static INLINE void
nvc0_program_update_context_state(struct nvc0_context *nvc0,
struct nvc0_program *prog, int stage)
{
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
if (prog && prog->need_tls) {
const uint32_t flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR;
if (!nvc0->state.tls_required)
BCTX_REFN_bo(nvc0->bufctx_3d, TLS, flags, nvc0->screen->tls);
nvc0->state.tls_required |= 1 << stage;
} else {
if (nvc0->state.tls_required == (1 << stage))
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_TLS);
nvc0->state.tls_required &= ~(1 << stage);
}
if (prog && prog->immd_size) {
BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
/* NOTE: may overlap code of a different shader */
PUSH_DATA (push, align(prog->immd_size, 0x100));
PUSH_DATAh(push, nvc0->screen->text->offset + prog->immd_base);
PUSH_DATA (push, nvc0->screen->text->offset + prog->immd_base);
BEGIN_NVC0(push, NVC0_3D(CB_BIND(stage)), 1);
PUSH_DATA (push, (14 << 4) | 1);
nvc0->state.c14_bound |= 1 << stage;
} else
if (nvc0->state.c14_bound & (1 << stage)) {
BEGIN_NVC0(push, NVC0_3D(CB_BIND(stage)), 1);
PUSH_DATA (push, (14 << 4) | 0);
nvc0->state.c14_bound &= ~(1 << stage);
}
}
static INLINE void
nv50_program_update_context_state(struct nv50_context *nv50,
struct nv50_program *prog, int stage)
{
const unsigned flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR;
if (prog && prog->tls_space) {
if (nv50->state.new_tls_space)
nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_TLS);
if (!nv50->state.tls_required || nv50->state.new_tls_space)
BCTX_REFN_bo(nv50->bufctx_3d, TLS, flags, nv50->screen->tls_bo);
nv50->state.new_tls_space = FALSE;
nv50->state.tls_required |= 1 << stage;
} else {
if (nv50->state.tls_required == (1 << stage))
nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_TLS);
nv50->state.tls_required &= ~(1 << stage);
}
}
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;
}
struct pipe_context *
nv50_create(struct pipe_screen *pscreen, void *priv)
{
struct nv50_screen *screen = nv50_screen(pscreen);
struct nv50_context *nv50;
struct pipe_context *pipe;
int ret;
uint32_t flags;
nv50 = CALLOC_STRUCT(nv50_context);
if (!nv50)
return NULL;
pipe = &nv50->base.pipe;
if (!nv50_blitctx_create(nv50))
goto out_err;
nv50->base.pushbuf = screen->base.pushbuf;
nv50->base.client = screen->base.client;
ret = nouveau_bufctx_new(screen->base.client, NV50_BIND_COUNT,
&nv50->bufctx_3d);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, 2, &nv50->bufctx);
if (ret)
goto out_err;
nv50->base.screen = &screen->base;
nv50->base.copy_data = nv50_m2mf_copy_linear;
nv50->base.push_data = nv50_sifc_linear_u8;
nv50->base.push_cb = nv50_cb_push;
nv50->screen = screen;
pipe->screen = pscreen;
pipe->priv = priv;
pipe->destroy = nv50_destroy;
pipe->draw_vbo = nv50_draw_vbo;
pipe->clear = nv50_clear;
pipe->flush = nv50_flush;
pipe->texture_barrier = nv50_texture_barrier;
if (!screen->cur_ctx) {
screen->cur_ctx = nv50;
nouveau_pushbuf_bufctx(screen->base.pushbuf, nv50->bufctx);
}
nv50_init_query_functions(nv50);
nv50_init_surface_functions(nv50);
