int
nvc0_screen_compute_setup(struct nvc0_screen *screen,
struct nouveau_pushbuf *push)
{
struct nouveau_object *chan = screen->base.channel;
struct nouveau_device *dev = screen->base.device;
uint32_t obj_class;
int ret;
int i;
switch (dev->chipset & ~0xf) {
case 0xc0:
case 0xd0:
/* In theory, GF110+ should also support NVC8_COMPUTE_CLASS but,
* in practice, a ILLEGAL_CLASS dmesg fail appears when using it. */
obj_class = NVC0_COMPUTE_CLASS;
break;
default:
NOUVEAU_ERR("unsupported chipset: NV%02x\n", dev->chipset);
return -1;
}
ret = nouveau_object_new(chan, 0xbeef90c0, obj_class, NULL, 0,
&screen->compute);
if (ret) {
NOUVEAU_ERR("Failed to allocate compute object: %d\n", ret);
return ret;
}
ret = nouveau_bo_new(dev, NV_VRAM_DOMAIN(&screen->base), 0, 1 << 12, NULL,
&screen->parm);
if (ret)
return ret;
BEGIN_NVC0(push, SUBC_COMPUTE(NV01_SUBCHAN_OBJECT), 1);
PUSH_DATA (push, screen->compute->oclass);
/* hardware limit */
BEGIN_NVC0(push, NVC0_COMPUTE(MP_LIMIT), 1);
PUSH_DATA (push, screen->mp_count);
BEGIN_NVC0(push, NVC0_COMPUTE(CALL_LIMIT_LOG), 1);
PUSH_DATA (push, 0xf);
BEGIN_NVC0(push, SUBC_COMPUTE(0x02a0), 1);
PUSH_DATA (push, 0x8000);
/* global memory setup */
BEGIN_NVC0(push, SUBC_COMPUTE(0x02c4), 1);
PUSH_DATA (push, 0);
BEGIN_NIC0(push, NVC0_COMPUTE(GLOBAL_BASE), 0x100);
for (i = 0; i <= 0xff; i++)
PUSH_DATA (push, (0xc << 28) | (i << 16) | i);
BEGIN_NVC0(push, SUBC_COMPUTE(0x02c4), 1);
PUSH_DATA (push, 1);
/* local memory and cstack setup */
BEGIN_NVC0(push, NVC0_COMPUTE(TEMP_ADDRESS_HIGH), 2);
PUSH_DATAh(push, screen->tls->offset);
PUSH_DATA (push, screen->tls->offset);
BEGIN_NVC0(push, NVC0_COMPUTE(TEMP_SIZE_HIGH), 2);
PUSH_DATAh(push, screen->tls->size);
PUSH_DATA (push, screen->tls->size);
BEGIN_NVC0(push, NVC0_COMPUTE(WARP_TEMP_ALLOC), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, NVC0_COMPUTE(LOCAL_BASE), 1);
PUSH_DATA (push, 1 << 24);
/* shared memory setup */
BEGIN_NVC0(push, NVC0_COMPUTE(CACHE_SPLIT), 1);
PUSH_DATA (push, NVC0_COMPUTE_CACHE_SPLIT_48K_SHARED_16K_L1);
BEGIN_NVC0(push, NVC0_COMPUTE(SHARED_BASE), 1);
PUSH_DATA (push, 2 << 24);
BEGIN_NVC0(push, NVC0_COMPUTE(SHARED_SIZE), 1);
PUSH_DATA (push, 0);
/* code segment setup */
BEGIN_NVC0(push, NVC0_COMPUTE(CODE_ADDRESS_HIGH), 2);
PUSH_DATAh(push, screen->text->offset);
PUSH_DATA (push, screen->text->offset);
/* TODO: textures & samplers */
return 0;
}
struct pipe_screen *
nv30_screen_create(struct nouveau_device *dev)
{
struct nv30_screen *screen = CALLOC_STRUCT(nv30_screen);
struct pipe_screen *pscreen;
struct nouveau_pushbuf *push;
struct nv04_fifo *fifo;
unsigned oclass = 0;
int ret, i;
if (!screen)
return NULL;
switch (dev->chipset & 0xf0) {
case 0x30:
if (RANKINE_0397_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV30_3D_CLASS;
else
if (RANKINE_0697_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV34_3D_CLASS;
else
if (RANKINE_0497_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV35_3D_CLASS;
break;
case 0x40:
if (CURIE_4097_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV40_3D_CLASS;
else
if (CURIE_4497_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV44_3D_CLASS;
break;
case 0x60:
if (CURIE_4497_CHIPSET6X & (1 << (dev->chipset & 0x0f)))
oclass = NV44_3D_CLASS;
break;
default:
break;
}
if (!oclass) {
NOUVEAU_ERR("unknown 3d class for 0x%02x\n", dev->chipset);
FREE(screen);
return NULL;
}
/*
* Some modern apps try to use msaa without keeping in mind the
* restrictions on videomem of older cards. Resulting in dmesg saying:
* [ 1197.850642] nouveau E[soffice.bin[3785]] fail ttm_validate
* [ 1197.850648] nouveau E[soffice.bin[3785]] validating bo list
* [ 1197.850654] nouveau E[soffice.bin[3785]] validate: -12
*
* Because we are running out of video memory, after which the program
* using the msaa visual freezes, and eventually the entire system freezes.
*
* To work around this we do not allow msaa visauls by default and allow
* the user to override this via NV30_MAX_MSAA.
*/
screen->max_sample_count = debug_get_num_option("NV30_MAX_MSAA", 0);
if (screen->max_sample_count > 4)
screen->max_sample_count = 4;
pscreen = &screen->base.base;
pscreen->destroy = nv30_screen_destroy;
pscreen->get_param = nv30_screen_get_param;
pscreen->get_paramf = nv30_screen_get_paramf;
pscreen->get_shader_param = nv30_screen_get_shader_param;
pscreen->context_create = nv30_context_create;
pscreen->is_format_supported = nv30_screen_is_format_supported;
nv30_resource_screen_init(pscreen);
nouveau_screen_init_vdec(&screen->base);
screen->base.fence.emit = nv30_screen_fence_emit;
screen->base.fence.update = nv30_screen_fence_update;
ret = nouveau_screen_init(&screen->base, dev);
if (ret)
FAIL_SCREEN_INIT("nv30_screen_init failed: %d\n", ret);
screen->base.vidmem_bindings |= PIPE_BIND_VERTEX_BUFFER;
screen->base.sysmem_bindings |= PIPE_BIND_VERTEX_BUFFER;
if (oclass == NV40_3D_CLASS) {
screen->base.vidmem_bindings |= PIPE_BIND_INDEX_BUFFER;
screen->base.sysmem_bindings |= PIPE_BIND_INDEX_BUFFER;
}
fifo = screen->base.channel->data;
push = screen->base.pushbuf;
push->rsvd_kick = 16;
ret = nouveau_object_new(screen->base.channel, 0x00000000, NV01_NULL_CLASS,
NULL, 0, &screen->null);
if (ret)
FAIL_SCREEN_INIT("error allocating null object: %d\n", ret);
/* DMA_FENCE refuses to accept DMA objects with "adjust" filled in,
* this means that the address pointed at by the DMA object must
* be 4KiB aligned, which means this object needs to be the first
* one allocated on the channel.
*/
ret = nouveau_object_new(screen->base.channel, 0xbeef1e00,
NOUVEAU_NOTIFIER_CLASS, &(struct nv04_notify) {
.length = 32 }, sizeof(struct nv04_notify),
void
nve4_set_surface_info(struct nouveau_pushbuf *push,
struct pipe_surface *psf,
struct nvc0_screen *screen)
{
struct nv50_surface *sf = nv50_surface(psf);
struct nv04_resource *res;
uint64_t address;
uint32_t *const info = push->cur;
uint8_t log2cpp;
if (psf && !nve4_su_format_map[psf->format])
NOUVEAU_ERR("unsupported surface format, try is_format_supported() !\n");
push->cur += 16;
if (!psf || !nve4_su_format_map[psf->format]) {
memset(info, 0, 16 * sizeof(*info));
info[0] = 0xbadf0000;
info[1] = 0x80004000;
info[12] = nve4_suldp_lib_offset[PIPE_FORMAT_R32G32B32A32_UINT] +
screen->lib_code->start;
return;
}
res = nv04_resource(sf->base.texture);
address = res->address + sf->offset;
info[8] = sf->width;
info[9] = sf->height;
info[10] = sf->depth;
switch (res->base.target) {
case PIPE_TEXTURE_1D_ARRAY:
info[11] = 1;
break;
case PIPE_TEXTURE_2D:
case PIPE_TEXTURE_RECT:
info[11] = 2;
break;
case PIPE_TEXTURE_3D:
info[11] = 3;
break;
case PIPE_TEXTURE_2D_ARRAY:
case PIPE_TEXTURE_CUBE:
case PIPE_TEXTURE_CUBE_ARRAY:
info[11] = 4;
break;
default:
info[11] = 0;
break;
}
log2cpp = (0xf000 & nve4_su_format_aux_map[sf->base.format]) >> 12;
info[12] = nve4_suldp_lib_offset[sf->base.format] + screen->lib_code->start;
/* limit in bytes for raw access */
info[13] = (0x06 << 22) | ((sf->width << log2cpp) - 1);
info[1] = nve4_su_format_map[sf->base.format];
#if 0
switch (util_format_get_blocksizebits(sf->base.format)) {
case 16: info[1] |= 1 << 16; break;
case 32: info[1] |= 2 << 16; break;
case 64: info[1] |= 3 << 16; break;
case 128: info[1] |= 4 << 16; break;
default:
break;
}
#else
info[1] |= log2cpp << 16;
info[1] |= 0x4000;
info[1] |= (0x0f00 & nve4_su_format_aux_map[sf->base.format]);
#endif
if (res->base.target == PIPE_BUFFER) {
info[0] = address >> 8;
info[2] = sf->width - 1;
info[2] |= (0xff & nve4_su_format_aux_map[sf->base.format]) << 22;
info[3] = 0;
info[4] = 0;
info[5] = 0;
info[6] = 0;
info[7] = 0;
info[14] = 0;
info[15] = 0;
} else {
struct pipe_screen *
nv30_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
{
struct nv30_screen *screen = CALLOC_STRUCT(nv30_screen);
struct nouveau_channel *chan;
struct pipe_screen *pscreen;
struct nouveau_stateobj *so;
unsigned rankine_class = 0;
int ret, i;
if (!screen)
return NULL;
pscreen = &screen->base.base;
ret = nouveau_screen_init(&screen->base, dev);
if (ret) {
nv30_screen_destroy(pscreen);
return NULL;
}
chan = screen->base.channel;
pscreen->winsys = ws;
pscreen->destroy = nv30_screen_destroy;
pscreen->get_param = nv30_screen_get_param;
pscreen->get_paramf = nv30_screen_get_paramf;
pscreen->is_format_supported = nv30_screen_surface_format_supported;
pscreen->context_create = nv30_create;
nv30_screen_init_miptree_functions(pscreen);
nv30_screen_init_transfer_functions(pscreen);
/* 3D object */
switch (dev->chipset & 0xf0) {
case 0x30:
if (NV30TCL_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f)))
rankine_class = 0x0397;
else
if (NV34TCL_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f)))
rankine_class = 0x0697;
else
if (NV35TCL_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f)))
rankine_class = 0x0497;
break;
default:
break;
}
if (!rankine_class) {
NOUVEAU_ERR("Unknown nv3x chipset: nv%02x\n", dev->chipset);
return NULL;
}
ret = nouveau_grobj_alloc(chan, 0xbeef3097, rankine_class,
&screen->rankine);
if (ret) {
NOUVEAU_ERR("Error creating 3D object: %d\n", ret);
return FALSE;
}
/* 2D engine setup */
screen->eng2d = nv04_surface_2d_init(&screen->base);
screen->eng2d->buf = nv30_surface_buffer;
/* Notifier for sync purposes */
ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync);
if (ret) {
NOUVEAU_ERR("Error creating notifier object: %d\n", ret);
nv30_screen_destroy(pscreen);
return NULL;
}
/* Query objects */
ret = nouveau_notifier_alloc(chan, 0xbeef0302, 32, &screen->query);
if (ret) {
NOUVEAU_ERR("Error initialising query objects: %d\n", ret);
nv30_screen_destroy(pscreen);
return NULL;
}
ret = nouveau_resource_init(&screen->query_heap, 0, 32);
if (ret) {
NOUVEAU_ERR("Error initialising query object heap: %d\n", ret);
nv30_screen_destroy(pscreen);
return NULL;
}
/* Vtxprog resources */
if (nouveau_resource_init(&screen->vp_exec_heap, 0, 256) ||
nouveau_resource_init(&screen->vp_data_heap, 0, 256)) {
nv30_screen_destroy(pscreen);
return NULL;
}
/* Static rankine initialisation */
so = so_new(36, 60, 0);
so_method(so, screen->rankine, NV34TCL_DMA_NOTIFY, 1);
so_data (so, screen->sync->handle);
so_method(so, screen->rankine, NV34TCL_DMA_TEXTURE0, 2);
so_data (so, chan->vram->handle);
so_data (so, chan->gart->handle);
so_method(so, screen->rankine, NV34TCL_DMA_COLOR1, 1);
so_data (so, chan->vram->handle);
so_method(so, screen->rankine, NV34TCL_DMA_COLOR0, 2);
so_data (so, chan->vram->handle);
so_data (so, chan->vram->handle);
so_method(so, screen->rankine, NV34TCL_DMA_VTXBUF0, 2);
so_data (so, chan->vram->handle);
so_data (so, chan->gart->handle);
/* so_method(so, screen->rankine, NV34TCL_DMA_FENCE, 2);
so_data (so, 0);
so_data (so, screen->query->handle);*/
so_method(so, screen->rankine, NV34TCL_DMA_IN_MEMORY7, 1);
so_data (so, chan->vram->handle);
so_method(so, screen->rankine, NV34TCL_DMA_IN_MEMORY8, 1);
so_data (so, chan->vram->handle);
for (i=1; i<8; i++) {
so_method(so, screen->rankine, NV34TCL_VIEWPORT_CLIP_HORIZ(i), 1);
so_data (so, 0);
so_method(so, screen->rankine, NV34TCL_VIEWPORT_CLIP_VERT(i), 1);
