void
nv30_sampler_view_init(struct pipe_context *pipe,
struct nvfx_sampler_view *sv)
{
struct pipe_resource* pt = sv->base.texture;
struct nvfx_texture_format *tf = &nvfx_texture_formats[sv->base.format];
unsigned txf;
unsigned level = pt->target == PIPE_TEXTURE_CUBE ? 0 : sv->base.u.tex.first_level;
assert(tf->fmt[0] >= 0);
txf = sv->u.init_fmt;
txf |= (level != sv->base.u.tex.last_level ? NV30_3D_TEX_FORMAT_MIPMAP : 0);
txf |= util_logbase2(u_minify(pt->width0, level)) << NV30_3D_TEX_FORMAT_BASE_SIZE_U__SHIFT;
txf |= util_logbase2(u_minify(pt->height0, level)) << NV30_3D_TEX_FORMAT_BASE_SIZE_V__SHIFT;
txf |= util_logbase2(u_minify(pt->depth0, level)) << NV30_3D_TEX_FORMAT_BASE_SIZE_W__SHIFT;
txf |= 0x10000;
sv->u.nv30.fmt[0] = tf->fmt[0] | txf;
sv->u.nv30.fmt[1] = tf->fmt[1] | txf;
sv->u.nv30.fmt[2] = tf->fmt[2] | txf;
sv->u.nv30.fmt[3] = tf->fmt[3] | txf;
sv->swizzle |= (nvfx_subresource_pitch(pt, 0) << NV30_3D_TEX_SWIZZLE_RECT_PITCH__SHIFT);
if(pt->height0 <= 1 || util_format_is_compressed(sv->base.format))
sv->u.nv30.rect = -1;
else
sv->u.nv30.rect = !!(pt->flags & NVFX_RESOURCE_FLAG_LINEAR);
sv->lod_offset = sv->base.u.tex.first_level - level;
sv->max_lod_limit = sv->base.u.tex.last_level - level;
}
static void amdgpu_buffer_set_metadata(struct pb_buffer *_buf,
struct radeon_bo_metadata *md)
{
struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(_buf);
struct amdgpu_bo_metadata metadata = {0};
uint32_t tiling_flags = 0;
if (md->macrotile == RADEON_LAYOUT_TILED)
tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 4); /* 2D_TILED_THIN1 */
else if (md->microtile == RADEON_LAYOUT_TILED)
tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 2); /* 1D_TILED_THIN1 */
else
tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 1); /* LINEAR_ALIGNED */
tiling_flags |= AMDGPU_TILING_SET(PIPE_CONFIG, md->pipe_config);
tiling_flags |= AMDGPU_TILING_SET(BANK_WIDTH, util_logbase2(md->bankw));
tiling_flags |= AMDGPU_TILING_SET(BANK_HEIGHT, util_logbase2(md->bankh));
if (md->tile_split)
tiling_flags |= AMDGPU_TILING_SET(TILE_SPLIT, eg_tile_split_rev(md->tile_split));
tiling_flags |= AMDGPU_TILING_SET(MACRO_TILE_ASPECT, util_logbase2(md->mtilea));
tiling_flags |= AMDGPU_TILING_SET(NUM_BANKS, util_logbase2(md->num_banks)-1);
if (md->scanout)
tiling_flags |= AMDGPU_TILING_SET(MICRO_TILE_MODE, 0); /* DISPLAY_MICRO_TILING */
else
tiling_flags |= AMDGPU_TILING_SET(MICRO_TILE_MODE, 1); /* THIN_MICRO_TILING */
metadata.tiling_info = tiling_flags;
metadata.size_metadata = md->size_metadata;
memcpy(metadata.umd_metadata, md->metadata, sizeof(md->metadata));
amdgpu_bo_set_metadata(bo->bo, &metadata);
}
/* Returns the number of pixels that the texture should be aligned to
* in the given dimension. */
unsigned r300_get_pixel_alignment(enum pipe_format format,
unsigned num_samples,
enum radeon_bo_layout microtile,
enum radeon_bo_layout macrotile,
enum r300_dim dim, boolean is_rs690)
{
static const unsigned table[2][5][3][2] =
{
{
/* Macro: linear linear linear
Micro: linear tiled square-tiled */
{{ 32, 1}, { 8, 4}, { 0, 0}}, /* 8 bits per pixel */
{{ 16, 1}, { 8, 2}, { 4, 4}}, /* 16 bits per pixel */
{{ 8, 1}, { 4, 2}, { 0, 0}}, /* 32 bits per pixel */
{{ 4, 1}, { 2, 2}, { 0, 0}}, /* 64 bits per pixel */
{{ 2, 1}, { 0, 0}, { 0, 0}} /* 128 bits per pixel */
},
{
/* Macro: tiled tiled tiled
Micro: linear tiled square-tiled */
{{256, 8}, {64, 32}, { 0, 0}}, /* 8 bits per pixel */
{{128, 8}, {64, 16}, {32, 32}}, /* 16 bits per pixel */
{{ 64, 8}, {32, 16}, { 0, 0}}, /* 32 bits per pixel */
{{ 32, 8}, {16, 16}, { 0, 0}}, /* 64 bits per pixel */
{{ 16, 8}, { 0, 0}, { 0, 0}} /* 128 bits per pixel */
}
};
static const unsigned aa_block[2] = {4, 8};
unsigned tile = 0;
unsigned pixsize = util_format_get_blocksize(format);
assert(macrotile <= RADEON_LAYOUT_TILED);
assert(microtile <= RADEON_LAYOUT_SQUARETILED);
assert(pixsize <= 16);
assert(dim <= DIM_HEIGHT);
if (num_samples > 1) {
/* Multisampled textures have their own alignment scheme. */
if (pixsize == 4)
tile = aa_block[dim];
/* XXX FP16 AA. */
} else {
/* Standard alignment. */
tile = table[macrotile][util_logbase2(pixsize)][microtile][dim];
if (macrotile == 0 && is_rs690 && dim == DIM_WIDTH) {
int align;
int h_tile;
h_tile = table[macrotile][util_logbase2(pixsize)][microtile][DIM_HEIGHT];
align = 64 / (pixsize * h_tile);
if (tile < align)
tile = align;
}
}
assert(tile);
return tile;
}
static void
nv30_clear_render_target(struct pipe_context *pipe, struct pipe_surface *ps,
const union pipe_color_union *color,
unsigned x, unsigned y, unsigned w, unsigned h)
{
struct nv30_context *nv30 = nv30_context(pipe);
struct nv30_surface *sf = nv30_surface(ps);
struct nv30_miptree *mt = nv30_miptree(ps->texture);
struct nouveau_pushbuf *push = nv30->base.pushbuf;
struct nouveau_object *eng3d = nv30->screen->eng3d;
struct nouveau_pushbuf_refn refn;
uint32_t rt_format;
rt_format = nv30_format(pipe->screen, ps->format)->hw;
if (util_format_get_blocksize(ps->format) == 4)
rt_format |= NV30_3D_RT_FORMAT_ZETA_Z24S8;
else
rt_format |= NV30_3D_RT_FORMAT_ZETA_Z16;
if (nv30_miptree(ps->texture)->swizzled) {
rt_format |= NV30_3D_RT_FORMAT_TYPE_SWIZZLED;
rt_format |= util_logbase2(sf->width) << 16;
rt_format |= util_logbase2(sf->height) << 24;
} else {
rt_format |= NV30_3D_RT_FORMAT_TYPE_LINEAR;
}
refn.bo = mt->base.bo;
refn.flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_WR;
if (nouveau_pushbuf_space(push, 16, 1, 0) ||
nouveau_pushbuf_refn (push, &refn, 1))
return;