nv50_init_state_functions(nv50);
nv50_init_resource_functions(pipe);
nv50->base.invalidate_resource_storage = nv50_invalidate_resource_storage;
#ifdef NV50_WITH_DRAW_MODULE
/* no software fallbacks implemented */
nv50->draw = draw_create(pipe);
assert(nv50->draw);
draw_set_rasterize_stage(nv50->draw, nv50_draw_render_stage(nv50));
#endif
nouveau_context_init_vdec(&nv50->base);
flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD;
BCTX_REFN_bo(nv50->bufctx_3d, SCREEN, flags, screen->code);
BCTX_REFN_bo(nv50->bufctx_3d, SCREEN, flags, screen->uniforms);
BCTX_REFN_bo(nv50->bufctx_3d, SCREEN, flags, screen->txc);
BCTX_REFN_bo(nv50->bufctx_3d, SCREEN, flags, screen->stack_bo);
flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
BCTX_REFN_bo(nv50->bufctx_3d, SCREEN, flags, screen->fence.bo);
BCTX_REFN_bo(nv50->bufctx, FENCE, flags, screen->fence.bo);
nv50->base.scratch.bo_size = 2 << 20;
return pipe;
out_err:
if (nv50) {
if (nv50->bufctx_3d)
nouveau_bufctx_del(&nv50->bufctx_3d);
if (nv50->bufctx)
nouveau_bufctx_del(&nv50->bufctx);
if (nv50->blit)
FREE(nv50->blit);
FREE(nv50);
}
return NULL;
}
struct pipe_context *
nvc0_create(struct pipe_screen *pscreen, void *priv, unsigned ctxflags)
{
struct nvc0_screen *screen = nvc0_screen(pscreen);
struct nvc0_context *nvc0;
struct pipe_context *pipe;
int ret;
uint32_t flags;
nvc0 = CALLOC_STRUCT(nvc0_context);
if (!nvc0)
return NULL;
pipe = &nvc0->base.pipe;
if (!nvc0_blitctx_create(nvc0))
goto out_err;
nvc0->base.pushbuf = screen->base.pushbuf;
nvc0->base.client = screen->base.client;
ret = nouveau_bufctx_new(screen->base.client, 2, &nvc0->bufctx);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_3D_COUNT,
&nvc0->bufctx_3d);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_CP_COUNT,
&nvc0->bufctx_cp);
if (ret)
goto out_err;
nvc0->screen = screen;
nvc0->base.screen = &screen->base;
pipe->screen = pscreen;
pipe->priv = priv;
pipe->destroy = nvc0_destroy;
pipe->draw_vbo = nvc0_draw_vbo;
pipe->clear = nvc0_clear;
pipe->launch_grid = (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) ?
nve4_launch_grid : nvc0_launch_grid;
pipe->flush = nvc0_flush;
pipe->texture_barrier = nvc0_texture_barrier;
pipe->memory_barrier = nvc0_memory_barrier;
pipe->get_sample_position = nvc0_context_get_sample_position;
nouveau_context_init(&nvc0->base);
nvc0_init_query_functions(nvc0);
nvc0_init_surface_functions(nvc0);
nvc0_init_state_functions(nvc0);
nvc0_init_transfer_functions(nvc0);
nvc0_init_resource_functions(pipe);
nvc0->base.invalidate_resource_storage = nvc0_invalidate_resource_storage;
pipe->create_video_codec = nvc0_create_decoder;
pipe->create_video_buffer = nvc0_video_buffer_create;
/* shader builtin library is per-screen, but we need a context for m2mf */
nvc0_program_library_upload(nvc0);
nvc0_program_init_tcp_empty(nvc0);
if (!nvc0->tcp_empty)
goto out_err;
/* set the empty tctl prog on next draw in case one is never set */
nvc0->dirty |= NVC0_NEW_TCTLPROG;
/* now that there are no more opportunities for errors, set the current
* context if there isn't already one.