so_data (so, 0);
}
so_method(so, screen->rankine, 0x220, 1);
so_data (so, 1);
so_method(so, screen->rankine, 0x03b0, 1);
so_data (so, 0x00100000);
so_method(so, screen->rankine, 0x1454, 1);
so_data (so, 0);
so_method(so, screen->rankine, 0x1d80, 1);
so_data (so, 3);
so_method(so, screen->rankine, 0x1450, 1);
so_data (so, 0x00030004);
/* NEW */
so_method(so, screen->rankine, 0x1e98, 1);
so_data (so, 0);
so_method(so, screen->rankine, 0x17e0, 3);
so_data (so, fui(0.0));
so_data (so, fui(0.0));
so_data (so, fui(1.0));
so_method(so, screen->rankine, 0x1f80, 16);
for (i=0; i<16; i++) {
so_data (so, (i==8) ? 0x0000ffff : 0);
}
so_method(so, screen->rankine, 0x120, 3);
so_data (so, 0);
so_data (so, 1);
so_data (so, 2);
so_method(so, screen->rankine, 0x1d88, 1);
so_data (so, 0x00001200);
so_method(so, screen->rankine, NV34TCL_RC_ENABLE, 1);
so_data (so, 0);
so_method(so, screen->rankine, NV34TCL_DEPTH_RANGE_NEAR, 2);
so_data (so, fui(0.0));
so_data (so, fui(1.0));
so_method(so, screen->rankine, NV34TCL_MULTISAMPLE_CONTROL, 1);
so_data (so, 0xffff0000);
/* enables use of vp rather than fixed-function somehow */
so_method(so, screen->rankine, 0x1e94, 1);
so_data (so, 0x13);
so_emit(chan, so);
so_ref(NULL, &so);
nouveau_pushbuf_flush(chan, 0);
return pscreen;
}
void
nve4_mp_pm_query_begin(struct nvc0_context *nvc0, struct nvc0_query *q)
{
struct nvc0_screen *screen = nvc0->screen;
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
const struct nve4_mp_pm_query_cfg *cfg;
unsigned i, c;
unsigned num_ab[2] = { 0, 0 };
cfg = &nve4_mp_pm_queries[q->type - PIPE_QUERY_DRIVER_SPECIFIC];
/* check if we have enough free counter slots */
for (i = 0; i < cfg->num_counters; ++i)
num_ab[cfg->ctr[i].sig_dom]++;
if (screen->pm.num_mp_pm_active[0] + num_ab[0] > 4 ||
screen->pm.num_mp_pm_active[1] + num_ab[1] > 4) {
NOUVEAU_ERR("Not enough free MP counter slots !\n");
return;
}
assert(cfg->num_counters <= 4);
PUSH_SPACE(push, 4 * 8 + 6);
if (!screen->pm.mp_counters_enabled) {
screen->pm.mp_counters_enabled = TRUE;
BEGIN_NVC0(push, SUBC_SW(0x06ac), 1);
PUSH_DATA (push, 0x1fcb);
}
/* set sequence field to 0 (used to check if result is available) */
for (i = 0; i < screen->mp_count; ++i)
q->data[i * 10 + 10] = 0;
for (i = 0; i < cfg->num_counters; ++i) {
const unsigned d = cfg->ctr[i].sig_dom;
if (!screen->pm.num_mp_pm_active[d]) {
uint32_t m = (1 << 22) | (1 << (7 + (8 * !d)));
if (screen->pm.num_mp_pm_active[!d])
m |= 1 << (7 + (8 * d));
BEGIN_NVC0(push, SUBC_SW(0x0600), 1);
PUSH_DATA (push, m);
}
screen->pm.num_mp_pm_active[d]++;
for (c = d * 4; c < (d * 4 + 4); ++c) {
if (!screen->pm.mp_counter[c]) {
q->ctr[i] = c;
screen->pm.mp_counter[c] = (struct pipe_query *)q;
break;
}
}
assert(c <= (d * 4 + 3)); /* must succeed, already checked for space */
/* configure and reset the counter(s) */
if (d == 0)
BEGIN_NVC0(push, NVE4_COMPUTE(MP_PM_A_SIGSEL(c & 3)), 1);
else
BEGIN_NVC0(push, NVE4_COMPUTE(MP_PM_B_SIGSEL(c & 3)), 1);
PUSH_DATA (push, cfg->ctr[i].sig_sel);
BEGIN_NVC0(push, NVE4_COMPUTE(MP_PM_SRCSEL(c)), 1);
PUSH_DATA (push, cfg->ctr[i].src_sel + 0x2108421 * (c & 3));
BEGIN_NVC0(push, NVE4_COMPUTE(MP_PM_FUNC(c)), 1);
PUSH_DATA (push, (cfg->ctr[i].func << 4) | cfg->ctr[i].mode);
BEGIN_NVC0(push, NVE4_COMPUTE(MP_PM_SET(c)), 1);
PUSH_DATA (push, 0);
}
}
static int
nv50_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
enum pipe_shader_cap param)
{
switch (shader) {
case PIPE_SHADER_VERTEX:
case PIPE_SHADER_GEOMETRY:
case PIPE_SHADER_FRAGMENT:
break;
case PIPE_SHADER_COMPUTE:
default:
return 0;
}
switch (param) {
case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
return 16384;
case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
return 4;
case PIPE_SHADER_CAP_MAX_INPUTS:
if (shader == PIPE_SHADER_VERTEX)
return 32;
return 15;
case PIPE_SHADER_CAP_MAX_OUTPUTS:
return 16;
case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
return 65536;
case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
return NV50_MAX_PIPE_CONSTBUFS;
case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
return shader != PIPE_SHADER_FRAGMENT;
case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
return 1;
case PIPE_SHADER_CAP_MAX_PREDS:
return 0;
case PIPE_SHADER_CAP_MAX_TEMPS:
return nv50_screen(pscreen)->max_tls_space / ONE_TEMP_SIZE;
case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
return 1;
case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
return 1;
case PIPE_SHADER_CAP_SUBROUTINES:
return 0; /* please inline, or provide function declarations */
case PIPE_SHADER_CAP_INTEGERS:
return 1;
case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
/* The chip could handle more sampler views than samplers */
case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
return MIN2(16, PIPE_MAX_SAMPLERS);
case PIPE_SHADER_CAP_PREFERRED_IR:
return PIPE_SHADER_IR_TGSI;
case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
return 32;
case PIPE_SHADER_CAP_DOUBLES:
case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
case PIPE_SHADER_CAP_SUPPORTED_IRS:
case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
return 0;
default:
NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
return 0;
}
}
bool
nv50_program_translate(struct nv50_program *prog, uint16_t chipset,
struct pipe_debug_callback *debug)
{
struct nv50_ir_prog_info *info;
int ret;
const uint8_t map_undef = (prog->type == PIPE_SHADER_VERTEX) ? 0x40 : 0x80;
info = CALLOC_STRUCT(nv50_ir_prog_info);
if (!info)
return false;
info->type = prog->type;
info->target = chipset;
info->bin.sourceRep = NV50_PROGRAM_IR_TGSI;
info->bin.source = (void *)prog->pipe.tokens;
info->io.auxCBSlot = 15;
info->io.ucpBase = NV50_CB_AUX_UCP_OFFSET;
info->io.genUserClip = prog->vp.clpd_nr;
info->io.resInfoCBSlot = 15;
info->io.suInfoBase = NV50_CB_AUX_TEX_MS_OFFSET;
info->io.sampleInfoBase = NV50_CB_AUX_SAMPLE_OFFSET;
info->io.msInfoCBSlot = 15;
info->io.msInfoBase = NV50_CB_AUX_MS_OFFSET;
info->assignSlots = nv50_program_assign_varying_slots;
prog->vp.bfc[0] = 0xff;
prog->vp.bfc[1] = 0xff;
prog->vp.edgeflag = 0xff;
prog->vp.clpd[0] = map_undef;
prog->vp.clpd[1] = map_undef;
prog->vp.psiz = map_undef;
prog->gp.has_layer = 0;
prog->gp.has_viewport = 0;
if (prog->type == PIPE_SHADER_COMPUTE)
info->prop.cp.inputOffset = 0x10;
info->driverPriv = prog;
#ifdef DEBUG
info->optLevel = debug_get_num_option("NV50_PROG_OPTIMIZE", 3);
info->dbgFlags = debug_get_num_option("NV50_PROG_DEBUG", 0);
#else
info->optLevel = 3;
#endif
ret = nv50_ir_generate_code(info);
if (ret) {
NOUVEAU_ERR("shader translation failed: %i\n", ret);
goto out;
}
FREE(info->bin.syms);
prog->code = info->bin.code;
prog->code_size = info->bin.codeSize;
prog->fixups = info->bin.relocData;
prog->interps = info->bin.interpData;
prog->max_gpr = MAX2(4, (info->bin.maxGPR >> 1) + 1);
prog->tls_space = info->bin.tlsSpace;
prog->vp.need_vertex_id = info->io.vertexId < PIPE_MAX_SHADER_INPUTS;
if (prog->type == PIPE_SHADER_FRAGMENT) {
if (info->prop.fp.writesDepth) {
prog->fp.flags[0] |= NV50_3D_FP_CONTROL_EXPORTS_Z;
prog->fp.flags[1] = 0x11;
}
if (info->prop.fp.usesDiscard)
prog->fp.flags[0] |= NV50_3D_FP_CONTROL_USES_KIL;
} else
if (prog->type == PIPE_SHADER_GEOMETRY) {
switch (info->prop.gp.outputPrim) {
case PIPE_PRIM_LINE_STRIP:
prog->gp.prim_type = NV50_3D_GP_OUTPUT_PRIMITIVE_TYPE_LINE_STRIP;
break;
case PIPE_PRIM_TRIANGLE_STRIP:
prog->gp.prim_type = NV50_3D_GP_OUTPUT_PRIMITIVE_TYPE_TRIANGLE_STRIP;
break;
case PIPE_PRIM_POINTS:
default:
assert(info->prop.gp.outputPrim == PIPE_PRIM_POINTS);
prog->gp.prim_type = NV50_3D_GP_OUTPUT_PRIMITIVE_TYPE_POINTS;
break;
}
prog->gp.vert_count = info->prop.gp.maxVertices;
} else
if (prog->type == PIPE_SHADER_COMPUTE) {
prog->cp.syms = info->bin.syms;
prog->cp.num_syms = info->bin.numSyms;
}
if (prog->pipe.stream_output.num_outputs)
prog->so = nv50_program_create_strmout_state(info,
&prog->pipe.stream_output);
pipe_debug_message(debug, SHADER_INFO,
"type: %d, local: %d, gpr: %d, inst: %d, bytes: %d",
prog->type, info->bin.tlsSpace, prog->max_gpr,
info->bin.instructions, info->bin.codeSize);
out:
FREE(info);
return !ret;
}
static int
nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
{
const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
switch (param) {
case PIPE_CAP_MAX_COMBINED_SAMPLERS:
return 64;
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
return 14;
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
return 12;
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
return 14;
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
return 512;
case PIPE_CAP_MIN_TEXEL_OFFSET:
return -8;
case PIPE_CAP_MAX_TEXEL_OFFSET:
return 7;
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
case PIPE_CAP_TEXTURE_SWIZZLE:
case PIPE_CAP_TEXTURE_SHADOW_MAP:
case PIPE_CAP_NPOT_TEXTURES:
case PIPE_CAP_ANISOTROPIC_FILTER:
case PIPE_CAP_SCALED_RESOLVE:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
return 1;
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
return 65536;
case PIPE_CAP_SEAMLESS_CUBE_MAP:
return nv50_screen(pscreen)->tesla->oclass >= NVA0_3D_CLASS;
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
return 0;
case PIPE_CAP_CUBE_MAP_ARRAY:
return 0;
/*
return nv50_screen(pscreen)->tesla->oclass >= NVA3_3D_CLASS;
*/
case PIPE_CAP_TWO_SIDED_STENCIL:
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_POINT_SPRITE:
return 1;
case PIPE_CAP_SM3:
return 1;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
return 140;
case PIPE_CAP_MAX_RENDER_TARGETS:
return 8;
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
return 1;
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
return 1;
case PIPE_CAP_QUERY_TIMESTAMP:
case PIPE_CAP_QUERY_TIME_ELAPSED:
case PIPE_CAP_OCCLUSION_QUERY:
return 1;
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
return 4;
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
return 64;
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
return (class_3d >= NVA0_3D_CLASS) ? 1 : 0;
case PIPE_CAP_BLEND_EQUATION_SEPARATE:
case PIPE_CAP_INDEP_BLEND_ENABLE:
return 1;
case PIPE_CAP_INDEP_BLEND_FUNC:
return nv50_screen(pscreen)->tesla->oclass >= NVA3_3D_CLASS;
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
return 1;
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
return 0;
case PIPE_CAP_SHADER_STENCIL_EXPORT:
return 0;
case PIPE_CAP_PRIMITIVE_RESTART:
case PIPE_CAP_TGSI_INSTANCEID:
case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
case PIPE_CAP_CONDITIONAL_RENDER:
case PIPE_CAP_TEXTURE_BARRIER:
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_START_INSTANCE:
return 1;
case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
return 0; /* state trackers will know better */
case PIPE_CAP_USER_CONSTANT_BUFFERS:
case PIPE_CAP_USER_INDEX_BUFFERS:
case PIPE_CAP_USER_VERTEX_BUFFERS:
return 1;
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
return 256;
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
return 1; /* 256 for binding as RT, but that's not possible in GL */