BEGIN_NV04(push, NV30_3D(RT_ENABLE), 1);
PUSH_DATA (push, NV30_3D_RT_ENABLE_COLOR0);
BEGIN_NV04(push, NV30_3D(RT_HORIZ), 3);
PUSH_DATA (push, sf->width << 16);
PUSH_DATA (push, sf->height << 16);
PUSH_DATA (push, rt_format);
BEGIN_NV04(push, NV30_3D(COLOR0_PITCH), 2);
if (eng3d->oclass < NV40_3D_CLASS)
PUSH_DATA (push, (sf->pitch << 16) | sf->pitch);
else
PUSH_DATA (push, sf->pitch);
PUSH_RELOC(push, mt->base.bo, sf->offset, NOUVEAU_BO_LOW, 0, 0);
BEGIN_NV04(push, NV30_3D(SCISSOR_HORIZ), 2);
PUSH_DATA (push, (w << 16) | x);
PUSH_DATA (push, (h << 16) | y);
BEGIN_NV04(push, NV30_3D(CLEAR_COLOR_VALUE), 2);
PUSH_DATA (push, pack_rgba(ps->format, color->f));
PUSH_DATA (push, NV30_3D_CLEAR_BUFFERS_COLOR_R |
NV30_3D_CLEAR_BUFFERS_COLOR_G |
NV30_3D_CLEAR_BUFFERS_COLOR_B |
NV30_3D_CLEAR_BUFFERS_COLOR_A);
nv30->dirty |= NV30_NEW_FRAMEBUFFER | NV30_NEW_SCISSOR;
}
void cayman_emit_msaa_config(struct radeon_winsys_cs *cs, int nr_samples,
int ps_iter_samples, int overrast_samples)
{
int setup_samples = nr_samples > 1 ? nr_samples :
overrast_samples > 1 ? overrast_samples : 0;
if (setup_samples > 1) {
/* indexed by log2(nr_samples) */
unsigned max_dist[] = {
0,
eg_max_dist_2x,
eg_max_dist_4x,
cm_max_dist_8x,
cm_max_dist_16x
};
unsigned log_samples = util_logbase2(setup_samples);
unsigned log_ps_iter_samples =
util_logbase2(util_next_power_of_two(ps_iter_samples));
radeon_set_context_reg_seq(cs, CM_R_028BDC_PA_SC_LINE_CNTL, 2);
radeon_emit(cs, S_028BDC_LAST_PIXEL(1) |
S_028BDC_EXPAND_LINE_WIDTH(1)); /* CM_R_028BDC_PA_SC_LINE_CNTL */
radeon_emit(cs, S_028BE0_MSAA_NUM_SAMPLES(log_samples) |
S_028BE0_MAX_SAMPLE_DIST(max_dist[log_samples]) |
S_028BE0_MSAA_EXPOSED_SAMPLES(log_samples)); /* CM_R_028BE0_PA_SC_AA_CONFIG */
if (nr_samples > 1) {
radeon_set_context_reg(cs, CM_R_028804_DB_EQAA,
S_028804_MAX_ANCHOR_SAMPLES(log_samples) |
S_028804_PS_ITER_SAMPLES(log_ps_iter_samples) |
S_028804_MASK_EXPORT_NUM_SAMPLES(log_samples) |
S_028804_ALPHA_TO_MASK_NUM_SAMPLES(log_samples) |
S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1,
EG_S_028A4C_PS_ITER_SAMPLE(ps_iter_samples > 1));
} else if (overrast_samples > 1) {
radeon_set_context_reg(cs, CM_R_028804_DB_EQAA,
S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
S_028804_STATIC_ANCHOR_ASSOCIATIONS(1) |
S_028804_OVERRASTERIZATION_AMOUNT(log_samples));
radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1, 0);
}
} else {
radeon_set_context_reg_seq(cs, CM_R_028BDC_PA_SC_LINE_CNTL, 2);
radeon_emit(cs, S_028BDC_LAST_PIXEL(1)); /* CM_R_028BDC_PA_SC_LINE_CNTL */
radeon_emit(cs, 0); /* CM_R_028BE0_PA_SC_AA_CONFIG */
radeon_set_context_reg(cs, CM_R_028804_DB_EQAA,
S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
radeon_set_context_reg(cs, EG_R_028A4C_PA_SC_MODE_CNTL_1, 0);
}
}
int nv50_tls_realloc(struct nv50_screen *screen, unsigned tls_space)
{
struct nouveau_pushbuf *push = screen->base.pushbuf;
int ret;
uint64_t tls_size;
if (tls_space < screen->cur_tls_space)
return 0;
if (tls_space > screen->max_tls_space) {
/* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
* LOCAL_WARPS_NO_CLAMP) */
NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
(unsigned)(tls_space / ONE_TEMP_SIZE),
(unsigned)(screen->max_tls_space / ONE_TEMP_SIZE));
return -ENOMEM;
}
nouveau_bo_ref(NULL, &screen->tls_bo);
ret = nv50_tls_alloc(screen, tls_space, &tls_size);
if (ret)
return ret;
BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->tls_bo->offset);
PUSH_DATA (push, screen->tls_bo->offset);
PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
return 1;
}
void
ImmediateValue::applyLog2()
{
switch (reg.type) {
case TYPE_S8:
case TYPE_S16:
case TYPE_S32:
assert(!this->isNegative());
// fall through
case TYPE_U8:
case TYPE_U16:
case TYPE_U32:
reg.data.u32 = util_logbase2(reg.data.u32);
break;
case TYPE_F32:
reg.data.f32 = log2f(reg.data.f32);
break;
case TYPE_F64:
reg.data.f64 = log2(reg.data.f64);
break;
default:
assert(0);
break;
}
}
static void amdgpu_bo_set_tiling(struct pb_buffer *_buf,
struct radeon_winsys_cs *rcs,
enum radeon_bo_layout microtiled,
enum radeon_bo_layout macrotiled,
unsigned pipe_config,
unsigned bankw, unsigned bankh,
unsigned tile_split,
unsigned stencil_tile_split,
unsigned mtilea, unsigned num_banks,
uint32_t pitch,
bool scanout)
{
struct amdgpu_winsys_bo *bo = get_amdgpu_winsys_bo(_buf);
struct amdgpu_bo_metadata metadata = {0};
uint32_t tiling_flags = 0;
if (macrotiled == RADEON_LAYOUT_TILED)
tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 4); /* 2D_TILED_THIN1 */
else if (microtiled == RADEON_LAYOUT_TILED)
tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 2); /* 1D_TILED_THIN1 */
else
tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 1); /* LINEAR_ALIGNED */
tiling_flags |= AMDGPU_TILING_SET(PIPE_CONFIG, pipe_config);
tiling_flags |= AMDGPU_TILING_SET(BANK_WIDTH, util_logbase2(bankw));
tiling_flags |= AMDGPU_TILING_SET(BANK_HEIGHT, util_logbase2(bankh));
if (tile_split)
tiling_flags |= AMDGPU_TILING_SET(TILE_SPLIT, eg_tile_split_rev(tile_split));
tiling_flags |= AMDGPU_TILING_SET(MACRO_TILE_ASPECT, util_logbase2(mtilea));
tiling_flags |= AMDGPU_TILING_SET(NUM_BANKS, util_logbase2(num_banks)-1);
if (scanout)
tiling_flags |= AMDGPU_TILING_SET(MICRO_TILE_MODE, 0); /* DISPLAY_MICRO_TILING */
else
tiling_flags |= AMDGPU_TILING_SET(MICRO_TILE_MODE, 1); /* THIN_MICRO_TILING */
metadata.tiling_info = tiling_flags;
amdgpu_bo_set_metadata(bo->bo, &metadata);
}
/**
* Called prior to emitting the TGSI code for each Mesa instruction.