*/
if (!screen->cur_ctx) {
nvc0->state = screen->save_state;
screen->cur_ctx = nvc0;
nouveau_pushbuf_bufctx(screen->base.pushbuf, nvc0->bufctx);
}
screen->base.pushbuf->kick_notify = nvc0_default_kick_notify;
/* add permanently resident buffers to bufctxts */
flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD;
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->text);
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->uniform_bo);
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->txc);
if (screen->compute) {
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->text);
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->txc);
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->parm);
}
flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR;
if (screen->poly_cache)
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->poly_cache);
if (screen->compute)
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->tls);
flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->fence.bo);
BCTX_REFN_bo(nvc0->bufctx, FENCE, flags, screen->fence.bo);
if (screen->compute)
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->fence.bo);
nvc0->base.scratch.bo_size = 2 << 20;
memset(nvc0->tex_handles, ~0, sizeof(nvc0->tex_handles));
util_dynarray_init(&nvc0->global_residents);
return pipe;
out_err:
if (nvc0) {
if (nvc0->bufctx_3d)
nouveau_bufctx_del(&nvc0->bufctx_3d);
if (nvc0->bufctx_cp)
nouveau_bufctx_del(&nvc0->bufctx_cp);
if (nvc0->bufctx)
nouveau_bufctx_del(&nvc0->bufctx);
FREE(nvc0->blit);
FREE(nvc0);
}
return NULL;
}
struct pipe_context *
nvc0_create(struct pipe_screen *pscreen, void *priv)
{
struct nvc0_screen *screen = nvc0_screen(pscreen);
struct nvc0_context *nvc0;
struct pipe_context *pipe;
int ret;
uint32_t flags;
nvc0 = CALLOC_STRUCT(nvc0_context);
if (!nvc0)
return NULL;
pipe = &nvc0->base.pipe;
if (!nvc0_blitctx_create(nvc0))
goto out_err;
nvc0->base.pushbuf = screen->base.pushbuf;
nvc0->base.client = screen->base.client;
ret = nouveau_bufctx_new(screen->base.client, 2, &nvc0->bufctx);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_3D_COUNT,
&nvc0->bufctx_3d);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_CP_COUNT,
&nvc0->bufctx_cp);
if (ret)
goto out_err;
nvc0->screen = screen;
nvc0->base.screen = &screen->base;
pipe->screen = pscreen;
pipe->priv = priv;
pipe->destroy = nvc0_destroy;
pipe->draw_vbo = nvc0_draw_vbo;
pipe->clear = nvc0_clear;
pipe->launch_grid = (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) ?
nve4_launch_grid : nvc0_launch_grid;
pipe->flush = nvc0_flush;
pipe->texture_barrier = nvc0_texture_barrier;
pipe->memory_barrier = nvc0_memory_barrier;
pipe->get_sample_position = nvc0_context_get_sample_position;
if (!screen->cur_ctx) {
screen->cur_ctx = nvc0;
nouveau_pushbuf_bufctx(screen->base.pushbuf, nvc0->bufctx);
}
screen->base.pushbuf->kick_notify = nvc0_default_kick_notify;
nvc0_init_query_functions(nvc0);
nvc0_init_surface_functions(nvc0);
nvc0_init_state_functions(nvc0);
nvc0_init_transfer_functions(nvc0);
nvc0_init_resource_functions(pipe);
nvc0->base.invalidate_resource_storage = nvc0_invalidate_resource_storage;
#ifdef NVC0_WITH_DRAW_MODULE
/* no software fallbacks implemented */
nvc0->draw = draw_create(pipe);
assert(nvc0->draw);
draw_set_rasterize_stage(nvc0->draw, nvc0_draw_render_stage(nvc0));
#endif
pipe->create_video_codec = nvc0_create_decoder;
pipe->create_video_buffer = nvc0_video_buffer_create;
/* shader builtin library is per-screen, but we need a context for m2mf */
nvc0_program_library_upload(nvc0);
/* add permanently resident buffers to bufctxts */
flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD;
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->text);