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_TGSI_TEXCOORD:
case PIPE_CAP_TEXTURE_MULTISAMPLE:
return 0;
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
return 1;
case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
return 0;
case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
case PIPE_CAP_ENDIANNESS:
return PIPE_ENDIAN_LITTLE;
default:
NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
return 0;
}
}
static void
nvc0_render_tri(struct draw_stage *stage, struct prim_header *prim)
{
NOUVEAU_ERR("\n");
}
static void
nvc0_render_reset_stipple_counter(struct draw_stage *stage)
{
NOUVEAU_ERR("\n");
}
static boolean
nv30_vertprog_parse_instruction(struct nv30_vpc *vpc,
const struct tgsi_full_instruction *finst)
{
struct nv30_sreg src[3], dst, tmp;
struct nv30_sreg none = nv30_sr(NV30SR_NONE, 0);
int mask;
int ai = -1, ci = -1;
int i;
if (finst->Instruction.Opcode == TGSI_OPCODE_END)
return TRUE;
vpc->temp_temp_count = 0;
for (i = 0; i < finst->Instruction.NumSrcRegs; i++) {
const struct tgsi_full_src_register *fsrc;
fsrc = &finst->FullSrcRegisters[i];
if (fsrc->SrcRegister.File == TGSI_FILE_TEMPORARY) {
src[i] = tgsi_src(vpc, fsrc);
}
}
for (i = 0; i < finst->Instruction.NumSrcRegs; i++) {
const struct tgsi_full_src_register *fsrc;
fsrc = &finst->FullSrcRegisters[i];
switch (fsrc->SrcRegister.File) {
case TGSI_FILE_INPUT:
if (ai == -1 || ai == fsrc->SrcRegister.Index) {
ai = fsrc->SrcRegister.Index;
src[i] = tgsi_src(vpc, fsrc);
} else {
src[i] = temp(vpc);
arith(vpc, 0, OP_MOV, src[i], MASK_ALL,
tgsi_src(vpc, fsrc), none, none);
}
break;
/*XXX: index comparison is broken now that consts come from
* two different register files.
*/
case TGSI_FILE_CONSTANT:
case TGSI_FILE_IMMEDIATE:
if (ci == -1 || ci == fsrc->SrcRegister.Index) {
ci = fsrc->SrcRegister.Index;
src[i] = tgsi_src(vpc, fsrc);
} else {
src[i] = temp(vpc);
arith(vpc, 0, OP_MOV, src[i], MASK_ALL,
tgsi_src(vpc, fsrc), none, none);
}
break;
case TGSI_FILE_TEMPORARY:
/* handled above */
break;
default:
NOUVEAU_ERR("bad src file\n");
return FALSE;
}
}
dst = tgsi_dst(vpc, &finst->FullDstRegisters[0]);
mask = tgsi_mask(finst->FullDstRegisters[0].DstRegister.WriteMask);
switch (finst->Instruction.Opcode) {
case TGSI_OPCODE_ABS:
arith(vpc, 0, OP_MOV, dst, mask, abs(src[0]), none, none);
break;
case TGSI_OPCODE_ADD:
arith(vpc, 0, OP_ADD, dst, mask, src[0], none, src[1]);
break;
case TGSI_OPCODE_ARL:
arith(vpc, 0, OP_ARL, dst, mask, src[0], none, none);
break;
case TGSI_OPCODE_DP3:
arith(vpc, 0, OP_DP3, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_DP4:
arith(vpc, 0, OP_DP4, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_DPH:
arith(vpc, 0, OP_DPH, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_DST:
arith(vpc, 0, OP_DST, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_EX2:
arith(vpc, 1, OP_EX2, dst, mask, none, none, src[0]);
break;
case TGSI_OPCODE_EXP:
arith(vpc, 1, OP_EXP, dst, mask, none, none, src[0]);
break;
case TGSI_OPCODE_FLR:
arith(vpc, 0, OP_FLR, dst, mask, src[0], none, none);
break;
case TGSI_OPCODE_FRC:
arith(vpc, 0, OP_FRC, dst, mask, src[0], none, none);
break;
case TGSI_OPCODE_LG2:
arith(vpc, 1, OP_LG2, dst, mask, none, none, src[0]);
break;
case TGSI_OPCODE_LIT:
arith(vpc, 1, OP_LIT, dst, mask, none, none, src[0]);
break;
case TGSI_OPCODE_LOG:
arith(vpc, 1, OP_LOG, dst, mask, none, none, src[0]);
break;
case TGSI_OPCODE_MAD:
arith(vpc, 0, OP_MAD, dst, mask, src[0], src[1], src[2]);
break;
case TGSI_OPCODE_MAX:
arith(vpc, 0, OP_MAX, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_MIN:
arith(vpc, 0, OP_MIN, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_MOV:
arith(vpc, 0, OP_MOV, dst, mask, src[0], none, none);
break;
case TGSI_OPCODE_MUL:
arith(vpc, 0, OP_MUL, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_POW:
tmp = temp(vpc);
arith(vpc, 1, OP_LG2, tmp, MASK_X, none, none,
swz(src[0], X, X, X, X));
arith(vpc, 0, OP_MUL, tmp, MASK_X, swz(tmp, X, X, X, X),
swz(src[1], X, X, X, X), none);
arith(vpc, 1, OP_EX2, dst, mask, none, none,
swz(tmp, X, X, X, X));
break;
case TGSI_OPCODE_RCP:
arith(vpc, 1, OP_RCP, dst, mask, none, none, src[0]);
break;
case TGSI_OPCODE_RET:
break;
case TGSI_OPCODE_RSQ:
arith(vpc, 1, OP_RSQ, dst, mask, none, none, src[0]);
break;
case TGSI_OPCODE_SGE:
arith(vpc, 0, OP_SGE, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_SGT:
arith(vpc, 0, OP_SGT, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_SLT:
arith(vpc, 0, OP_SLT, dst, mask, src[0], src[1], none);
break;
case TGSI_OPCODE_SUB:
arith(vpc, 0, OP_ADD, dst, mask, src[0], none, neg(src[1]));
break;
case TGSI_OPCODE_XPD:
tmp = temp(vpc);
arith(vpc, 0, OP_MUL, tmp, mask,
swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none);
arith(vpc, 0, OP_MAD, dst, (mask & ~MASK_W),
swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y),
neg(tmp));
break;
default:
NOUVEAU_ERR("invalid opcode %d\n", finst->Instruction.Opcode);
return FALSE;
}
return TRUE;
}
int
nv50_screen_compute_setup(struct nv50_screen *screen,
struct nouveau_pushbuf *push)
{
struct nouveau_device *dev = screen->base.device;
struct nouveau_object *chan = screen->base.channel;
struct nv04_fifo *fifo = (struct nv04_fifo *)chan->data;
unsigned obj_class;
int i, ret;
switch (dev->chipset & 0xf0) {
case 0x50:
case 0x80:
case 0x90:
obj_class = NV50_COMPUTE_CLASS;
break;
case 0xa0:
switch (dev->chipset) {
case 0xa3:
case 0xa5:
case 0xa8:
obj_class = NVA3_COMPUTE_CLASS;
break;
default:
obj_class = NV50_COMPUTE_CLASS;
break;
}
break;
default:
NOUVEAU_ERR("unsupported chipset: NV%02x\n", dev->chipset);
return -1;
}
ret = nouveau_object_new(chan, 0xbeef50c0, obj_class, NULL, 0,
&screen->compute);
if (ret)
return ret;
BEGIN_NV04(push, SUBC_COMPUTE(NV01_SUBCHAN_OBJECT), 1);
PUSH_DATA (push, screen->compute->handle);
BEGIN_NV04(push, NV50_COMPUTE(UNK02A0), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_COMPUTE(DMA_STACK), 1);
PUSH_DATA (push, fifo->vram);
BEGIN_NV04(push, NV50_COMPUTE(STACK_ADDRESS_HIGH), 2);
PUSH_DATAh(push, screen->stack_bo->offset);
PUSH_DATA (push, screen->stack_bo->offset);
BEGIN_NV04(push, NV50_COMPUTE(STACK_SIZE_LOG), 1);
PUSH_DATA (push, 4);
BEGIN_NV04(push, NV50_COMPUTE(UNK0290), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_COMPUTE(LANES32_ENABLE), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_COMPUTE(REG_MODE), 1);
PUSH_DATA (push, NV50_COMPUTE_REG_MODE_STRIPED);
BEGIN_NV04(push, NV50_COMPUTE(UNK0384), 1);
PUSH_DATA (push, 0x100);
BEGIN_NV04(push, NV50_COMPUTE(DMA_GLOBAL), 1);
PUSH_DATA (push, fifo->vram);
for (i = 0; i < 15; i++) {
BEGIN_NV04(push, NV50_COMPUTE(GLOBAL_ADDRESS_HIGH(i)), 2);
PUSH_DATA (push, 0);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_COMPUTE(GLOBAL_LIMIT(i)), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_COMPUTE(GLOBAL_MODE(i)), 1);
PUSH_DATA (push, NV50_COMPUTE_GLOBAL_MODE_LINEAR);
}
BEGIN_NV04(push, NV50_COMPUTE(GLOBAL_ADDRESS_HIGH(15)), 2);
PUSH_DATA (push, 0);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_COMPUTE(GLOBAL_LIMIT(15)), 1);
PUSH_DATA (push, ~0);
BEGIN_NV04(push, NV50_COMPUTE(GLOBAL_MODE(15)), 1);
PUSH_DATA (push, NV50_COMPUTE_GLOBAL_MODE_LINEAR);
BEGIN_NV04(push, NV50_COMPUTE(LOCAL_WARPS_LOG_ALLOC), 1);
PUSH_DATA (push, 7);
BEGIN_NV04(push, NV50_COMPUTE(LOCAL_WARPS_NO_CLAMP), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_COMPUTE(STACK_WARPS_LOG_ALLOC), 1);
PUSH_DATA (push, 7);
BEGIN_NV04(push, NV50_COMPUTE(STACK_WARPS_NO_CLAMP), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_COMPUTE(USER_PARAM_COUNT), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_COMPUTE(DMA_TEXTURE), 1);
PUSH_DATA (push, fifo->vram);
BEGIN_NV04(push, NV50_COMPUTE(TEX_LIMITS), 1);
PUSH_DATA (push, 0x54);
BEGIN_NV04(push, NV50_COMPUTE(LINKED_TSC), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_COMPUTE(DMA_TIC), 1);
PUSH_DATA (push, fifo->vram);
BEGIN_NV04(push, NV50_COMPUTE(TIC_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->txc->offset);
PUSH_DATA (push, screen->txc->offset);
PUSH_DATA (push, NV50_TIC_MAX_ENTRIES - 1);
BEGIN_NV04(push, NV50_COMPUTE(DMA_TSC), 1);
PUSH_DATA (push, fifo->vram);
BEGIN_NV04(push, NV50_COMPUTE(TSC_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->txc->offset + 65536);
PUSH_DATA (push, screen->txc->offset + 65536);
PUSH_DATA (push, NV50_TSC_MAX_ENTRIES - 1);
BEGIN_NV04(push, NV50_COMPUTE(DMA_CODE_CB), 1);
PUSH_DATA (push, fifo->vram);
BEGIN_NV04(push, NV50_COMPUTE(DMA_LOCAL), 1);
PUSH_DATA (push, fifo->vram);
BEGIN_NV04(push, NV50_COMPUTE(LOCAL_ADDRESS_HIGH), 2);
PUSH_DATAh(push, screen->tls_bo->offset + 65536);
PUSH_DATA (push, screen->tls_bo->offset + 65536);
BEGIN_NV04(push, NV50_COMPUTE(LOCAL_SIZE_LOG), 1);
PUSH_DATA (push, util_logbase2((screen->max_tls_space / ONE_TEMP_SIZE) * 2));
return 0;
}
struct pipe_screen *
nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
{
struct nv50_screen *screen = CALLOC_STRUCT(nv50_screen);
struct nouveau_channel *chan;
struct pipe_screen *pscreen;
struct nouveau_stateobj *so;
unsigned chipset = dev->chipset;
unsigned tesla_class = 0;
int ret, i;
if (!screen)
return NULL;
pscreen = &screen->base.base;
ret = nouveau_screen_init(&screen->base, dev);
if (ret) {
nv50_screen_destroy(pscreen);
return NULL;
}
chan = screen->base.channel;
pscreen->winsys = ws;
pscreen->destroy = nv50_screen_destroy;
pscreen->get_param = nv50_screen_get_param;
pscreen->get_paramf = nv50_screen_get_paramf;
pscreen->is_format_supported = nv50_screen_is_format_supported;
pscreen->context_create = nv50_create;
screen->base.pre_pipebuffer_map_callback = nv50_pre_pipebuffer_map;
nv50_screen_init_miptree_functions(pscreen);
nv50_transfer_init_screen_functions(pscreen);
/* DMA engine object */
ret = nouveau_grobj_alloc(chan, 0xbeef5039,
NV50_MEMORY_TO_MEMORY_FORMAT, &screen->m2mf);
if (ret) {
NOUVEAU_ERR("Error creating M2MF object: %d\n", ret);
nv50_screen_destroy(pscreen);
return NULL;
}
/* 2D object */
ret = nouveau_grobj_alloc(chan, 0xbeef502d, NV50_2D, &screen->eng2d);
if (ret) {
NOUVEAU_ERR("Error creating 2D object: %d\n", ret);
nv50_screen_destroy(pscreen);
return NULL;
}
/* 3D object */
switch (chipset & 0xf0) {
case 0x50:
tesla_class = NV50TCL;
break;
case 0x80:
case 0x90:
tesla_class = NV84TCL;
break;
case 0xa0:
switch (chipset) {
case 0xa0:
case 0xaa:
case 0xac:
tesla_class = NVA0TCL;
break;
default:
tesla_class = NVA8TCL;
break;
}
break;
default:
NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", chipset);
nv50_screen_destroy(pscreen);
return NULL;
}
ret = nouveau_grobj_alloc(chan, 0xbeef5097, tesla_class,
&screen->tesla);
if (ret) {
NOUVEAU_ERR("Error creating 3D object: %d\n", ret);
nv50_screen_destroy(pscreen);
return NULL;
}
/* Sync notifier */
ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync);
if (ret) {
NOUVEAU_ERR("Error creating notifier object: %d\n", ret);
nv50_screen_destroy(pscreen);
return NULL;
}
/* Static M2MF init */
so = so_new(1, 3, 0);
so_method(so, screen->m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_DMA_NOTIFY, 3);