* Allocate additional space for instructions if needed.
* Update the insn[] array so the next Mesa instruction points to
* the next TGSI instruction.
*/
static void set_insn_start( struct st_translate *t,
unsigned start )
{
if (t->insn_count + 1 >= t->insn_size) {
t->insn_size = 1 << (util_logbase2(t->insn_size) + 1);
t->insn = realloc(t->insn, t->insn_size * sizeof t->insn[0]);
if (t->insn == NULL) {
t->error = TRUE;
return;
}
}
t->insn[t->insn_count++] = start;
}
void r300_texture_setup_format_state(struct r300_screen *screen,
struct r300_texture_desc *desc,
unsigned level,
struct r300_texture_format_state *out)
{
struct pipe_resource *pt = &desc->b.b;
boolean is_r500 = screen->caps.is_r500;
/* Mask out all the fields we change. */
out->format0 = 0;
out->format1 &= ~R300_TX_FORMAT_TEX_COORD_TYPE_MASK;
out->format2 &= R500_TXFORMAT_MSB;
out->tile_config = 0;
/* Set sampler state. */
out->format0 = R300_TX_WIDTH((u_minify(pt->width0, level) - 1) & 0x7ff) |
R300_TX_HEIGHT((u_minify(pt->height0, level) - 1) & 0x7ff);
if (desc->uses_stride_addressing) {
/* rectangles love this */
out->format0 |= R300_TX_PITCH_EN;
out->format2 = (desc->stride_in_pixels[level] - 1) & 0x1fff;
} else {
/* Power of two textures (3D, mipmaps, and no pitch),
* also NPOT textures with a width being POT. */
out->format0 |=
R300_TX_DEPTH(util_logbase2(u_minify(pt->depth0, level)) & 0xf);
}
if (pt->target == PIPE_TEXTURE_CUBE) {
out->format1 |= R300_TX_FORMAT_CUBIC_MAP;
}
if (pt->target == PIPE_TEXTURE_3D) {
out->format1 |= R300_TX_FORMAT_3D;
}
/* large textures on r500 */
if (is_r500)
{
if (pt->width0 > 2048) {
out->format2 |= R500_TXWIDTH_BIT11;
}
if (pt->height0 > 2048) {
out->format2 |= R500_TXHEIGHT_BIT11;
}
}
out->tile_config = R300_TXO_MACRO_TILE(desc->macrotile[level]) |
R300_TXO_MICRO_TILE(desc->microtile);
}
/* Returns the number of pixels that the texture should be aligned to
* in the given dimension. */
unsigned r300_get_pixel_alignment(enum pipe_format format,
unsigned num_samples,
enum r300_buffer_tiling microtile,
enum r300_buffer_tiling macrotile,
enum r300_dim dim)
{
static const unsigned table[2][5][3][2] =
{
{
/* Macro: linear linear linear
Micro: linear tiled square-tiled */
{{ 32, 1}, { 8, 4}, { 0, 0}}, /* 8 bits per pixel */
{{ 16, 1}, { 8, 2}, { 4, 4}}, /* 16 bits per pixel */
{{ 8, 1}, { 4, 2}, { 0, 0}}, /* 32 bits per pixel */
{{ 4, 1}, { 0, 0}, { 2, 2}}, /* 64 bits per pixel */
{{ 2, 1}, { 0, 0}, { 0, 0}} /* 128 bits per pixel */
},
{
/* Macro: tiled tiled tiled
Micro: linear tiled square-tiled */
{{256, 8}, {64, 32}, { 0, 0}}, /* 8 bits per pixel */
{{128, 8}, {64, 16}, {32, 32}}, /* 16 bits per pixel */
{{ 64, 8}, {32, 16}, { 0, 0}}, /* 32 bits per pixel */
{{ 32, 8}, { 0, 0}, {16, 16}}, /* 64 bits per pixel */
{{ 16, 8}, { 0, 0}, { 0, 0}} /* 128 bits per pixel */
}
};
static const unsigned aa_block[2] = {4, 8};
unsigned tile = 0;
unsigned pixsize = util_format_get_blocksize(format);
assert(macrotile <= R300_BUFFER_TILED);
assert(microtile <= R300_BUFFER_SQUARETILED);
assert(pixsize <= 16);
assert(dim <= DIM_HEIGHT);
if (num_samples > 1) {
/* Multisampled textures have their own alignment scheme. */
if (pixsize == 4)
tile = aa_block[dim];
/* XXX FP16 AA. */
} else {
/* Standard alignment. */
tile = table[macrotile][util_logbase2(pixsize)][microtile][dim];
}
assert(tile);
return tile;
}
/**
* Return the width (dim==TILE_WIDTH) or height (dim==TILE_HEIGHT) of one tile
* of the given texture.