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->uniform_bo);
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->txc);
if (screen->compute) {
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->text);
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->txc);
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->parm);
}
flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR;
if (screen->poly_cache)
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->poly_cache);
if (screen->compute)
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->tls);
flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->fence.bo);
BCTX_REFN_bo(nvc0->bufctx, FENCE, flags, screen->fence.bo);
if (screen->compute)
BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->fence.bo);
nvc0->base.scratch.bo_size = 2 << 20;
memset(nvc0->tex_handles, ~0, sizeof(nvc0->tex_handles));
util_dynarray_init(&nvc0->global_residents);
return pipe;
out_err:
if (nvc0) {
if (nvc0->bufctx_3d)
nouveau_bufctx_del(&nvc0->bufctx_3d);
if (nvc0->bufctx_cp)
nouveau_bufctx_del(&nvc0->bufctx_cp);
if (nvc0->bufctx)
nouveau_bufctx_del(&nvc0->bufctx);
if (nvc0->blit)
FREE(nvc0->blit);
FREE(nvc0);
}
return NULL;
}
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);
}
struct pipe_context *
nv50_create(struct pipe_screen *pscreen, void *priv, unsigned ctxflags)
{
struct nv50_screen *screen = nv50_screen(pscreen);
struct nv50_context *nv50;
struct pipe_context *pipe;
int ret;
uint32_t flags;
nv50 = CALLOC_STRUCT(nv50_context);
if (!nv50)
return NULL;
pipe = &nv50->base.pipe;
if (!nv50_blitctx_create(nv50))
goto out_err;
nv50->base.pushbuf = screen->base.pushbuf;
nv50->base.client = screen->base.client;
ret = nouveau_bufctx_new(screen->base.client, 2, &nv50->bufctx);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, NV50_BIND_3D_COUNT,
&nv50->bufctx_3d);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, NV50_BIND_CP_COUNT,
&nv50->bufctx_cp);
if (ret)
goto out_err;
nv50->base.screen = &screen->base;
nv50->base.copy_data = nv50_m2mf_copy_linear;
nv50->base.push_data = nv50_sifc_linear_u8;
/* FIXME: Make it possible to use this again. The problem is that there is
* some clever logic in the card that allows for multiple renders to happen
* when there are only constbuf changes. However that relies on the
* constbuf updates happening to the right constbuf slots. Currently
* implementation just makes it go through a separate slot which doesn't
* properly update the right constbuf data.
nv50->base.push_cb = nv50_cb_push;
*/
nv50->screen = screen;
pipe->screen = pscreen;
pipe->priv = priv;
pipe->destroy = nv50_destroy;
pipe->draw_vbo = nv50_draw_vbo;
pipe->clear = nv50_clear;
pipe->launch_grid = nv50_launch_grid;
pipe->flush = nv50_flush;
pipe->texture_barrier = nv50_texture_barrier;
pipe->memory_barrier = nv50_memory_barrier;
pipe->get_sample_position = nv50_context_get_sample_position;
if (!screen->cur_ctx) {
/* Restore the last context's state here, normally handled during
* context switch
*/
nv50->state = screen->save_state;
screen->cur_ctx = nv50;
nouveau_pushbuf_bufctx(screen->base.pushbuf, nv50->bufctx);
}
nv50->base.pushbuf->kick_notify = nv50_default_kick_notify;
nouveau_context_init(&nv50->base);
nv50_init_query_functions(nv50);
nv50_init_surface_functions(nv50);
nv50_init_state_functions(nv50);
nv50_init_resource_functions(pipe);
nv50->base.invalidate_resource_storage = nv50_invalidate_resource_storage;
if (screen->base.device->chipset < 0x84 ||
debug_get_bool_option("NOUVEAU_PMPEG", false)) {
/* PMPEG */
nouveau_context_init_vdec(&nv50->base);
} else if (screen->base.device->chipset < 0x98 ||
screen->base.device->chipset == 0xa0) {
/* VP2 */