so_data (so, screen->sync->handle);
so_data (so, chan->vram->handle);
so_data (so, chan->vram->handle);
so_emit(chan, so);
so_ref (NULL, &so);
/* Static 2D init */
so = so_new(4, 7, 0);
so_method(so, screen->eng2d, NV50_2D_DMA_NOTIFY, 4);
so_data (so, screen->sync->handle);
so_data (so, chan->vram->handle);
so_data (so, chan->vram->handle);
so_data (so, chan->vram->handle);
so_method(so, screen->eng2d, NV50_2D_OPERATION, 1);
so_data (so, NV50_2D_OPERATION_SRCCOPY);
so_method(so, screen->eng2d, NV50_2D_CLIP_ENABLE, 1);
so_data (so, 0);
so_method(so, screen->eng2d, 0x0888, 1);
so_data (so, 1);
so_emit(chan, so);
so_ref(NULL, &so);
/* Static tesla init */
so = so_new(47, 95, 24);
so_method(so, screen->tesla, NV50TCL_COND_MODE, 1);
so_data (so, NV50TCL_COND_MODE_ALWAYS);
so_method(so, screen->tesla, NV50TCL_DMA_NOTIFY, 1);
so_data (so, screen->sync->handle);
so_method(so, screen->tesla, NV50TCL_DMA_ZETA, 11);
for (i = 0; i < 11; i++)
so_data(so, chan->vram->handle);
so_method(so, screen->tesla, NV50TCL_DMA_COLOR(0),
NV50TCL_DMA_COLOR__SIZE);
for (i = 0; i < NV50TCL_DMA_COLOR__SIZE; i++)
so_data(so, chan->vram->handle);
so_method(so, screen->tesla, NV50TCL_RT_CONTROL, 1);
so_data (so, 1);
/* activate all 32 lanes (threads) in a warp */
so_method(so, screen->tesla, NV50TCL_WARP_HALVES, 1);
so_data (so, 0x2);
so_method(so, screen->tesla, 0x1400, 1);
so_data (so, 0xf);
/* max TIC (bits 4:8) & TSC (ignored) bindings, per program type */
for (i = 0; i < 3; ++i) {
so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(i), 1);
so_data (so, 0x54);
}
/* origin is top left (set to 1 for bottom left) */
so_method(so, screen->tesla, NV50TCL_Y_ORIGIN_BOTTOM, 1);
so_data (so, 0);
so_method(so, screen->tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1);
so_data (so, 8);
/* constant buffers for immediates and VP/FP parameters */
ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, (32 * 4) * 4,
&screen->constbuf_misc[0]);
if (ret) {
nv50_screen_destroy(pscreen);
return NULL;
}
for (i = 0; i < 3; i++) {
ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, (256 * 4) * 4,
&screen->constbuf_parm[i]);
if (ret) {
nv50_screen_destroy(pscreen);
return NULL;
}
}
if (nouveau_resource_init(&screen->immd_heap[0], 0, 128) ||
nouveau_resource_init(&screen->parm_heap[0], 0, 512) ||
nouveau_resource_init(&screen->parm_heap[1], 0, 512))
{
NOUVEAU_ERR("Error initialising constant buffers.\n");
nv50_screen_destroy(pscreen);
return NULL;
}
/*
// map constant buffers:
// B = buffer ID (maybe more than 1 byte)
// N = CB index used in shader instruction
// P = program type (0 = VP, 2 = GP, 3 = FP)
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x000BBNP1);
*/
so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
so_reloc (so, screen->constbuf_misc[0], 0, NOUVEAU_BO_VRAM |
NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
so_reloc (so, screen->constbuf_misc[0], 0, NOUVEAU_BO_VRAM |
NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
so_data (so, (NV50_CB_PMISC << 16) | 0x00000200);
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x00000001 | (NV50_CB_PMISC << 12));
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x00000021 | (NV50_CB_PMISC << 12));
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x00000031 | (NV50_CB_PMISC << 12));
/* bind auxiliary constbuf to immediate data bo */
so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
so_reloc (so, screen->constbuf_misc[0], (128 * 4) * 4,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
so_reloc (so, screen->constbuf_misc[0], (128 * 4) * 4,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
so_data (so, (NV50_CB_AUX << 16) | 0x00000200);
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x00000201 | (NV50_CB_AUX << 12));
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x00000221 | (NV50_CB_AUX << 12));
so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
so_data (so, (NV50_CB_PVP << 16) | 0x00000800);
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x00000101 | (NV50_CB_PVP << 12));
so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
so_reloc (so, screen->constbuf_parm[PIPE_SHADER_GEOMETRY], 0,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
so_reloc (so, screen->constbuf_parm[PIPE_SHADER_GEOMETRY], 0,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
so_data (so, (NV50_CB_PGP << 16) | 0x00000800);
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x00000121 | (NV50_CB_PGP << 12));
so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3);
so_reloc (so, screen->constbuf_parm[PIPE_SHADER_FRAGMENT], 0,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
so_reloc (so, screen->constbuf_parm[PIPE_SHADER_FRAGMENT], 0,
NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
so_data (so, (NV50_CB_PFP << 16) | 0x00000800);
so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1);
so_data (so, 0x00000131 | (NV50_CB_PFP << 12));
ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, PIPE_SHADER_TYPES*32*32,
&screen->tic);
if (ret) {
nv50_screen_destroy(pscreen);
return NULL;
}
so_method(so, screen->tesla, NV50TCL_TIC_ADDRESS_HIGH, 3);
so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM |
NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM |
NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
so_data (so, PIPE_SHADER_TYPES * 32 - 1);
ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, PIPE_SHADER_TYPES*32*32,
&screen->tsc);
if (ret) {
nv50_screen_destroy(pscreen);
return NULL;
}
so_method(so, screen->tesla, NV50TCL_TSC_ADDRESS_HIGH, 3);
so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM |
NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM |
NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
so_data (so, 0x00000000); /* ignored if TSC_LINKED (0x1234) = 1 */
/* Vertex array limits - max them out */
for (i = 0; i < 16; i++) {
so_method(so, screen->tesla, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i), 2);
so_data (so, 0x000000ff);
so_data (so, 0xffffffff);
}
so_method(so, screen->tesla, NV50TCL_DEPTH_RANGE_NEAR(0), 2);
so_data (so, fui(0.0));
so_data (so, fui(1.0));
/* no dynamic combination of TIC & TSC entries => only BIND_TIC used */
so_method(so, screen->tesla, NV50TCL_LINKED_TSC, 1);
so_data (so, 1);
/* activate first scissor rectangle */
so_method(so, screen->tesla, NV50TCL_SCISSOR_ENABLE(0), 1);
so_data (so, 1);
so_method(so, screen->tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
so_data (so, 1); /* default edgeflag to TRUE */
so_emit(chan, so);
so_ref (so, &screen->static_init);
so_ref (NULL, &so);
nouveau_pushbuf_flush(chan, 0);
return pscreen;
}
static int
nv30_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
{
struct nv30_screen *screen = nv30_screen(pscreen);
struct nouveau_object *eng3d = screen->eng3d;
struct nouveau_device *dev = nouveau_screen(pscreen)->device;
switch (param) {
/* non-boolean capabilities */
case PIPE_CAP_MAX_RENDER_TARGETS:
return (eng3d->oclass >= NV40_3D_CLASS) ? 4 : 1;
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
return 13;
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
return 10;
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
return 13;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
return 120;
case PIPE_CAP_ENDIANNESS:
return PIPE_ENDIAN_LITTLE;
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
return 16;
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
case PIPE_CAP_MAX_VIEWPORTS:
return 1;
case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
return 2048;
/* supported capabilities */
case PIPE_CAP_TWO_SIDED_STENCIL:
case PIPE_CAP_ANISOTROPIC_FILTER:
case PIPE_CAP_POINT_SPRITE:
case PIPE_CAP_OCCLUSION_QUERY:
case PIPE_CAP_QUERY_TIME_ELAPSED:
case PIPE_CAP_QUERY_TIMESTAMP:
case PIPE_CAP_TEXTURE_SHADOW_MAP:
case PIPE_CAP_TEXTURE_SWIZZLE:
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
case PIPE_CAP_TGSI_TEXCOORD:
case PIPE_CAP_USER_CONSTANT_BUFFERS:
case PIPE_CAP_USER_INDEX_BUFFERS:
case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
return 1;
/* nv35 capabilities */
case PIPE_CAP_DEPTH_BOUNDS_TEST:
return eng3d->oclass == NV35_3D_CLASS || eng3d->oclass >= NV40_3D_CLASS;
/* nv4x capabilities */
case PIPE_CAP_BLEND_EQUATION_SEPARATE:
case PIPE_CAP_NPOT_TEXTURES:
case PIPE_CAP_CONDITIONAL_RENDER:
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
case PIPE_CAP_PRIMITIVE_RESTART:
return (eng3d->oclass >= NV40_3D_CLASS) ? 1 : 0;
/* unsupported */
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
case PIPE_CAP_SM3:
case PIPE_CAP_INDEP_BLEND_ENABLE:
case PIPE_CAP_INDEP_BLEND_FUNC:
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
case PIPE_CAP_SHADER_STENCIL_EXPORT:
case PIPE_CAP_TGSI_INSTANCEID:
case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR: /* XXX: yes? */
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
case PIPE_CAP_MIN_TEXEL_OFFSET:
case PIPE_CAP_MAX_TEXEL_OFFSET:
case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
case PIPE_CAP_MAX_VERTEX_STREAMS:
case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
case PIPE_CAP_TEXTURE_BARRIER:
case PIPE_CAP_SEAMLESS_CUBE_MAP:
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_CUBE_MAP_ARRAY:
case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
case PIPE_CAP_START_INSTANCE:
case PIPE_CAP_TEXTURE_MULTISAMPLE:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
case PIPE_CAP_TEXTURE_GATHER_SM5:
case PIPE_CAP_FAKE_SW_MSAA:
case PIPE_CAP_TEXTURE_QUERY_LOD:
case PIPE_CAP_SAMPLE_SHADING:
case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
case PIPE_CAP_USER_VERTEX_BUFFERS:
case PIPE_CAP_COMPUTE:
case PIPE_CAP_DRAW_INDIRECT:
case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
case PIPE_CAP_SAMPLER_VIEW_TARGET:
case PIPE_CAP_CLIP_HALFZ:
case PIPE_CAP_VERTEXID_NOBASE:
case PIPE_CAP_POLYGON_OFFSET_CLAMP:
case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
return 0;
case PIPE_CAP_VENDOR_ID:
return 0x10de;
case PIPE_CAP_DEVICE_ID: {
uint64_t device_id;
if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
return -1;
}
return device_id;
}
case PIPE_CAP_ACCELERATED:
return 1;
case PIPE_CAP_VIDEO_MEMORY:
return dev->vram_size >> 20;
case PIPE_CAP_UMA:
return 0;
}
debug_printf("unknown param %d\n", param);
return 0;
}
bool
nv50_program_upload_code(struct nv50_context *nv50, struct nv50_program *prog)
{
struct nouveau_heap *heap;
int ret;
uint32_t size = align(prog->code_size, 0x40);
uint8_t prog_type;
switch (prog->type) {
case PIPE_SHADER_VERTEX: heap = nv50->screen->vp_code_heap; break;
case PIPE_SHADER_GEOMETRY: heap = nv50->screen->gp_code_heap; break;
case PIPE_SHADER_FRAGMENT: heap = nv50->screen->fp_code_heap; break;
case PIPE_SHADER_COMPUTE: heap = nv50->screen->fp_code_heap; break;
default:
assert(!"invalid program type");
return false;
}
ret = nouveau_heap_alloc(heap, size, prog, &prog->mem);
if (ret) {
/* Out of space: evict everything to compactify the code segment, hoping
* the working set is much smaller and drifts slowly. Improve me !