*/
static unsigned r300_texture_get_tile_size(struct r300_texture* tex,
int dim, boolean macrotile)
{
unsigned pixsize, tile_size;
pixsize = util_format_get_blocksize(tex->tex.format);
tile_size = microblock_table[util_logbase2(pixsize)][tex->microtile][dim];
if (macrotile) {
tile_size *= 8;
}
assert(tile_size);
return tile_size;
}
void
lp_debug_draw_bins_by_cmd_length( struct lp_scene *scene )
{
unsigned x, y;
for (y = 0; y < scene->tiles_y; y++) {
for (x = 0; x < scene->tiles_x; x++) {
const char *bits = " ...,-~:;=o+xaw*#XAWWWWWWWWWWWWWWWW";
unsigned sz = lp_scene_bin_size(scene, x, y);
unsigned sz2 = util_logbase2(sz);
debug_printf("%c", bits[MIN2(sz2,32)]);
}
debug_printf("\n");
}
}
static char *
swizzle2d_ptr(struct nv30_rect *rect, char *base, int x, int y, int z)
{
unsigned k = util_logbase2(MIN2(rect->w, rect->h));
unsigned km = (1 << k) - 1;
unsigned nx = rect->w >> k;
unsigned tx = x >> k;
unsigned ty = y >> k;
unsigned m;
m = swizzle2d(x & km, 0);
m |= swizzle2d(y & km, 1);
m += ((ty * nx) + tx) << k << k;
return base + (m * rect->cpp);
}
/**
* Compute the partial offset of a pixel block along an arbitrary axis.
*
* @param coord coordinate in pixels
* @param stride number of bytes between rows of successive pixel blocks
* @param block_length number of pixels in a pixels block along the coordinate
* axis
* @param out_offset resulting relative offset of the pixel block in bytes
* @param out_subcoord resulting sub-block pixel coordinate
*/
void
lp_build_sample_partial_offset(struct lp_build_context *bld,
unsigned block_length,
LLVMValueRef coord,
LLVMValueRef stride,
LLVMValueRef *out_offset,
LLVMValueRef *out_subcoord)
{
LLVMBuilderRef builder = bld->gallivm->builder;
LLVMValueRef offset;
LLVMValueRef subcoord;
if (block_length == 1) {
subcoord = bld->zero;
}
else {
/*
* Pixel blocks have power of two dimensions. LLVM should convert the
* rem/div to bit arithmetic.
* TODO: Verify this.
* It does indeed BUT it does transform it to scalar (and back) when doing so
* (using roughly extract, shift/and, mov, unpack) (llvm 2.7).
* The generated code looks seriously unfunny and is quite expensive.
*/
#if 0
LLVMValueRef block_width = lp_build_const_int_vec(bld->type, block_length);
subcoord = LLVMBuildURem(builder, coord, block_width, "");
coord = LLVMBuildUDiv(builder, coord, block_width, "");
#else
unsigned logbase2 = util_logbase2(block_length);
LLVMValueRef block_shift = lp_build_const_int_vec(bld->gallivm, bld->type, logbase2);
LLVMValueRef block_mask = lp_build_const_int_vec(bld->gallivm, bld->type, block_length - 1);
subcoord = LLVMBuildAnd(builder, coord, block_mask, "");
coord = LLVMBuildLShr(builder, coord, block_shift, "");
#endif
}
offset = lp_build_mul(bld, coord, stride);
assert(out_offset);
assert(out_subcoord);
*out_offset = offset;
*out_subcoord = subcoord;
}
/**
* Make note of a branch to a label in the TGSI code.
* After we've emitted all instructions, we'll go over the list
* of labels built here and patch the TGSI code with the actual
* location of each label.
*/
static unsigned *get_label( struct st_translate *t,
unsigned branch_target )
{
unsigned i;
if (t->labels_count + 1 >= t->labels_size) {
t->labels_size = 1 << (util_logbase2(t->labels_size) + 1);
t->labels = realloc(t->labels, t->labels_size * sizeof t->labels[0]);
if (t->labels == NULL) {
static unsigned dummy;
t->error = TRUE;
return &dummy;
}
}
i = t->labels_count++;
t->labels[i].branch_target = branch_target;
return &t->labels[i].token;
}
/**
* Try to allocate a pipe_resource object for the given st_texture_object.
*
* We use the given st_texture_image as a clue to determine the size of the
* mipmap image at level=0.
*
* \return GL_TRUE for success, GL_FALSE if out of memory.
*/
static GLboolean
guess_and_alloc_texture(struct st_context *st,
struct st_texture_object *stObj,
const struct st_texture_image *stImage)
{
GLuint lastLevel, width, height, depth;
GLuint bindings;
GLuint ptWidth, ptHeight, ptDepth, ptLayers;
enum pipe_format fmt;
DBG("%s\n", __FUNCTION__);
assert(!stObj->pt);
if (!guess_base_level_size(stObj->base.Target,
stImage->base.Width2,
stImage->base.Height2,
stImage->base.Depth2,
stImage->base.Level,
&width, &height, &depth)) {
/* we can't determine the image size at level=0 */
stObj->width0 = stObj->height0 = stObj->depth0 = 0;
/* this is not an out of memory error */
return GL_TRUE;
}
/* At this point, (width x height x depth) is the expected size of
* the level=0 mipmap image.
*/
/* Guess a reasonable value for lastLevel. With OpenGL we have no
* idea how many mipmap levels will be in a texture until we start
* to render with it. Make an educated guess here but be prepared
* to re-allocating a texture buffer with space for more (or fewer)
* mipmap levels later.