pipe->create_video_codec = nv84_create_decoder;
pipe->create_video_buffer = nv84_video_buffer_create;
} else {
/* VP3/4 */
pipe->create_video_codec = nv98_create_decoder;
pipe->create_video_buffer = nv98_video_buffer_create;
}
flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD;
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->code);
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->uniforms);
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->txc);
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->stack_bo);
if (screen->compute) {
BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->code);
BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->txc);
BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->stack_bo);
}
flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->fence.bo);
BCTX_REFN_bo(nv50->bufctx, FENCE, flags, screen->fence.bo);
if (screen->compute)
BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->fence.bo);
nv50->base.scratch.bo_size = 2 << 20;
util_dynarray_init(&nv50->global_residents);
return pipe;
out_err:
if (nv50->bufctx_3d)
nouveau_bufctx_del(&nv50->bufctx_3d);
if (nv50->bufctx_cp)
nouveau_bufctx_del(&nv50->bufctx_cp);
if (nv50->bufctx)
nouveau_bufctx_del(&nv50->bufctx);
FREE(nv50->blit);
FREE(nv50);
return NULL;
}
static void
nv50_hw_sm_end_query(struct nv50_context *nv50, struct nv50_hw_query *hq)
{
struct nv50_screen *screen = nv50->screen;
struct pipe_context *pipe = &nv50->base.pipe;
struct nouveau_pushbuf *push = nv50->base.pushbuf;
struct nv50_hw_sm_query *hsq = nv50_hw_sm_query(hq);
struct pipe_grid_info info = {};
uint32_t mask;
uint32_t input[3];
const uint block[3] = { 32, 1, 1 };
const uint grid[3] = { screen->MPsInTP, screen->TPs, 1 };
int c, i;
if (unlikely(!screen->pm.prog)) {
struct nv50_program *prog = CALLOC_STRUCT(nv50_program);
prog->type = PIPE_SHADER_COMPUTE;
prog->translated = true;
prog->max_gpr = 7;
prog->parm_size = 8;
prog->code = (uint32_t *)nv50_read_hw_sm_counters_code;
prog->code_size = sizeof(nv50_read_hw_sm_counters_code);
screen->pm.prog = prog;
}
/* disable all counting */
PUSH_SPACE(push, 8);
for (c = 0; c < 4; c++) {
if (screen->pm.mp_counter[c]) {
BEGIN_NV04(push, NV50_COMPUTE(MP_PM_CONTROL(c)), 1);
PUSH_DATA (push, 0);
}
}
/* release counters for this query */
for (c = 0; c < 4; c++) {
if (screen->pm.mp_counter[c] == hsq) {
screen->pm.num_hw_sm_active--;
screen->pm.mp_counter[c] = NULL;
}
}
BCTX_REFN_bo(nv50->bufctx_cp, CP_QUERY, NOUVEAU_BO_GART | NOUVEAU_BO_WR,
hq->bo);
PUSH_SPACE(push, 2);
BEGIN_NV04(push, SUBC_COMPUTE(NV50_GRAPH_SERIALIZE), 1);
PUSH_DATA (push, 0);
pipe->bind_compute_state(pipe, screen->pm.prog);
input[0] = hq->bo->offset + hq->base_offset;
input[1] = hq->sequence;
for (i = 0; i < 3; i++) {
info.block[i] = block[i];
info.grid[i] = grid[i];
}
info.pc = 0;
info.input = input;
pipe->launch_grid(pipe, &info);
nouveau_bufctx_reset(nv50->bufctx_cp, NV50_BIND_CP_QUERY);
/* re-active other counters */
PUSH_SPACE(push, 8);
mask = 0;
for (c = 0; c < 4; c++) {
const struct nv50_hw_sm_query_cfg *cfg;
unsigned i;
hsq = screen->pm.mp_counter[c];
if (!hsq)
continue;
cfg = nv50_hw_sm_query_get_cfg(nv50, &hsq->base);
for (i = 0; i < cfg->num_counters; i++) {
uint16_t func;
if (mask & (1 << hsq->ctr[i]))
break;
mask |= 1 << hsq->ctr[i];
func = nv50_hw_sm_get_func(hsq->ctr[i]);
BEGIN_NV04(push, NV50_COMPUTE(MP_PM_CONTROL(hsq->ctr[i])), 1);
PUSH_DATA (push, (cfg->ctr[i].sig << 24) | (func << 8)
| cfg->ctr[i].unit | cfg->ctr[i].mode);
}
}
}
static void
nve4_mp_pm_query_end(struct nvc0_context *nvc0, struct nvc0_query *q)
{
struct nvc0_screen *screen = nvc0->screen;
struct pipe_context *pipe = &nvc0->base.pipe;
