdPict->format) ||
!NV40EXAPicture(pNv, psPix, psPict, 0))
return FALSE;
if (pmPict) {
if (!NV40EXAPicture(pNv, pmPix, pmPict, 1))
return FALSE;
if (pdPict->format == PICT_a8) {
fragprog = PFP_C_A8;
} else
if (pmPict->componentAlpha && PICT_FORMAT_RGB(pmPict->format)) {
if (blend->src_alpha)
fragprog = PFP_CCASA;
else
fragprog = PFP_CCA;
} else {
fragprog = PFP_C;
}
} else {
if (pdPict->format == PICT_a8)
fragprog = PFP_S_A8;
else
fragprog = PFP_S;
}
BEGIN_NV04(push, NV30_3D(FP_ACTIVE_PROGRAM), 1);
PUSH_MTHD (push, NV30_3D(FP_ACTIVE_PROGRAM), pNv->scratch, fragprog,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW |
NOUVEAU_BO_OR,
NV30_3D_FP_ACTIVE_PROGRAM_DMA0,
NV30_3D_FP_ACTIVE_PROGRAM_DMA1);
BEGIN_NV04(push, NV30_3D(FP_CONTROL), 1);
PUSH_DATA (push, 0x02000000);
/* Appears to be some kind of cache flush, needed here at least
* sometimes.. funky text rendering otherwise :)
*/
BEGIN_NV04(push, NV40_3D(TEX_CACHE_CTL), 1);
PUSH_DATA (push, 2);
BEGIN_NV04(push, NV40_3D(TEX_CACHE_CTL), 1);
PUSH_DATA (push, 1);
nouveau_pushbuf_bufctx(push, pNv->bufctx);
if (nouveau_pushbuf_validate(push)) {
nouveau_pushbuf_bufctx(push, NULL);
return FALSE;
}
return TRUE;
}
static void
nvc0_push_upload_vertex_ids(struct push_context *ctx,
struct nvc0_context *nvc0,
const struct pipe_draw_info *info)
{
struct nouveau_pushbuf *push = ctx->push;
struct nouveau_bo *bo;
uint64_t va;
uint32_t *data;
uint32_t format;
unsigned index_size = nvc0->idxbuf.index_size;
unsigned i;
unsigned a = nvc0->vertex->num_elements;
if (!index_size || info->index_bias)
index_size = 4;
data = (uint32_t *)nouveau_scratch_get(&nvc0->base,
info->count * index_size, &va, &bo);
BCTX_REFN_bo(nvc0->bufctx_3d, VTX_TMP, NOUVEAU_BO_GART | NOUVEAU_BO_RD,
bo);
nouveau_pushbuf_validate(push);
if (info->indexed) {
if (!info->index_bias) {
memcpy(data, ctx->idxbuf, info->count * index_size);
} else {
switch (nvc0->idxbuf.index_size) {
case 1:
copy_indices_u8(data, ctx->idxbuf, info->index_bias, info->count);
break;
case 2:
copy_indices_u16(data, ctx->idxbuf, info->index_bias, info->count);
break;
default:
copy_indices_u32(data, ctx->idxbuf, info->index_bias, info->count);
break;
}
}
} else {
for (i = 0; i < info->count; ++i)
data[i] = i + (info->start + info->index_bias);
}
format = (1 << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT) |
NVC0_3D_VERTEX_ATTRIB_FORMAT_TYPE_UINT;
switch (index_size) {
case 1:
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_8;
break;
case 2:
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_16;
break;
default:
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_32;
break;
}
PUSH_SPACE(push, 12);
if (unlikely(nvc0->state.instance_elts & 2)) {
nvc0->state.instance_elts &= ~2;
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(1)), 0);
}
BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(a)), 1);
PUSH_DATA (push, format);
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(1)), 3);
PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | index_size);
PUSH_DATAh(push, va);
PUSH_DATA (push, va);
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(1)), 2);
PUSH_DATAh(push, va + info->count * index_size - 1);
PUSH_DATA (push, va + info->count * index_size - 1);
#define NVC0_3D_VERTEX_ID_REPLACE_SOURCE_ATTR_X(a) \
(((0x80 + (a) * 0x10) / 4) << NVC0_3D_VERTEX_ID_REPLACE_SOURCE__SHIFT)
BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_REPLACE), 1);
PUSH_DATA (push, NVC0_3D_VERTEX_ID_REPLACE_SOURCE_ATTR_X(a) | 1);
}
static void
nve4_m2mf_transfer_rect(struct nvc0_context *nvc0,
const struct nv50_m2mf_rect *dst,
const struct nv50_m2mf_rect *src,
uint32_t nblocksx, uint32_t nblocksy)
{
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nouveau_bufctx *bctx = nvc0->bufctx;
uint32_t exec;
uint32_t src_base = src->base;
uint32_t dst_base = dst->base;
const int cpp = dst->cpp;
assert(dst->cpp == src->cpp);
nouveau_bufctx_refn(bctx, 0, dst->bo, dst->domain | NOUVEAU_BO_WR);
nouveau_bufctx_refn(bctx, 0, src->bo, src->domain | NOUVEAU_BO_RD);
nouveau_pushbuf_bufctx(push, bctx);
nouveau_pushbuf_validate(push);
exec = 0x200 /* 2D_ENABLE */ | 0x6 /* UNK */;
if (!nouveau_bo_memtype(dst->bo)) {
assert(!dst->z);
dst_base += dst->y * dst->pitch + dst->x * cpp;
exec |= 0x100; /* DST_MODE_2D_LINEAR */
}
if (!nouveau_bo_memtype(src->bo)) {
assert(!src->z);
src_base += src->y * src->pitch + src->x * cpp;
exec |= 0x080; /* SRC_MODE_2D_LINEAR */
}
BEGIN_NVC0(push, SUBC_COPY(0x070c), 6);
PUSH_DATA (push, 0x1000 | dst->tile_mode);
PUSH_DATA (push, dst->pitch);
PUSH_DATA (push, dst->height);
PUSH_DATA (push, dst->depth);
PUSH_DATA (push, dst->z);
PUSH_DATA (push, (dst->y << 16) | (dst->x * cpp));
BEGIN_NVC0(push, SUBC_COPY(0x0728), 6);
PUSH_DATA (push, 0x1000 | src->tile_mode);
PUSH_DATA (push, src->pitch);
PUSH_DATA (push, src->height);
PUSH_DATA (push, src->depth);
PUSH_DATA (push, src->z);
PUSH_DATA (push, (src->y << 16) | (src->x * cpp));
BEGIN_NVC0(push, SUBC_COPY(0x0400), 8);
PUSH_DATAh(push, src->bo->offset + src_base);
PUSH_DATA (push, src->bo->offset + src_base);
PUSH_DATAh(push, dst->bo->offset + dst_base);
PUSH_DATA (push, dst->bo->offset + dst_base);
PUSH_DATA (push, src->pitch);
PUSH_DATA (push, dst->pitch);
PUSH_DATA (push, nblocksx * cpp);
PUSH_DATA (push, nblocksy);
BEGIN_NVC0(push, SUBC_COPY(0x0300), 1);
PUSH_DATA (push, exec);
nouveau_bufctx_reset(bctx, 0);
}