*/
while (heap->next) {
struct nv50_program *evict = heap->next->priv;
if (evict)
nouveau_heap_free(&evict->mem);
}
debug_printf("WARNING: out of code space, evicting all shaders.\n");
ret = nouveau_heap_alloc(heap, size, prog, &prog->mem);
if (ret) {
NOUVEAU_ERR("shader too large (0x%x) to fit in code space ?\n", size);
return false;
}
}
if (prog->type == PIPE_SHADER_COMPUTE) {
/* CP code must be uploaded in FP code segment. */
prog_type = 1;
} else {
prog->code_base = prog->mem->start;
prog_type = prog->type;
}
ret = nv50_tls_realloc(nv50->screen, prog->tls_space);
if (ret < 0) {
nouveau_heap_free(&prog->mem);
return false;
}
if (ret > 0)
nv50->state.new_tls_space = true;
if (prog->fixups)
nv50_ir_relocate_code(prog->fixups, prog->code, prog->code_base, 0, 0);
if (prog->interps)
nv50_ir_change_interp(prog->interps, prog->code,
prog->fp.force_persample_interp,
false /* flatshade */);
nv50_sifc_linear_u8(&nv50->base, nv50->screen->code,
(prog_type << NV50_CODE_BO_SIZE_LOG2) + prog->code_base,
NOUVEAU_BO_VRAM, prog->code_size, prog->code);
BEGIN_NV04(nv50->base.pushbuf, NV50_3D(CODE_CB_FLUSH), 1);
PUSH_DATA (nv50->base.pushbuf, 0);
return true;
}
void
nv50_constbufs_validate(struct nv50_context *nv50)
{
struct nouveau_pushbuf *push = nv50->base.pushbuf;
unsigned s;
for (s = 0; s < 3; ++s) {
unsigned p;
if (s == PIPE_SHADER_FRAGMENT)
p = NV50_3D_SET_PROGRAM_CB_PROGRAM_FRAGMENT;
else
if (s == PIPE_SHADER_GEOMETRY)
p = NV50_3D_SET_PROGRAM_CB_PROGRAM_GEOMETRY;
else
p = NV50_3D_SET_PROGRAM_CB_PROGRAM_VERTEX;
while (nv50->constbuf_dirty[s]) {
const unsigned i = (unsigned)ffs(nv50->constbuf_dirty[s]) - 1;
assert(i < NV50_MAX_PIPE_CONSTBUFS);
nv50->constbuf_dirty[s] &= ~(1 << i);
if (nv50->constbuf[s][i].user) {
const unsigned b = NV50_CB_PVP + s;
unsigned start = 0;
unsigned words = nv50->constbuf[s][0].size / 4;
if (i) {
NOUVEAU_ERR("user constbufs only supported in slot 0\n");
continue;
}
if (!nv50->state.uniform_buffer_bound[s]) {
nv50->state.uniform_buffer_bound[s] = true;
BEGIN_NV04(push, NV50_3D(SET_PROGRAM_CB), 1);
PUSH_DATA (push, (b << 12) | (i << 8) | p | 1);
}
while (words) {
unsigned nr = MIN2(words, NV04_PFIFO_MAX_PACKET_LEN);
PUSH_SPACE(push, nr + 3);
BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
PUSH_DATA (push, (start << 8) | b);
BEGIN_NI04(push, NV50_3D(CB_DATA(0)), nr);
PUSH_DATAp(push, &nv50->constbuf[s][0].u.data[start * 4], nr);
start += nr;
words -= nr;
}
} else {
struct nv04_resource *res =
nv04_resource(nv50->constbuf[s][i].u.buf);
if (res) {
/* TODO: allocate persistent bindings */
const unsigned b = s * 16 + i;
assert(nouveau_resource_mapped_by_gpu(&res->base));
BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
PUSH_DATAh(push, res->address + nv50->constbuf[s][i].offset);
PUSH_DATA (push, res->address + nv50->constbuf[s][i].offset);
PUSH_DATA (push, (b << 16) |
(nv50->constbuf[s][i].size & 0xffff));
BEGIN_NV04(push, NV50_3D(SET_PROGRAM_CB), 1);
PUSH_DATA (push, (b << 12) | (i << 8) | p | 1);
BCTX_REFN(nv50->bufctx_3d, CB(s, i), res, RD);
nv50->cb_dirty = 1; /* Force cache flush for UBO. */
} else {
BEGIN_NV04(push, NV50_3D(SET_PROGRAM_CB), 1);
PUSH_DATA (push, (i << 8) | p | 0);
}
if (i == 0)
nv50->state.uniform_buffer_bound[s] = false;
}
}
}
}
struct pipe_sampler_view *
nv50_create_texture_view(struct pipe_context *pipe,
struct pipe_resource *texture,
const struct pipe_sampler_view *templ,
uint32_t flags,
enum pipe_texture_target target)
{
const struct util_format_description *desc;
uint64_t addr;
uint32_t *tic;
uint32_t swz[4];
uint32_t depth;
struct nv50_tic_entry *view;
struct nv50_miptree *mt = nv50_miptree(texture);
boolean tex_int;
view = MALLOC_STRUCT(nv50_tic_entry);
if (!view)
return NULL;
view->pipe = *templ;
view->pipe.reference.count = 1;
view->pipe.texture = NULL;
view->pipe.context = pipe;
view->id = -1;
pipe_resource_reference(&view->pipe.texture, texture);
tic = &view->tic[0];
desc = util_format_description(view->pipe.format);
/* TIC[0] */
tic[0] = nv50_format_table[view->pipe.format].tic;
tex_int = util_format_is_pure_integer(view->pipe.format);
swz[0] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_r, tex_int);
swz[1] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_g, tex_int);
swz[2] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_b, tex_int);
swz[3] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_a, tex_int);
tic[0] = (tic[0] & ~NV50_TIC_0_SWIZZLE__MASK) |
(swz[0] << NV50_TIC_0_MAPR__SHIFT) |
(swz[1] << NV50_TIC_0_MAPG__SHIFT) |
(swz[2] << NV50_TIC_0_MAPB__SHIFT) |
(swz[3] << NV50_TIC_0_MAPA__SHIFT);
addr = mt->base.address;
if (mt->base.base.target == PIPE_TEXTURE_1D_ARRAY ||
mt->base.base.target == PIPE_TEXTURE_2D_ARRAY) {
addr += view->pipe.u.tex.first_layer * mt->layer_stride;
depth = view->pipe.u.tex.last_layer - view->pipe.u.tex.first_layer + 1;
} else {
depth = mt->base.base.depth0;
}
tic[2] = 0x10001000 | NV50_TIC_2_NO_BORDER;
if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
tic[2] |= NV50_TIC_2_COLORSPACE_SRGB;
if (!(flags & NV50_TEXVIEW_SCALED_COORDS))
tic[2] |= NV50_TIC_2_NORMALIZED_COORDS;
if (unlikely(!nouveau_bo_memtype(nv04_resource(texture)->bo))) {
if (target == PIPE_BUFFER) {
addr += view->pipe.u.buf.first_element * desc->block.bits / 8;
tic[2] |= NV50_TIC_2_LINEAR | NV50_TIC_2_TARGET_BUFFER;
tic[3] = 0;
tic[4] = /* width */
view->pipe.u.buf.last_element - view->pipe.u.buf.first_element + 1;
tic[5] = 0;
} else {
tic[2] |= NV50_TIC_2_LINEAR | NV50_TIC_2_TARGET_RECT;
tic[3] = mt->level[0].pitch;
tic[4] = mt->base.base.width0;
tic[5] = (1 << 16) | (mt->base.base.height0);
}
tic[6] =
tic[7] = 0;
tic[1] = addr;
tic[2] |= addr >> 32;
return &view->pipe;
}
tic[1] = addr;
tic[2] |= (addr >> 32) & 0xff;
tic[2] |=
((mt->level[0].tile_mode & 0x0f0) << (22 - 4)) |
((mt->level[0].tile_mode & 0xf00) << (25 - 8));
switch (target) {
case PIPE_TEXTURE_1D:
tic[2] |= NV50_TIC_2_TARGET_1D;
break;
case PIPE_TEXTURE_2D:
tic[2] |= NV50_TIC_2_TARGET_2D;
break;
case PIPE_TEXTURE_RECT:
tic[2] |= NV50_TIC_2_TARGET_RECT;
break;
case PIPE_TEXTURE_3D:
tic[2] |= NV50_TIC_2_TARGET_3D;
break;
case PIPE_TEXTURE_CUBE:
depth /= 6;
tic[2] |= NV50_TIC_2_TARGET_CUBE;
break;
case PIPE_TEXTURE_1D_ARRAY:
tic[2] |= NV50_TIC_2_TARGET_1D_ARRAY;
break;
case PIPE_TEXTURE_2D_ARRAY:
tic[2] |= NV50_TIC_2_TARGET_2D_ARRAY;
break;
case PIPE_TEXTURE_CUBE_ARRAY:
depth /= 6;
tic[2] |= NV50_TIC_2_TARGET_CUBE_ARRAY;
break;
case PIPE_BUFFER:
assert(0); /* should be linear and handled above ! */
tic[2] |= NV50_TIC_2_TARGET_BUFFER | NV50_TIC_2_LINEAR;
break;
default:
NOUVEAU_ERR("invalid texture target: %d\n", mt->base.base.target);
return FALSE;
}
tic[3] = (flags & NV50_TEXVIEW_FILTER_MSAA8) ? 0x20000000 : 0x00300000;
tic[4] = (1 << 31) | (mt->base.base.width0 << mt->ms_x);
tic[5] = (mt->base.base.height0 << mt->ms_y) & 0xffff;
tic[5] |= depth << 16;
tic[5] |= mt->base.base.last_level << NV50_TIC_5_LAST_LEVEL__SHIFT;
tic[6] = (mt->ms_x > 1) ? 0x88000000 : 0x03000000; /* sampling points */
tic[7] = (view->pipe.u.tex.last_level << 4) | view->pipe.u.tex.first_level;
if (unlikely(!(tic[2] & NV50_TIC_2_NORMALIZED_COORDS)))
if (mt->base.base.last_level)
tic[5] &= ~NV50_TIC_5_LAST_LEVEL__MASK;
return &view->pipe;
}
static int
nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
enum pipe_shader_cap param)
{
const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
switch (shader) {
case PIPE_SHADER_VERTEX:
case PIPE_SHADER_GEOMETRY:
case PIPE_SHADER_FRAGMENT:
break;
case PIPE_SHADER_TESS_CTRL:
case PIPE_SHADER_TESS_EVAL:
if (class_3d >= GM107_3D_CLASS)
return 0;
break;
case PIPE_SHADER_COMPUTE:
if (class_3d > NVE4_3D_CLASS)
return 0;
break;
default:
return 0;
}
switch (param) {
case PIPE_SHADER_CAP_PREFERRED_IR:
return PIPE_SHADER_IR_TGSI;
case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
return 16384;
case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
return 16;
case PIPE_SHADER_CAP_MAX_INPUTS:
if (shader == PIPE_SHADER_VERTEX)
return 32;
/* NOTE: These only count our slots for GENERIC varyings.