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
uint32_t mask;
uint32_t input[3];
const uint block[3] = { 32, 4, 1 };
const uint grid[3] = { screen->mp_count, 1, 1 };
unsigned c;
const struct nve4_mp_pm_query_cfg *cfg;
cfg = &nve4_mp_pm_queries[q->type - PIPE_QUERY_DRIVER_SPECIFIC];
if (unlikely(!screen->pm.prog)) {
struct nvc0_program *prog = CALLOC_STRUCT(nvc0_program);
prog->type = PIPE_SHADER_COMPUTE;
prog->translated = TRUE;
prog->num_gprs = 14;
prog->code = (uint32_t *)nve4_read_mp_pm_counters_code;
prog->code_size = sizeof(nve4_read_mp_pm_counters_code);
prog->parm_size = 12;
screen->pm.prog = prog;
}
/* disable all counting */
PUSH_SPACE(push, 8);
for (c = 0; c < 8; ++c)
if (screen->pm.mp_counter[c])
IMMED_NVC0(push, NVE4_COMPUTE(MP_PM_FUNC(c)), 0);
/* release counters for this query */
for (c = 0; c < 8; ++c) {
if (nvc0_query(screen->pm.mp_counter[c]) == q) {
screen->pm.num_mp_pm_active[c / 4]--;
screen->pm.mp_counter[c] = NULL;
}
}
BCTX_REFN_bo(nvc0->bufctx_cp, CP_QUERY, NOUVEAU_BO_GART | NOUVEAU_BO_WR,
q->bo);
PUSH_SPACE(push, 1);
IMMED_NVC0(push, SUBC_COMPUTE(NV50_GRAPH_SERIALIZE), 0);
pipe->bind_compute_state(pipe, screen->pm.prog);
input[0] = (q->bo->offset + q->base);
input[1] = (q->bo->offset + q->base) >> 32;
input[2] = q->sequence;
pipe->launch_grid(pipe, block, grid, 0, input);
nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_QUERY);
/* re-activate other counters */
PUSH_SPACE(push, 16);
mask = 0;
for (c = 0; c < 8; ++c) {
unsigned i;
q = nvc0_query(screen->pm.mp_counter[c]);
if (!q)
continue;
cfg = &nve4_mp_pm_queries[q->type - PIPE_QUERY_DRIVER_SPECIFIC];
for (i = 0; i < cfg->num_counters; ++i) {
if (mask & (1 << q->ctr[i]))
break;
mask |= 1 << q->ctr[i];
BEGIN_NVC0(push, NVE4_COMPUTE(MP_PM_FUNC(q->ctr[i])), 1);
PUSH_DATA (push, (cfg->ctr[i].func << 4) | cfg->ctr[i].mode);
}
}
}
struct pipe_context *
nvc0_create(struct pipe_screen *pscreen, void *priv)
{
struct nvc0_screen *screen = nvc0_screen(pscreen);
struct nvc0_context *nvc0;
struct pipe_context *pipe;
int ret;
uint32_t flags;
nvc0 = CALLOC_STRUCT(nvc0_context);
if (!nvc0)
return NULL;
pipe = &nvc0->base.pipe;
nvc0->base.pushbuf = screen->base.pushbuf;
ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_COUNT,
&nvc0->bufctx_3d);
if (!ret)
nouveau_bufctx_new(screen->base.client, 2, &nvc0->bufctx);
if (ret)
goto out_err;
nvc0->screen = screen;
nvc0->base.screen = &screen->base;
pipe->screen = pscreen;
pipe->priv = priv;
pipe->destroy = nvc0_destroy;
pipe->draw_vbo = nvc0_draw_vbo;
pipe->clear = nvc0_clear;
pipe->flush = nvc0_flush;
pipe->texture_barrier = nvc0_texture_barrier;
if (!screen->cur_ctx) {
screen->cur_ctx = nvc0;
nouveau_pushbuf_bufctx(screen->base.pushbuf, nvc0->bufctx);
}
screen->base.pushbuf->kick_notify = nvc0_default_kick_notify;
nvc0_init_query_functions(nvc0);
nvc0_init_surface_functions(nvc0);
nvc0_init_state_functions(nvc0);
nvc0_init_transfer_functions(nvc0);
nvc0_init_resource_functions(pipe);
#ifdef NVC0_WITH_DRAW_MODULE
/* no software fallbacks implemented */
nvc0->draw = draw_create(pipe);
assert(nvc0->draw);
draw_set_rasterize_stage(nvc0->draw, nvc0_draw_render_stage(nvc0));
#endif
nouveau_context_init_vdec(&nvc0->base);
/* shader builtin library is per-screen, but we need a context for m2mf */
nvc0_program_library_upload(nvc0);
/* add permanently resident buffers to bufctxts */
flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD;
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->text);
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->uniform_bo);
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->txc);
flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
BCTX_REFN_bo(nvc0->bufctx_3d, SCREEN, flags, screen->fence.bo);