*/
if ((stObj->base.Sampler.MinFilter == GL_NEAREST ||
stObj->base.Sampler.MinFilter == GL_LINEAR ||
stImage->base._BaseFormat == GL_DEPTH_COMPONENT ||
stImage->base._BaseFormat == GL_DEPTH_STENCIL_EXT) &&
!stObj->base.GenerateMipmap &&
stImage->base.Level == 0) {
/* only alloc space for a single mipmap level */
lastLevel = 0;
}
else {
/* alloc space for a full mipmap */
GLuint l2width = util_logbase2(width);
GLuint l2height = util_logbase2(height);
GLuint l2depth = util_logbase2(depth);
lastLevel = MAX2(MAX2(l2width, l2height), l2depth);
}
/* Save the level=0 dimensions */
stObj->width0 = width;
stObj->height0 = height;
stObj->depth0 = depth;
fmt = st_mesa_format_to_pipe_format(stImage->base.TexFormat);
bindings = default_bindings(st, fmt);
st_gl_texture_dims_to_pipe_dims(stObj->base.Target,
width, height, depth,
&ptWidth, &ptHeight, &ptDepth, &ptLayers);
stObj->pt = st_texture_create(st,
gl_target_to_pipe(stObj->base.Target),
fmt,
lastLevel,
ptWidth,
ptHeight,
ptDepth,
ptLayers,
bindings);
stObj->lastLevel = lastLevel;
DBG("%s returning %d\n", __FUNCTION__, (stObj->pt != NULL));
return stObj->pt != NULL;
}
static void
nv30_clear_depth_stencil(struct pipe_context *pipe, struct pipe_surface *ps,
unsigned buffers, double depth, unsigned stencil,
unsigned x, unsigned y, unsigned w, unsigned h)
{
struct nv30_context *nv30 = nv30_context(pipe);
struct nv30_surface *sf = nv30_surface(ps);
struct nv30_miptree *mt = nv30_miptree(ps->texture);
struct nouveau_pushbuf *push = nv30->base.pushbuf;
struct nouveau_object *eng3d = nv30->screen->eng3d;
struct nouveau_pushbuf_refn refn;
uint32_t rt_format, mode = 0;
rt_format = nv30_format(pipe->screen, ps->format)->hw;
if (util_format_get_blocksize(ps->format) == 4)
rt_format |= NV30_3D_RT_FORMAT_COLOR_A8R8G8B8;
else
rt_format |= NV30_3D_RT_FORMAT_COLOR_R5G6B5;
if (nv30_miptree(ps->texture)->swizzled) {
rt_format |= NV30_3D_RT_FORMAT_TYPE_SWIZZLED;
rt_format |= util_logbase2(sf->width) << 16;
rt_format |= util_logbase2(sf->height) << 24;
} else {
rt_format |= NV30_3D_RT_FORMAT_TYPE_LINEAR;
}
if (buffers & PIPE_CLEAR_DEPTH)
mode |= NV30_3D_CLEAR_BUFFERS_DEPTH;
if (buffers & PIPE_CLEAR_STENCIL)
mode |= NV30_3D_CLEAR_BUFFERS_STENCIL;
refn.bo = mt->base.bo;
refn.flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_WR;
if (nouveau_pushbuf_space(push, 32, 1, 0) ||
nouveau_pushbuf_refn (push, &refn, 1))
return;
BEGIN_NV04(push, NV30_3D(RT_ENABLE), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV30_3D(RT_HORIZ), 3);
PUSH_DATA (push, sf->width << 16);
PUSH_DATA (push, sf->height << 16);
PUSH_DATA (push, rt_format);
if (eng3d->oclass < NV40_3D_CLASS) {
BEGIN_NV04(push, NV30_3D(COLOR0_PITCH), 1);
PUSH_DATA (push, (sf->pitch << 16) | sf->pitch);
} else {
BEGIN_NV04(push, NV40_3D(ZETA_PITCH), 1);
PUSH_DATA (push, sf->pitch);
}
BEGIN_NV04(push, NV30_3D(ZETA_OFFSET), 1);
PUSH_RELOC(push, mt->base.bo, sf->offset, NOUVEAU_BO_LOW, 0, 0);
BEGIN_NV04(push, NV30_3D(SCISSOR_HORIZ), 2);
PUSH_DATA (push, (w << 16) | x);
PUSH_DATA (push, (h << 16) | y);
BEGIN_NV04(push, NV30_3D(CLEAR_DEPTH_VALUE), 1);
PUSH_DATA (push, pack_zeta(ps->format, depth, stencil));
BEGIN_NV04(push, NV30_3D(CLEAR_BUFFERS), 1);
PUSH_DATA (push, mode);
nv30->dirty |= NV30_NEW_FRAMEBUFFER | NV30_NEW_SCISSOR;
}
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;
}
static struct pipe_sampler_view *
nv30_sampler_view_create(struct pipe_context *pipe, struct pipe_resource *pt,
const struct pipe_sampler_view *tmpl)
{
const struct nv30_texfmt *fmt = nv30_texfmt(pipe->screen, tmpl->format);
struct nouveau_object *eng3d = nv30_context(pipe)->screen->eng3d;
struct nv30_miptree *mt = nv30_miptree(pt);
struct nv30_sampler_view *so;
so = MALLOC_STRUCT(nv30_sampler_view);
if (!so)
return NULL;
so->pipe = *tmpl;
so->pipe.reference.count = 1;
so->pipe.texture = NULL;
so->pipe.context = pipe;
pipe_resource_reference(&so->pipe.texture, pt);
so->fmt = NV30_3D_TEX_FORMAT_NO_BORDER;
switch (pt->target) {
case PIPE_TEXTURE_1D:
so->fmt |= NV30_3D_TEX_FORMAT_DIMS_1D;
break;
case PIPE_TEXTURE_CUBE:
so->fmt |= NV30_3D_TEX_FORMAT_CUBIC;
case PIPE_TEXTURE_2D:
case PIPE_TEXTURE_RECT:
so->fmt |= NV30_3D_TEX_FORMAT_DIMS_2D;
break;
case PIPE_TEXTURE_3D:
so->fmt |= NV30_3D_TEX_FORMAT_DIMS_3D;
break;
default:
assert(0);
so->fmt |= NV30_3D_TEX_FORMAT_DIMS_1D;
break;
}
so->filt = fmt->filter;
so->wrap = fmt->wrap;
so->swz = fmt->swizzle;
so->swz |= swizzle(fmt, 3, tmpl->swizzle_a);
so->swz |= swizzle(fmt, 0, tmpl->swizzle_r) << 2;
so->swz |= swizzle(fmt, 1, tmpl->swizzle_g) << 4;
so->swz |= swizzle(fmt, 2, tmpl->swizzle_b) << 6;
/* apparently, we need to ignore the t coordinate for 1D textures to
* fix piglit tex1d-2dborder
*/
so->wrap_mask = ~0;
if (pt->target == PIPE_TEXTURE_1D) {
so->wrap_mask &= ~NV30_3D_TEX_WRAP_T__MASK;
so->wrap |= NV30_3D_TEX_WRAP_T_REPEAT;
}
/* yet more hardware suckage, can't filter 32-bit float formats */
switch (tmpl->format) {
case PIPE_FORMAT_R32_FLOAT:
case PIPE_FORMAT_R32G32B32A32_FLOAT:
so->filt_mask = ~(NV30_3D_TEX_FILTER_MIN__MASK |
NV30_3D_TEX_FILTER_MAG__MASK);
so->filt |= NV30_3D_TEX_FILTER_MIN_NEAREST |
NV30_3D_TEX_FILTER_MAG_NEAREST;
break;
default:
so->filt_mask = ~0;
break;
}
so->npot_size0 = (pt->width0 << 16) | pt->height0;
if (eng3d->oclass >= NV40_3D_CLASS) {
so->npot_size1 = (pt->depth0 << 20) | mt->uniform_pitch;
if (!mt->swizzled)
so->fmt |= NV40_3D_TEX_FORMAT_LINEAR;
so->fmt |= 0x00008000;
so->fmt |= (pt->last_level + 1) << NV40_3D_TEX_FORMAT_MIPMAP_COUNT__SHIFT;
} else {
so->swz |= mt->uniform_pitch << NV30_3D_TEX_SWIZZLE_RECT_PITCH__SHIFT;
if (pt->last_level)
so->fmt |= NV30_3D_TEX_FORMAT_MIPMAP;
so->fmt |= util_logbase2(pt->width0) << 20;
so->fmt |= util_logbase2(pt->height0) << 24;
so->fmt |= util_logbase2(pt->depth0) << 28;
so->fmt |= 0x00010000;
}
so->base_lod = so->pipe.u.tex.first_level << 8;
so->high_lod = MIN2(pt->last_level, so->pipe.u.tex.last_level) << 8;
return &so->pipe;
}
static int
svga_get_param(struct pipe_screen *screen, enum pipe_cap param)
{
struct svga_screen *svgascreen = svga_screen(screen);
struct svga_winsys_screen *sws = svgascreen->sws;
SVGA3dDevCapResult result;
switch (param) {
case PIPE_CAP_NPOT_TEXTURES:
case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
return 1;
case PIPE_CAP_TWO_SIDED_STENCIL:
return 1;
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
/*
* "In virtually every OpenGL implementation and hardware,