* The address space may be larger, but the actual hard limit seems to be
* less than what the address space layout permits, so don't add TEXCOORD,
* COLOR, etc. here.
*/
if (shader == PIPE_SHADER_FRAGMENT)
return 0x1f0 / 16;
/* Actually this counts CLIPVERTEX, which occupies the last generic slot,
* and excludes 0x60 per-patch inputs.
*/
return 0x200 / 16;
case PIPE_SHADER_CAP_MAX_OUTPUTS:
return 32;
case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
return 65536;
case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
if (shader == PIPE_SHADER_COMPUTE && class_3d >= NVE4_3D_CLASS)
return NVE4_MAX_PIPE_CONSTBUFS_COMPUTE;
return NVC0_MAX_PIPE_CONSTBUFS;
case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
return shader != PIPE_SHADER_FRAGMENT;
case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
return 1;
case PIPE_SHADER_CAP_MAX_PREDS:
return 0;
case PIPE_SHADER_CAP_MAX_TEMPS:
return NVC0_CAP_MAX_PROGRAM_TEMPS;
case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
return 1;
case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
return 0;
case PIPE_SHADER_CAP_SUBROUTINES:
return 1;
case PIPE_SHADER_CAP_INTEGERS:
return 1;
case PIPE_SHADER_CAP_DOUBLES:
return 1;
case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
return 1;
case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
return 0;
case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
return 16; /* would be 32 in linked (OpenGL-style) mode */
case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
return 16; /* XXX not sure if more are really safe */
case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
return 32;
default:
NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param);
return 0;
}
}
struct pipe_screen *
nv30_screen_create(struct nouveau_device *dev)
{
struct nv30_screen *screen = CALLOC_STRUCT(nv30_screen);
struct pipe_screen *pscreen;
struct nouveau_pushbuf *push;
struct nv04_fifo *fifo;
unsigned oclass = 0;
int ret, i;
if (!screen)
return NULL;
switch (dev->chipset & 0xf0) {
case 0x30:
if (RANKINE_0397_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV30_3D_CLASS;
else
if (RANKINE_0697_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV34_3D_CLASS;
else
if (RANKINE_0497_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV35_3D_CLASS;
break;
case 0x40:
if (CURIE_4097_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV40_3D_CLASS;
else
if (CURIE_4497_CHIPSET & (1 << (dev->chipset & 0x0f)))
oclass = NV44_3D_CLASS;
break;
case 0x60:
if (CURIE_4497_CHIPSET6X & (1 << (dev->chipset & 0x0f)))
oclass = NV44_3D_CLASS;
break;
default:
break;
}
if (!oclass) {
NOUVEAU_ERR("unknown 3d class for 0x%02x\n", dev->chipset);
FREE(screen);
return NULL;
}
pscreen = &screen->base.base;
pscreen->destroy = nv30_screen_destroy;
pscreen->get_param = nv30_screen_get_param;
pscreen->get_paramf = nv30_screen_get_paramf;
pscreen->get_shader_param = nv30_screen_get_shader_param;
pscreen->context_create = nv30_context_create;
pscreen->is_format_supported = nv30_screen_is_format_supported;
nv30_resource_screen_init(pscreen);
nouveau_screen_init_vdec(&screen->base);
screen->base.fence.emit = nv30_screen_fence_emit;
screen->base.fence.update = nv30_screen_fence_update;
ret = nouveau_screen_init(&screen->base, dev);
if (ret)
FAIL_SCREEN_INIT("nv30_screen_init failed: %d\n", ret);
screen->base.vidmem_bindings |= PIPE_BIND_VERTEX_BUFFER;
screen->base.sysmem_bindings |= PIPE_BIND_VERTEX_BUFFER;
if (oclass == NV40_3D_CLASS) {
screen->base.vidmem_bindings |= PIPE_BIND_INDEX_BUFFER;
screen->base.sysmem_bindings |= PIPE_BIND_INDEX_BUFFER;
}
fifo = screen->base.channel->data;
push = screen->base.pushbuf;
push->rsvd_kick = 16;
ret = nouveau_object_new(screen->base.channel, 0x00000000, NV01_NULL_CLASS,
NULL, 0, &screen->null);
if (ret)
FAIL_SCREEN_INIT("error allocating null object: %d\n", ret);
/* DMA_FENCE refuses to accept DMA objects with "adjust" filled in,
* this means that the address pointed at by the DMA object must
* be 4KiB aligned, which means this object needs to be the first
* one allocated on the channel.
*/
ret = nouveau_object_new(screen->base.channel, 0xbeef1e00,
NOUVEAU_NOTIFIER_CLASS, &(struct nv04_notify) {
.length = 32 }, sizeof(struct nv04_notify),
boolean
nv50_program_translate(struct nv50_program *prog, uint16_t chipset)
{
struct nv50_ir_prog_info *info;
int ret;
const uint8_t map_undef = (prog->type == PIPE_SHADER_VERTEX) ? 0x40 : 0x80;
info = CALLOC_STRUCT(nv50_ir_prog_info);
if (!info)
return FALSE;
info->type = prog->type;
info->target = chipset;
info->bin.sourceRep = NV50_PROGRAM_IR_TGSI;
info->bin.source = (void *)prog->pipe.tokens;
info->io.ucpBinding = 15;
info->io.ucpBase = 0;
info->io.genUserClip = prog->vp.clpd_nr;
info->assignSlots = nv50_program_assign_varying_slots;
prog->vp.bfc[0] = 0xff;
prog->vp.bfc[1] = 0xff;
prog->vp.edgeflag = 0xff;
prog->vp.clpd[0] = map_undef;
prog->vp.clpd[1] = map_undef;
prog->vp.psiz = map_undef;
prog->gp.primid = 0x80;
info->driverPriv = prog;
#ifdef DEBUG
info->optLevel = debug_get_num_option("NV50_PROG_OPTIMIZE", 3);
info->dbgFlags = debug_get_num_option("NV50_PROG_DEBUG", 0);
#else
info->optLevel = 3;
#endif
ret = nv50_ir_generate_code(info);
if (ret) {
NOUVEAU_ERR("shader translation failed: %i\n", ret);
goto out;
}
FREE(info->bin.syms);
prog->code = info->bin.code;
prog->code_size = info->bin.codeSize;
prog->fixups = info->bin.relocData;
prog->max_gpr = MAX2(4, (info->bin.maxGPR >> 1) + 1);
prog->tls_space = info->bin.tlsSpace;
if (prog->type == PIPE_SHADER_FRAGMENT) {
if (info->prop.fp.writesDepth) {
prog->fp.flags[0] |= NV50_3D_FP_CONTROL_EXPORTS_Z;
prog->fp.flags[1] = 0x11;
}
if (info->prop.fp.usesDiscard)
prog->fp.flags[0] |= NV50_3D_FP_CONTROL_USES_KIL;
}
if (prog->pipe.stream_output.num_outputs)
prog->so = nv50_program_create_strmout_state(info,
&prog->pipe.stream_output);
out:
FREE(info);
return !ret;
}
static int
nvc0_2d_texture_set(struct nouveau_pushbuf *push, boolean dst,
struct nv50_miptree *mt, unsigned level, unsigned layer)
{
struct nouveau_bo *bo = mt->base.bo;
uint32_t width, height, depth;
uint32_t format;
uint32_t mthd = dst ? NVC0_2D_DST_FORMAT : NVC0_2D_SRC_FORMAT;
uint32_t offset = mt->level[level].offset;
format = nvc0_2d_format(mt->base.base.format);
if (!format) {
NOUVEAU_ERR("invalid/unsupported surface format: %s\n",
util_format_name(mt->base.base.format));
return 1;
}
width = u_minify(mt->base.base.width0, level) << mt->ms_x;
height = u_minify(mt->base.base.height0, level) << mt->ms_y;
depth = u_minify(mt->base.base.depth0, level);
/* layer has to be < depth, and depth > tile depth / 2 */
if (!mt->layout_3d) {
offset += mt->layer_stride * layer;
layer = 0;
depth = 1;
} else
if (!dst) {
offset += nvc0_mt_zslice_offset(mt, level, layer);
layer = 0;
}
if (nouveau_bo_memtype(bo)) {
BEGIN_NVC0(push, SUBC_2D(mthd), 2);
PUSH_DATA (push, format);
PUSH_DATA (push, 1);
BEGIN_NVC0(push, SUBC_2D(mthd + 0x14), 5);
PUSH_DATA (push, mt->level[level].pitch);
PUSH_DATA (push, width);
PUSH_DATA (push, height);
PUSH_DATAh(push, bo->offset + offset);
PUSH_DATA (push, bo->offset + offset);
} else {
BEGIN_NVC0(push, SUBC_2D(mthd), 5);
PUSH_DATA (push, format);
PUSH_DATA (push, 0);
PUSH_DATA (push, mt->level[level].tile_mode);
PUSH_DATA (push, depth);
PUSH_DATA (push, layer);
BEGIN_NVC0(push, SUBC_2D(mthd + 0x18), 4);
PUSH_DATA (push, width);
PUSH_DATA (push, height);
PUSH_DATAh(push, bo->offset + offset);
PUSH_DATA (push, bo->offset + offset);
}
#if 0
if (dst) {
BEGIN_NVC0(push, SUBC_2D(NVC0_2D_CLIP_X), 4);
PUSH_DATA (push, 0);
PUSH_DATA (push, 0);
PUSH_DATA (push, width);
PUSH_DATA (push, height);
}
#endif
return 0;
}
static int
nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
{
const uint16_t class_3d = nouveau_screen(pscreen)->class_3d;
struct nouveau_device *dev = nouveau_screen(pscreen)->device;
switch (param) {
/* non-boolean caps */
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
return 14;
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
return 12;
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
return 14;
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
return 512;
case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MIN_TEXEL_OFFSET:
return -8;
case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MAX_TEXEL_OFFSET:
return 7;
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
return 128 * 1024 * 1024;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
return 330;
case PIPE_CAP_MAX_RENDER_TARGETS:
return 8;
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
return 1;
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
return 4;
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
return 64;
case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
return 1024;
case PIPE_CAP_MAX_VERTEX_STREAMS:
return 1;
case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
return 2048;
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
return 256;
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
return 16; /* 256 for binding as RT, but that's not possible in GL */
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
return NOUVEAU_MIN_BUFFER_MAP_ALIGN;
case PIPE_CAP_MAX_VIEWPORTS:
return NV50_MAX_VIEWPORTS;
case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50;
case PIPE_CAP_ENDIANNESS:
return PIPE_ENDIAN_LITTLE;
case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
return (class_3d >= NVA3_3D_CLASS) ? 4 : 0;
case PIPE_CAP_MAX_WINDOW_RECTANGLES:
return NV50_MAX_WINDOW_RECTANGLES;
/* supported caps */
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
case PIPE_CAP_TEXTURE_SWIZZLE:
case PIPE_CAP_TEXTURE_SHADOW_MAP:
case PIPE_CAP_NPOT_TEXTURES:
case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
case PIPE_CAP_ANISOTROPIC_FILTER:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
case PIPE_CAP_TWO_SIDED_STENCIL:
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_POINT_SPRITE:
case PIPE_CAP_SM3:
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
case PIPE_CAP_QUERY_TIMESTAMP:
case PIPE_CAP_QUERY_TIME_ELAPSED:
case PIPE_CAP_OCCLUSION_QUERY:
case PIPE_CAP_BLEND_EQUATION_SEPARATE:
case PIPE_CAP_INDEP_BLEND_ENABLE:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
case PIPE_CAP_PRIMITIVE_RESTART:
case PIPE_CAP_TGSI_INSTANCEID:
case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
case PIPE_CAP_CONDITIONAL_RENDER:
case PIPE_CAP_TEXTURE_BARRIER:
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_START_INSTANCE:
case PIPE_CAP_USER_CONSTANT_BUFFERS:
case PIPE_CAP_USER_INDEX_BUFFERS:
case PIPE_CAP_USER_VERTEX_BUFFERS:
case PIPE_CAP_TEXTURE_MULTISAMPLE:
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
case PIPE_CAP_SAMPLER_VIEW_TARGET:
case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
case PIPE_CAP_CLIP_HALFZ:
case PIPE_CAP_POLYGON_OFFSET_CLAMP:
case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
case PIPE_CAP_DEPTH_BOUNDS_TEST:
case PIPE_CAP_TGSI_TXQS:
case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
case PIPE_CAP_SHAREABLE_SHADERS:
case PIPE_CAP_CLEAR_TEXTURE:
case PIPE_CAP_COMPUTE:
case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
case PIPE_CAP_INVALIDATE_BUFFER:
case PIPE_CAP_STRING_MARKER:
case PIPE_CAP_CULL_DISTANCE:
return 1;
case PIPE_CAP_SEAMLESS_CUBE_MAP:
return 1; /* class_3d >= NVA0_3D_CLASS; */
/* supported on nva0+ */
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
return class_3d >= NVA0_3D_CLASS;
/* supported on nva3+ */
case PIPE_CAP_CUBE_MAP_ARRAY:
case PIPE_CAP_INDEP_BLEND_FUNC:
case PIPE_CAP_TEXTURE_QUERY_LOD:
case PIPE_CAP_SAMPLE_SHADING:
case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
return class_3d >= NVA3_3D_CLASS;
/* unsupported caps */
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
case PIPE_CAP_SHADER_STENCIL_EXPORT:
case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_TGSI_TEXCOORD:
case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
case PIPE_CAP_TEXTURE_GATHER_SM5:
case PIPE_CAP_FAKE_SW_MSAA:
case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
case PIPE_CAP_DRAW_INDIRECT:
case PIPE_CAP_MULTI_DRAW_INDIRECT:
case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
case PIPE_CAP_VERTEXID_NOBASE:
case PIPE_CAP_MULTISAMPLE_Z_RESOLVE: /* potentially supported on some hw */
case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
case PIPE_CAP_DRAW_PARAMETERS:
case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_GENERATE_MIPMAP:
case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
case PIPE_CAP_QUERY_BUFFER_OBJECT:
case PIPE_CAP_QUERY_MEMORY_INFO:
case PIPE_CAP_PCI_GROUP:
case PIPE_CAP_PCI_BUS:
case PIPE_CAP_PCI_DEVICE:
case PIPE_CAP_PCI_FUNCTION:
case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
case PIPE_CAP_TGSI_VOTE:
case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
return 0;
case PIPE_CAP_VENDOR_ID:
return 0x10de;
case PIPE_CAP_DEVICE_ID: {
uint64_t device_id;
if (nouveau_getparam(dev, NOUVEAU_GETPARAM_PCI_DEVICE, &device_id)) {
NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
return -1;
}
return device_id;
}
case PIPE_CAP_ACCELERATED:
return 1;
case PIPE_CAP_VIDEO_MEMORY:
return dev->vram_size >> 20;
case PIPE_CAP_UMA:
return 0;
}
NOUVEAU_ERR("unknown PIPE_CAP %d\n", param);
return 0;
}
static int
nvfx_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
{
struct nvfx_screen *screen = nvfx_screen(pscreen);
switch (param) {
case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
return 16;
case PIPE_CAP_NPOT_TEXTURES:
return screen->advertise_npot;
case PIPE_CAP_TWO_SIDED_STENCIL:
return 1;
case PIPE_CAP_GLSL:
return 1;
case PIPE_CAP_ANISOTROPIC_FILTER:
return 1;
case PIPE_CAP_POINT_SPRITE:
return 1;
case PIPE_CAP_MAX_RENDER_TARGETS:
return screen->use_nv4x ? 4 : 1;
case PIPE_CAP_OCCLUSION_QUERY:
return 1;
case PIPE_CAP_TIMER_QUERY:
return 0;
case PIPE_CAP_TEXTURE_SHADOW_MAP:
return 1;
case PIPE_CAP_TEXTURE_SWIZZLE:
return 1;
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
return 13;
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
return 10;
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
return 13;
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
return !!screen->use_nv4x;
case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
return 1;
case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
return 0; /* We have 4 on nv40 - but unsupported currently */
case PIPE_CAP_BLEND_EQUATION_SEPARATE:
return screen->advertise_blend_equation_separate;
case PIPE_CAP_MAX_COMBINED_SAMPLERS:
return 16;
case PIPE_CAP_INDEP_BLEND_ENABLE:
/* TODO: on nv40 we have separate color masks */
/* TODO: nv40 mrt blending is probably broken */
return 0;
case PIPE_CAP_INDEP_BLEND_FUNC:
return 0;
case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
return 0;
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
return 1;
case PIPE_CAP_DEPTH_CLAMP:
return 0; // TODO: implement depth clamp
case PIPE_CAP_PRIMITIVE_RESTART:
return 0; // TODO: implement primitive restart
case PIPE_CAP_SHADER_STENCIL_EXPORT:
return 0;
default:
NOUVEAU_ERR("Warning: unknown PIPE_CAP %d\n", param);
return 0;
}
}
struct pipe_sampler_view *
nvc0_create_texture_view(struct pipe_context *pipe,
struct pipe_resource *texture,
const struct pipe_sampler_view *templ,
uint32_t flags,
enum pipe_texture_target target)
{
const struct util_format_description *desc;
uint64_t address;
uint32_t *tic;
uint32_t swz[4];
uint32_t width, height;
uint32_t depth;
struct nv50_tic_entry *view;
struct nv50_miptree *mt;
boolean tex_int;
view = MALLOC_STRUCT(nv50_tic_entry);
if (!view)
return NULL;
mt = nv50_miptree(texture);
view->pipe = *templ;
view->pipe.reference.count = 1;
view->pipe.texture = NULL;
view->pipe.context = pipe;
view->id = -1;
pipe_resource_reference(&view->pipe.texture, texture);
tic = &view->tic[0];
desc = util_format_description(view->pipe.format);
tic[0] = nvc0_format_table[view->pipe.format].tic;
tex_int = util_format_is_pure_integer(view->pipe.format);
swz[0] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_r, tex_int);
swz[1] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_g, tex_int);
swz[2] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_b, tex_int);
swz[3] = nv50_tic_swizzle(tic[0], view->pipe.swizzle_a, tex_int);
tic[0] = (tic[0] & ~NV50_TIC_0_SWIZZLE__MASK) |
(swz[0] << NV50_TIC_0_MAPR__SHIFT) |
(swz[1] << NV50_TIC_0_MAPG__SHIFT) |
(swz[2] << NV50_TIC_0_MAPB__SHIFT) |
(swz[3] << NV50_TIC_0_MAPA__SHIFT);
address = mt->base.address;
tic[2] = 0x10001000 | NV50_TIC_2_NO_BORDER;
if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
tic[2] |= NV50_TIC_2_COLORSPACE_SRGB;
if (!(flags & NV50_TEXVIEW_SCALED_COORDS))
tic[2] |= NV50_TIC_2_NORMALIZED_COORDS;
/* check for linear storage type */
if (unlikely(!nouveau_bo_memtype(nv04_resource(texture)->bo))) {
if (texture->target == PIPE_BUFFER) {
assert(!(tic[2] & NV50_TIC_2_NORMALIZED_COORDS));
address +=
view->pipe.u.buf.first_element * desc->block.bits / 8;
tic[2] |= NV50_TIC_2_LINEAR | NV50_TIC_2_TARGET_BUFFER;
tic[3] = 0;
tic[4] = /* width */
view->pipe.u.buf.last_element - view->pipe.u.buf.first_element + 1;
tic[5] = 0;
} else {
/* must be 2D texture without mip maps */
tic[2] |= NV50_TIC_2_LINEAR | NV50_TIC_2_TARGET_RECT;
tic[3] = mt->level[0].pitch;
tic[4] = mt->base.base.width0;
tic[5] = (1 << 16) | mt->base.base.height0;
}
tic[6] =
tic[7] = 0;
tic[1] = address;
tic[2] |= address >> 32;
return &view->pipe;
}
tic[2] |=
((mt->level[0].tile_mode & 0x0f0) << (22 - 4)) |
((mt->level[0].tile_mode & 0xf00) << (25 - 8));
depth = MAX2(mt->base.base.array_size, mt->base.base.depth0);
if (mt->base.base.array_size > 1) {
/* there doesn't seem to be a base layer field in TIC */
address += view->pipe.u.tex.first_layer * mt->layer_stride;
depth = view->pipe.u.tex.last_layer - view->pipe.u.tex.first_layer + 1;
}
tic[1] = address;
tic[2] |= address >> 32;
switch (target) {
case PIPE_TEXTURE_1D:
tic[2] |= NV50_TIC_2_TARGET_1D;
break;
case PIPE_TEXTURE_2D:
tic[2] |= NV50_TIC_2_TARGET_2D;
break;
case PIPE_TEXTURE_RECT:
tic[2] |= NV50_TIC_2_TARGET_2D;
break;
case PIPE_TEXTURE_3D:
tic[2] |= NV50_TIC_2_TARGET_3D;
break;
case PIPE_TEXTURE_CUBE:
depth /= 6;
tic[2] |= NV50_TIC_2_TARGET_CUBE;
break;
case PIPE_TEXTURE_1D_ARRAY:
tic[2] |= NV50_TIC_2_TARGET_1D_ARRAY;
break;
case PIPE_TEXTURE_2D_ARRAY:
tic[2] |= NV50_TIC_2_TARGET_2D_ARRAY;
break;
case PIPE_TEXTURE_CUBE_ARRAY:
depth /= 6;
tic[2] |= NV50_TIC_2_TARGET_CUBE_ARRAY;
break;
default:
NOUVEAU_ERR("unexpected/invalid texture target: %d\n",
mt->base.base.target);
return FALSE;
}
tic[3] = (flags & NV50_TEXVIEW_FILTER_MSAA8) ? 0x20000000 : 0x00300000;
if (flags & NV50_TEXVIEW_ACCESS_RESOLVE) {
width = mt->base.base.width0 << mt->ms_x;
height = mt->base.base.height0 << mt->ms_y;
} else {
width = mt->base.base.width0;
height = mt->base.base.height0;
}
tic[4] = (1 << 31) | width;
tic[5] = height & 0xffff;
tic[5] |= depth << 16;
tic[5] |= mt->base.base.last_level << 28;
/* sampling points: (?) */
if (flags & NV50_TEXVIEW_ACCESS_RESOLVE)
tic[6] = (mt->ms_x > 1) ? 0x88000000 : 0x03000000;
else
tic[6] = 0x03000000;
tic[7] = (view->pipe.u.tex.last_level << 4) | view->pipe.u.tex.first_level;
tic[7] |= mt->ms_mode << 12;
return &view->pipe;
}
struct pipe_screen *
nvfx_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
{
static const unsigned query_sizes[] = {(4096 - 4 * 32) / 32, 3 * 1024 / 32, 2 * 1024 / 32, 1024 / 32};
struct nvfx_screen *screen = CALLOC_STRUCT(nvfx_screen);
struct nouveau_channel *chan;
struct pipe_screen *pscreen;
unsigned eng3d_class = 0;
int ret, i;
if (!screen)
return NULL;
pscreen = &screen->base.base;
ret = nouveau_screen_init(&screen->base, dev);
if (ret) {
nvfx_screen_destroy(pscreen);
return NULL;
}
chan = screen->base.channel;
screen->cur_ctx = NULL;
chan->user_private = screen;
chan->flush_notify = nvfx_channel_flush_notify;
pscreen->winsys = ws;
pscreen->destroy = nvfx_screen_destroy;
pscreen->get_param = nvfx_screen_get_param;
pscreen->get_shader_param = nvfx_screen_get_shader_param;
pscreen->get_paramf = nvfx_screen_get_paramf;
pscreen->is_format_supported = nvfx_screen_is_format_supported;
pscreen->context_create = nvfx_create;
switch (dev->chipset & 0xf0) {
case 0x30:
if (NV30_3D_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f)))
eng3d_class = NV30_3D;
else if (NV34_3D_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f)))
eng3d_class = NV34_3D;
else if (NV35_3D_CHIPSET_3X_MASK & (1 << (dev->chipset & 0x0f)))
eng3d_class = NV35_3D;
break;
case 0x40:
if (NV4X_GRCLASS4097_CHIPSETS & (1 << (dev->chipset & 0x0f)))
eng3d_class = NV40_3D;
else if (NV4X_GRCLASS4497_CHIPSETS & (1 << (dev->chipset & 0x0f)))
eng3d_class = NV44_3D;
screen->is_nv4x = ~0;
break;
case 0x60:
if (NV6X_GRCLASS4497_CHIPSETS & (1 << (dev->chipset & 0x0f)))
eng3d_class = NV44_3D;
screen->is_nv4x = ~0;
break;
}
if (!eng3d_class) {
NOUVEAU_ERR("Unknown nv3x/nv4x chipset: nv%02x\n", dev->chipset);
return NULL;
}
screen->advertise_npot = !!screen->is_nv4x;
screen->advertise_blend_equation_separate = !!screen->is_nv4x;
screen->use_nv4x = screen->is_nv4x;
if(screen->is_nv4x) {
if(debug_get_bool_option("NVFX_SIMULATE_NV30", FALSE))
screen->use_nv4x = 0;
if(!debug_get_bool_option("NVFX_NPOT", TRUE))
screen->advertise_npot = 0;
if(!debug_get_bool_option("NVFX_BLEND_EQ_SEP", TRUE))
screen->advertise_blend_equation_separate = 0;
}
screen->force_swtnl = debug_get_bool_option("NVFX_SWTNL", FALSE);
screen->trace_draw = debug_get_bool_option("NVFX_TRACE_DRAW", FALSE);
screen->buffer_allocation_cost = debug_get_num_option("NVFX_BUFFER_ALLOCATION_COST", 16384);
screen->inline_cost_per_hardware_cost = atof(debug_get_option("NVFX_INLINE_COST_PER_HARDWARE_COST", "1.0"));
screen->static_reuse_threshold = atof(debug_get_option("NVFX_STATIC_REUSE_THRESHOLD", "2.0"));
/* We don't advertise these by default because filtering and blending doesn't work as
* it should, due to several restrictions.