BCTX_REFN_bo(nvc0->bufctx, FENCE, flags, screen->fence.bo);
nvc0->base.scratch.bo_size = 2 << 20;
return pipe;
out_err:
if (nvc0) {
if (nvc0->bufctx_3d)
nouveau_bufctx_del(&nvc0->bufctx_3d);
if (nvc0->bufctx)
nouveau_bufctx_del(&nvc0->bufctx);
FREE(nvc0);
}
return NULL;
}
struct pipe_context *
nv50_create(struct pipe_screen *pscreen, void *priv, unsigned ctxflags)
{
struct nv50_screen *screen = nv50_screen(pscreen);
struct nv50_context *nv50;
struct pipe_context *pipe;
int ret;
uint32_t flags;
nv50 = CALLOC_STRUCT(nv50_context);
if (!nv50)
return NULL;
pipe = &nv50->base.pipe;
if (!nv50_blitctx_create(nv50))
goto out_err;
nv50->base.pushbuf = screen->base.pushbuf;
nv50->base.client = screen->base.client;
ret = nouveau_bufctx_new(screen->base.client, 2, &nv50->bufctx);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, NV50_BIND_3D_COUNT,
&nv50->bufctx_3d);
if (!ret)
ret = nouveau_bufctx_new(screen->base.client, NV50_BIND_CP_COUNT,
&nv50->bufctx_cp);
if (ret)
goto out_err;
nv50->base.screen = &screen->base;
nv50->base.copy_data = nv50_m2mf_copy_linear;
nv50->base.push_data = nv50_sifc_linear_u8;
nv50->base.push_cb = nv50_cb_push;
nv50->screen = screen;
pipe->screen = pscreen;
pipe->priv = priv;
pipe->destroy = nv50_destroy;
pipe->draw_vbo = nv50_draw_vbo;
pipe->clear = nv50_clear;
pipe->launch_grid = nv50_launch_grid;
pipe->flush = nv50_flush;
pipe->texture_barrier = nv50_texture_barrier;
pipe->memory_barrier = nv50_memory_barrier;
pipe->get_sample_position = nv50_context_get_sample_position;
pipe->emit_string_marker = nv50_emit_string_marker;
if (!screen->cur_ctx) {
/* Restore the last context's state here, normally handled during
* context switch
*/
nv50->state = screen->save_state;
screen->cur_ctx = nv50;
nouveau_pushbuf_bufctx(screen->base.pushbuf, nv50->bufctx);
}
nv50->base.pushbuf->kick_notify = nv50_default_kick_notify;
nouveau_context_init(&nv50->base);
nv50_init_query_functions(nv50);
nv50_init_surface_functions(nv50);
nv50_init_state_functions(nv50);
nv50_init_resource_functions(pipe);
nv50->base.invalidate_resource_storage = nv50_invalidate_resource_storage;
if (screen->base.device->chipset < 0x84 ||
debug_get_bool_option("NOUVEAU_PMPEG", false)) {
/* PMPEG */
nouveau_context_init_vdec(&nv50->base);
} else if (screen->base.device->chipset < 0x98 ||
screen->base.device->chipset == 0xa0) {
/* VP2 */
pipe->create_video_codec = nv84_create_decoder;
pipe->create_video_buffer = nv84_video_buffer_create;
} else {
/* VP3/4 */
pipe->create_video_codec = nv98_create_decoder;
pipe->create_video_buffer = nv98_video_buffer_create;
}
flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RD;
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->code);
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->uniforms);
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->txc);
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->stack_bo);
if (screen->compute) {
BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->code);
BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->txc);
BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->stack_bo);
}
flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
BCTX_REFN_bo(nv50->bufctx_3d, 3D_SCREEN, flags, screen->fence.bo);
BCTX_REFN_bo(nv50->bufctx, FENCE, flags, screen->fence.bo);
if (screen->compute)
BCTX_REFN_bo(nv50->bufctx_cp, CP_SCREEN, flags, screen->fence.bo);
nv50->base.scratch.bo_size = 2 << 20;
util_dynarray_init(&nv50->global_residents);
return pipe;
out_err:
if (nv50->bufctx_3d)
nouveau_bufctx_del(&nv50->bufctx_3d);
if (nv50->bufctx_cp)
nouveau_bufctx_del(&nv50->bufctx_cp);
if (nv50->bufctx)
nouveau_bufctx_del(&nv50->bufctx);
FREE(nv50->blit);
FREE(nv50);
return NULL;
}