* GL_MAX_DUAL_SOURCE_DRAW_BUFFERS is 1"
* http://www.opengl.org/wiki/Blending
*/
return sws->have_vgpu10 ? 1 : 0;
case PIPE_CAP_ANISOTROPIC_FILTER:
return 1;
case PIPE_CAP_POINT_SPRITE:
return 1;
case PIPE_CAP_TGSI_TEXCOORD:
return 0;
case PIPE_CAP_MAX_RENDER_TARGETS:
return svgascreen->max_color_buffers;
case PIPE_CAP_OCCLUSION_QUERY:
return 1;
case PIPE_CAP_QUERY_TIME_ELAPSED:
return 0;
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
return sws->have_vgpu10;
case PIPE_CAP_TEXTURE_SHADOW_MAP:
return 1;
case PIPE_CAP_TEXTURE_SWIZZLE:
return 1;
case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
return 0;
case PIPE_CAP_USER_VERTEX_BUFFERS:
case PIPE_CAP_USER_INDEX_BUFFERS:
return 0;
case PIPE_CAP_USER_CONSTANT_BUFFERS:
return 1;
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
return 256;
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
{
unsigned levels = SVGA_MAX_TEXTURE_LEVELS;
if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_WIDTH, &result))
levels = MIN2(util_logbase2(result.u) + 1, levels);
else
levels = 12 /* 2048x2048 */;
if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_HEIGHT, &result))
levels = MIN2(util_logbase2(result.u) + 1, levels);
else
levels = 12 /* 2048x2048 */;
return levels;
}
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VOLUME_EXTENT, &result))
return 8; /* max 128x128x128 */
return MIN2(util_logbase2(result.u) + 1, SVGA_MAX_TEXTURE_LEVELS);
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
/*
* No mechanism to query the host, and at least limited to 2048x2048 on
* certain hardware.
*/
return MIN2(screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS),
12 /* 2048x2048 */);
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
return sws->have_vgpu10 ? SVGA3D_MAX_SURFACE_ARRAYSIZE : 0;
case PIPE_CAP_BLEND_EQUATION_SEPARATE: /* req. for GL 1.5 */
return 1;
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
return 1;
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
return sws->have_vgpu10;
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
return 0;
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
return !sws->have_vgpu10;
case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
return 1; /* The color outputs of vertex shaders are not clamped */
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
return 0; /* The driver can't clamp vertex colors */
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
return 0; /* The driver can't clamp fragment colors */
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
return 1; /* expected for GL_ARB_framebuffer_object */
case PIPE_CAP_GLSL_FEATURE_LEVEL:
return sws->have_vgpu10 ? 330 : 120;
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
return 0;
case PIPE_CAP_SM3:
return 1;
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_INDEP_BLEND_ENABLE:
case PIPE_CAP_CONDITIONAL_RENDER:
case PIPE_CAP_QUERY_TIMESTAMP:
case PIPE_CAP_TGSI_INSTANCEID:
case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
case PIPE_CAP_SEAMLESS_CUBE_MAP:
case PIPE_CAP_FAKE_SW_MSAA:
return sws->have_vgpu10;
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
return sws->have_vgpu10 ? SVGA3D_DX_MAX_SOTARGETS : 0;
case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
return sws->have_vgpu10 ? 4 : 0;
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
return sws->have_vgpu10 ? SVGA3D_MAX_STREAMOUT_DECLS : 0;
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
return 0;
case PIPE_CAP_TEXTURE_MULTISAMPLE:
return svgascreen->ms_samples ? 1 : 0;
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
return SVGA3D_DX_MAX_RESOURCE_SIZE;
case PIPE_CAP_MIN_TEXEL_OFFSET:
return sws->have_vgpu10 ? VGPU10_MIN_TEXEL_FETCH_OFFSET : 0;
case PIPE_CAP_MAX_TEXEL_OFFSET:
return sws->have_vgpu10 ? VGPU10_MAX_TEXEL_FETCH_OFFSET : 0;
case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
return 0;
case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
return sws->have_vgpu10 ? 256 : 0;
case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
return sws->have_vgpu10 ? 1024 : 0;
case PIPE_CAP_PRIMITIVE_RESTART:
return 1; /* may be a sw fallback, depending on restart index */
case PIPE_CAP_GENERATE_MIPMAP:
return sws->have_vgpu10;
/* Unsupported features */
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
case PIPE_CAP_SHADER_STENCIL_EXPORT:
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_INDEP_BLEND_FUNC:
case PIPE_CAP_TEXTURE_BARRIER:
case PIPE_CAP_MAX_VERTEX_STREAMS:
case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
case PIPE_CAP_COMPUTE:
case PIPE_CAP_START_INSTANCE:
case PIPE_CAP_CUBE_MAP_ARRAY:
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
case PIPE_CAP_TEXTURE_GATHER_SM5:
case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
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_DRAW_INDIRECT:
case PIPE_CAP_MULTI_DRAW_INDIRECT:
case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
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_TGSI_PACK_HALF_FLOAT:
case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_INVALIDATE_BUFFER:
case PIPE_CAP_STRING_MARKER:
case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
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:
return 0;
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
return 64;
case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
return 1; /* need 4-byte alignment for all offsets and strides */
case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
return 2048;
case PIPE_CAP_MAX_VIEWPORTS:
return 1;
case PIPE_CAP_ENDIANNESS:
return PIPE_ENDIAN_LITTLE;
case PIPE_CAP_VENDOR_ID:
return 0x15ad; /* VMware Inc. */
case PIPE_CAP_DEVICE_ID:
return 0x0405; /* assume SVGA II */
case PIPE_CAP_ACCELERATED:
return 0; /* XXX: */
case PIPE_CAP_VIDEO_MEMORY:
/* XXX: Query the host ? */
return 1;
case PIPE_CAP_UMA:
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:
case PIPE_CAP_DEPTH_BOUNDS_TEST:
case PIPE_CAP_TGSI_TXQS:
case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
case PIPE_CAP_SHAREABLE_SHADERS:
case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
case PIPE_CAP_CLEAR_TEXTURE:
case PIPE_CAP_DRAW_PARAMETERS:
case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
case PIPE_CAP_QUERY_BUFFER_OBJECT:
return 0;
}
debug_printf("Unexpected PIPE_CAP_ query %u\n", param);
return 0;
}
/*
* Do a cached lookup.