* The only exception is fp16 on nv40.
*/
screen->advertise_fp16 = debug_get_bool_option("NVFX_FP16", !!screen->use_nv4x);
screen->advertise_fp32 = debug_get_bool_option("NVFX_FP32", 0);
screen->vertex_buffer_reloc_flags = nvfx_screen_get_vertex_buffer_flags(screen);
/* surely both nv3x and nv44 support index buffers too: find out how and test that */
if(eng3d_class == NV40_3D)
screen->index_buffer_reloc_flags = screen->vertex_buffer_reloc_flags;
if(!screen->force_swtnl && screen->vertex_buffer_reloc_flags == screen->index_buffer_reloc_flags)
screen->base.vertex_buffer_flags = screen->base.index_buffer_flags = screen->vertex_buffer_reloc_flags;
nvfx_screen_init_resource_functions(pscreen);
ret = nouveau_grobj_alloc(chan, 0xbeef3097, eng3d_class, &screen->eng3d);
if (ret) {
NOUVEAU_ERR("Error creating 3D object: %d\n", ret);
return FALSE;
}
/* 2D engine setup */
nvfx_screen_surface_init(pscreen);
/* Notifier for sync purposes */
ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync);
if (ret) {
NOUVEAU_ERR("Error creating notifier object: %d\n", ret);
nvfx_screen_destroy(pscreen);
return NULL;
}
/* Query objects */
for(i = 0; i < sizeof(query_sizes) / sizeof(query_sizes[0]); ++i)
{
ret = nouveau_notifier_alloc(chan, 0xbeef0302, query_sizes[i], &screen->query);
if(!ret)
break;
}
if (ret) {
NOUVEAU_ERR("Error initialising query objects: %d\n", ret);
nvfx_screen_destroy(pscreen);
return NULL;
}
ret = nouveau_resource_init(&screen->query_heap, 0, query_sizes[i]);
if (ret) {
NOUVEAU_ERR("Error initialising query object heap: %d\n", ret);
nvfx_screen_destroy(pscreen);
return NULL;
}
LIST_INITHEAD(&screen->query_list);
/* Vtxprog resources */
if (nouveau_resource_init(&screen->vp_exec_heap, 0, screen->use_nv4x ? 512 : 256) ||
nouveau_resource_init(&screen->vp_data_heap, 0, screen->use_nv4x ? 468 : 256)) {
nvfx_screen_destroy(pscreen);
return NULL;
}
BIND_RING(chan, screen->eng3d, 7);
/* Static eng3d initialisation */
/* note that we just started using the channel, so we must have space in the pushbuffer */
OUT_RING(chan, RING_3D(NV30_3D_DMA_NOTIFY, 1));
OUT_RING(chan, screen->sync->handle);
OUT_RING(chan, RING_3D(NV30_3D_DMA_TEXTURE0, 2));
OUT_RING(chan, chan->vram->handle);
OUT_RING(chan, chan->gart->handle);
OUT_RING(chan, RING_3D(NV30_3D_DMA_COLOR1, 1));
OUT_RING(chan, chan->vram->handle);
OUT_RING(chan, RING_3D(NV30_3D_DMA_COLOR0, 2));
OUT_RING(chan, chan->vram->handle);
OUT_RING(chan, chan->vram->handle);
OUT_RING(chan, RING_3D(NV30_3D_DMA_VTXBUF0, 2));
OUT_RING(chan, chan->vram->handle);
OUT_RING(chan, chan->gart->handle);
OUT_RING(chan, RING_3D(NV30_3D_DMA_FENCE, 2));
OUT_RING(chan, 0);
OUT_RING(chan, screen->query->handle);
OUT_RING(chan, RING_3D(NV30_3D_DMA_UNK1AC, 2));
OUT_RING(chan, chan->vram->handle);
OUT_RING(chan, chan->vram->handle);
if(!screen->is_nv4x)
nv30_screen_init(screen);
else
nv40_screen_init(screen);
return pscreen;
}
struct pipe_screen *
nv50_screen_create(struct nouveau_device *dev)
{
struct nv50_screen *screen;
struct pipe_screen *pscreen;
struct nouveau_object *chan;
uint64_t value;
uint32_t tesla_class;
unsigned stack_size;
int ret;
screen = CALLOC_STRUCT(nv50_screen);
if (!screen)
return NULL;
pscreen = &screen->base.base;
ret = nouveau_screen_init(&screen->base, dev);
if (ret) {
NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret);
goto fail;
}
/* TODO: Prevent FIFO prefetch before transfer of index buffers and
* admit them to VRAM.
*/
screen->base.vidmem_bindings |= PIPE_BIND_CONSTANT_BUFFER |
PIPE_BIND_VERTEX_BUFFER;
screen->base.sysmem_bindings |=
PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER;
screen->base.pushbuf->user_priv = screen;
screen->base.pushbuf->rsvd_kick = 5;
chan = screen->base.channel;
pscreen->destroy = nv50_screen_destroy;
pscreen->context_create = nv50_create;
pscreen->is_format_supported = nv50_screen_is_format_supported;
pscreen->get_param = nv50_screen_get_param;
pscreen->get_shader_param = nv50_screen_get_shader_param;
pscreen->get_paramf = nv50_screen_get_paramf;
nv50_screen_init_resource_functions(pscreen);
if (screen->base.device->chipset < 0x84 ||
debug_get_bool_option("NOUVEAU_PMPEG", FALSE)) {
/* PMPEG */
nouveau_screen_init_vdec(&screen->base);
} else if (screen->base.device->chipset < 0x98 ||
screen->base.device->chipset == 0xa0) {
/* VP2 */
screen->base.base.get_video_param = nv84_screen_get_video_param;
screen->base.base.is_video_format_supported = nv84_screen_video_supported;
} else {
/* VP3/4 */
screen->base.base.get_video_param = nouveau_vp3_screen_get_video_param;
screen->base.base.is_video_format_supported = nouveau_vp3_screen_video_supported;
}
ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0, 4096,
NULL, &screen->fence.bo);
if (ret) {
NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret);
goto fail;
}
nouveau_bo_map(screen->fence.bo, 0, NULL);
screen->fence.map = screen->fence.bo->map;
screen->base.fence.emit = nv50_screen_fence_emit;
screen->base.fence.update = nv50_screen_fence_update;
ret = nouveau_object_new(chan, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS,
&(struct nv04_notify){ .length = 32 },
sizeof(struct nv04_notify), &screen->sync);
void
nvc0_launch_grid(struct pipe_context *pipe,
const uint *block_layout, const uint *grid_layout,
uint32_t label,
const void *input)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nvc0_program *cp = nvc0->compprog;
unsigned s, i;
int ret;
ret = !nvc0_compute_state_validate(nvc0);
if (ret)
goto out;
nvc0_compute_upload_input(nvc0, input);
BEGIN_NVC0(push, NVC0_COMPUTE(CP_START_ID), 1);
PUSH_DATA (push, nvc0_program_symbol_offset(cp, label));
BEGIN_NVC0(push, NVC0_COMPUTE(LOCAL_POS_ALLOC), 3);
PUSH_DATA (push, align(cp->cp.lmem_size, 0x10));
PUSH_DATA (push, 0);
PUSH_DATA (push, 0x800); /* WARP_CSTACK_SIZE */
BEGIN_NVC0(push, NVC0_COMPUTE(SHARED_SIZE), 3);
PUSH_DATA (push, align(cp->cp.smem_size, 0x100));
PUSH_DATA (push, block_layout[0] * block_layout[1] * block_layout[2]);
PUSH_DATA (push, cp->num_barriers);
BEGIN_NVC0(push, NVC0_COMPUTE(CP_GPR_ALLOC), 1);
PUSH_DATA (push, cp->num_gprs);
/* grid/block setup */
BEGIN_NVC0(push, NVC0_COMPUTE(GRIDDIM_YX), 2);
PUSH_DATA (push, (grid_layout[1] << 16) | grid_layout[0]);
PUSH_DATA (push, grid_layout[2]);
BEGIN_NVC0(push, NVC0_COMPUTE(BLOCKDIM_YX), 2);
PUSH_DATA (push, (block_layout[1] << 16) | block_layout[0]);
PUSH_DATA (push, block_layout[2]);
/* launch preliminary setup */
BEGIN_NVC0(push, NVC0_COMPUTE(GRIDID), 1);
PUSH_DATA (push, 0x1);
BEGIN_NVC0(push, SUBC_COMPUTE(0x036c), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, NVC0_COMPUTE(FLUSH), 1);
PUSH_DATA (push, NVC0_COMPUTE_FLUSH_GLOBAL | NVC0_COMPUTE_FLUSH_UNK8);
/* kernel launching */
BEGIN_NVC0(push, NVC0_COMPUTE(COMPUTE_BEGIN), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, SUBC_COMPUTE(0x0a08), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, NVC0_COMPUTE(LAUNCH), 1);
PUSH_DATA (push, 0x1000);
BEGIN_NVC0(push, NVC0_COMPUTE(COMPUTE_END), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, SUBC_COMPUTE(0x0360), 1);
PUSH_DATA (push, 0x1);
/* rebind all the 3D constant buffers
* (looks like binding a CB on COMPUTE clobbers 3D state) */
nvc0->dirty |= NVC0_NEW_CONSTBUF;
for (s = 0; s < 6; s++) {
for (i = 0; i < NVC0_MAX_PIPE_CONSTBUFS; i++)
if (nvc0->constbuf[s][i].u.buf)
nvc0->constbuf_dirty[s] |= 1 << i;
}
memset(nvc0->state.uniform_buffer_bound, 0,
sizeof(nvc0->state.uniform_buffer_bound));
out:
if (ret)
NOUVEAU_ERR("Failed to launch grid !\n");
}