*
* Returns (vectors of) 4x8 rgba aos value
*/
LLVMValueRef
lp_build_fetch_cached_texels(struct gallivm_state *gallivm,
const struct util_format_description *format_desc,
unsigned n,
LLVMValueRef base_ptr,
LLVMValueRef offset,
LLVMValueRef i,
LLVMValueRef j,
LLVMValueRef cache)
{
LLVMBuilderRef builder = gallivm->builder;
unsigned count, low_bit, log2size;
LLVMValueRef color, offset_stored, addr, ptr_addrtrunc, tmp;
LLVMValueRef ij_index, hash_index, hash_mask, block_index;
LLVMTypeRef i8t = LLVMInt8TypeInContext(gallivm->context);
LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
LLVMTypeRef i64t = LLVMInt64TypeInContext(gallivm->context);
struct lp_type type;
struct lp_build_context bld32;
memset(&type, 0, sizeof type);
type.width = 32;
type.length = n;
assert(format_desc->block.width == 4);
assert(format_desc->block.height == 4);
lp_build_context_init(&bld32, gallivm, type);
/*
* compute hash - we use direct mapped cache, the hash function could
* be better but it needs to be simple
* per-element:
* compare offset with offset stored at tag (hash)
* if not equal decode/store block, update tag
* extract color from cache
* assemble result vector
*/
/* TODO: not ideal with 32bit pointers... */
low_bit = util_logbase2(format_desc->block.bits / 8);
log2size = util_logbase2(LP_BUILD_FORMAT_CACHE_SIZE);
addr = LLVMBuildPtrToInt(builder, base_ptr, i64t, "");
ptr_addrtrunc = LLVMBuildPtrToInt(builder, base_ptr, i32t, "");
ptr_addrtrunc = lp_build_broadcast_scalar(&bld32, ptr_addrtrunc);
/* For the hash function, first mask off the unused lowest bits. Then just
do some xor with address bits - only use lower 32bits */
ptr_addrtrunc = LLVMBuildAdd(builder, offset, ptr_addrtrunc, "");
ptr_addrtrunc = LLVMBuildLShr(builder, ptr_addrtrunc,
lp_build_const_int_vec(gallivm, type, low_bit), "");
/* This only really makes sense for size 64,128,256 */
hash_index = ptr_addrtrunc;
ptr_addrtrunc = LLVMBuildLShr(builder, ptr_addrtrunc,
lp_build_const_int_vec(gallivm, type, 2*log2size), "");
hash_index = LLVMBuildXor(builder, ptr_addrtrunc, hash_index, "");
tmp = LLVMBuildLShr(builder, hash_index,
lp_build_const_int_vec(gallivm, type, log2size), "");
hash_index = LLVMBuildXor(builder, hash_index, tmp, "");
hash_mask = lp_build_const_int_vec(gallivm, type, LP_BUILD_FORMAT_CACHE_SIZE - 1);
hash_index = LLVMBuildAnd(builder, hash_index, hash_mask, "");
ij_index = LLVMBuildShl(builder, i, lp_build_const_int_vec(gallivm, type, 2), "");
ij_index = LLVMBuildAdd(builder, ij_index, j, "");
block_index = LLVMBuildShl(builder, hash_index,
lp_build_const_int_vec(gallivm, type, 4), "");
block_index = LLVMBuildAdd(builder, ij_index, block_index, "");
if (n > 1) {
color = LLVMGetUndef(LLVMVectorType(i32t, n));
for (count = 0; count < n; count++) {
LLVMValueRef index, cond, colorx;
LLVMValueRef block_indexx, hash_indexx, addrx, offsetx, ptr_addrx;
struct lp_build_if_state if_ctx;
index = lp_build_const_int32(gallivm, count);
offsetx = LLVMBuildExtractElement(builder, offset, index, "");
addrx = LLVMBuildZExt(builder, offsetx, i64t, "");
addrx = LLVMBuildAdd(builder, addrx, addr, "");
block_indexx = LLVMBuildExtractElement(builder, block_index, index, "");
hash_indexx = LLVMBuildLShr(builder, block_indexx,
lp_build_const_int32(gallivm, 4), "");
offset_stored = lookup_tag_data(gallivm, cache, hash_indexx);
cond = LLVMBuildICmp(builder, LLVMIntNE, offset_stored, addrx, "");
lp_build_if(&if_ctx, gallivm, cond);
{
ptr_addrx = LLVMBuildIntToPtr(builder, addrx,
LLVMPointerType(i8t, 0), "");
update_cached_block(gallivm, format_desc, ptr_addrx, hash_indexx, cache);
#if LP_BUILD_FORMAT_CACHE_DEBUG
update_cache_access(gallivm, cache, 1,
LP_BUILD_FORMAT_CACHE_MEMBER_ACCESS_MISS);
#endif
}
lp_build_endif(&if_ctx);
colorx = lookup_cached_pixel(gallivm, cache, block_indexx);
color = LLVMBuildInsertElement(builder, color, colorx,
lp_build_const_int32(gallivm, count), "");
}
}
else {
LLVMValueRef cond;
struct lp_build_if_state if_ctx;
tmp = LLVMBuildZExt(builder, offset, i64t, "");
addr = LLVMBuildAdd(builder, tmp, addr, "");
offset_stored = lookup_tag_data(gallivm, cache, hash_index);
cond = LLVMBuildICmp(builder, LLVMIntNE, offset_stored, addr, "");
lp_build_if(&if_ctx, gallivm, cond);
{
tmp = LLVMBuildIntToPtr(builder, addr, LLVMPointerType(i8t, 0), "");
update_cached_block(gallivm, format_desc, tmp, hash_index, cache);
#if LP_BUILD_FORMAT_CACHE_DEBUG
update_cache_access(gallivm, cache, 1,
LP_BUILD_FORMAT_CACHE_MEMBER_ACCESS_MISS);
#endif
}
lp_build_endif(&if_ctx);
color = lookup_cached_pixel(gallivm, cache, block_index);
}
#if LP_BUILD_FORMAT_CACHE_DEBUG
update_cache_access(gallivm, cache, n,
LP_BUILD_FORMAT_CACHE_MEMBER_ACCESS_TOTAL);
#endif
return LLVMBuildBitCast(builder, color, LLVMVectorType(i8t, n * 4), "");
}
static int
svga_get_param(struct pipe_screen *screen, enum pipe_cap param)
{
struct svga_screen *svgascreen = svga_screen(screen);
struct svga_winsys_screen *sws = svgascreen->sws;
SVGA3dDevCapResult result;
switch (param) {
case PIPE_CAP_NPOT_TEXTURES:
case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
return 1;
case PIPE_CAP_TWO_SIDED_STENCIL:
return 1;
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
return 0;
case PIPE_CAP_ANISOTROPIC_FILTER:
return 1;
case PIPE_CAP_POINT_SPRITE:
return 1;
case PIPE_CAP_TGSI_TEXCOORD:
return 0;
case PIPE_CAP_MAX_RENDER_TARGETS:
return svgascreen->max_color_buffers;
case PIPE_CAP_OCCLUSION_QUERY:
return 1;
case PIPE_CAP_QUERY_TIME_ELAPSED:
return 0;
case PIPE_CAP_TEXTURE_SHADOW_MAP:
return 1;
case PIPE_CAP_TEXTURE_SWIZZLE:
return 1;
case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
return 0;
case PIPE_CAP_USER_VERTEX_BUFFERS:
case PIPE_CAP_USER_INDEX_BUFFERS:
return 0;
case PIPE_CAP_USER_CONSTANT_BUFFERS:
return 1;
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
return 16;
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
{
unsigned levels = SVGA_MAX_TEXTURE_LEVELS;
if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_WIDTH, &result))
levels = MIN2(util_logbase2(result.u) + 1, levels);
else
levels = 12 /* 2048x2048 */;
if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_HEIGHT, &result))
levels = MIN2(util_logbase2(result.u) + 1, levels);
else
levels = 12 /* 2048x2048 */;
return levels;
}
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VOLUME_EXTENT, &result))
return 8; /* max 128x128x128 */
return MIN2(util_logbase2(result.u) + 1, SVGA_MAX_TEXTURE_LEVELS);
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
/*
* No mechanism to query the host, and at least limited to 2048x2048 on
* certain hardware.
*/
return MIN2(screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS),
12 /* 2048x2048 */);
case PIPE_CAP_BLEND_EQUATION_SEPARATE: /* req. for GL 1.5 */
return 1;
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
return 1;
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
return 0;
case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
return 0;
case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
return 1;
case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
return 1; /* The color outputs of vertex shaders are not clamped */
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
return 0; /* The driver can't clamp vertex colors */
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
return 0; /* The driver can't clamp fragment colors */
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
return 1; /* expected for GL_ARB_framebuffer_object */
case PIPE_CAP_GLSL_FEATURE_LEVEL:
return 120;
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
return 0;
case PIPE_CAP_SM3:
return 1;
/* Unsupported features */
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
case PIPE_CAP_SHADER_STENCIL_EXPORT:
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_SEAMLESS_CUBE_MAP:
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_INDEP_BLEND_ENABLE:
case PIPE_CAP_INDEP_BLEND_FUNC:
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
case PIPE_CAP_PRIMITIVE_RESTART:
case PIPE_CAP_TGSI_INSTANCEID:
case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
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_CONDITIONAL_RENDER:
case PIPE_CAP_TEXTURE_BARRIER:
case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
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_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_COMPUTE:
case PIPE_CAP_START_INSTANCE:
case PIPE_CAP_QUERY_TIMESTAMP:
case PIPE_CAP_TEXTURE_MULTISAMPLE:
case PIPE_CAP_CUBE_MAP_ARRAY:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
case PIPE_CAP_TEXTURE_GATHER_SM5:
case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
case PIPE_CAP_RASTERIZER_LOWER_LEFT_ORIGIN:
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_DRAW_INDIRECT:
case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
return 0;
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
return 64;
case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
return 1;
case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
return 2048;
case PIPE_CAP_MAX_VIEWPORTS:
return 1;
case PIPE_CAP_ENDIANNESS:
return PIPE_ENDIAN_LITTLE;
case PIPE_CAP_VENDOR_ID:
return 0x15ad; /* VMware Inc. */
case PIPE_CAP_DEVICE_ID:
return 0x0405; /* assume SVGA II */
case PIPE_CAP_ACCELERATED:
return 0; /* XXX: */
case PIPE_CAP_VIDEO_MEMORY:
/* XXX: Query the host ? */
return 1;
case PIPE_CAP_UMA:
return 0;
}
debug_printf("Unexpected PIPE_CAP_ query %u\n", param);
return 0;
}
static int
svga_get_param(struct pipe_screen *screen, enum pipe_cap param)
{
struct svga_screen *svgascreen = svga_screen(screen);
struct svga_winsys_screen *sws = svgascreen->sws;
SVGA3dDevCapResult result;
switch (param) {
case PIPE_CAP_MAX_COMBINED_SAMPLERS:
return 16;
case PIPE_CAP_NPOT_TEXTURES:
return 1;
case PIPE_CAP_TWO_SIDED_STENCIL:
return 1;
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
return 0;
case PIPE_CAP_ANISOTROPIC_FILTER:
return 1;
case PIPE_CAP_POINT_SPRITE:
return 1;
case PIPE_CAP_MAX_RENDER_TARGETS:
if(!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_RENDER_TARGETS, &result))
return 1;
if(!result.u)
return 1;
return MIN2(result.u, PIPE_MAX_COLOR_BUFS);
case PIPE_CAP_OCCLUSION_QUERY:
return 1;
case PIPE_CAP_QUERY_TIME_ELAPSED:
return 0;
case PIPE_CAP_TEXTURE_SHADOW_MAP:
return 1;
case PIPE_CAP_TEXTURE_SWIZZLE:
return 1;
case PIPE_CAP_USER_VERTEX_BUFFERS:
case PIPE_CAP_USER_INDEX_BUFFERS:
return 0;
case PIPE_CAP_USER_CONSTANT_BUFFERS:
return 1;
case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
return 16;
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
{
unsigned levels = SVGA_MAX_TEXTURE_LEVELS;
if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_WIDTH, &result))
levels = MIN2(util_logbase2(result.u) + 1, levels);
else
levels = 12 /* 2048x2048 */;
if (sws->get_cap(sws, SVGA3D_DEVCAP_MAX_TEXTURE_HEIGHT, &result))
levels = MIN2(util_logbase2(result.u) + 1, levels);
else
levels = 12 /* 2048x2048 */;
return levels;
}
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
if (!sws->get_cap(sws, SVGA3D_DEVCAP_MAX_VOLUME_EXTENT, &result))
return 8; /* max 128x128x128 */
return MIN2(util_logbase2(result.u) + 1, SVGA_MAX_TEXTURE_LEVELS);
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
/*
* No mechanism to query the host, and at least limited to 2048x2048 on
* certain hardware.
*/
return MIN2(screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS),
12 /* 2048x2048 */);
case PIPE_CAP_BLEND_EQUATION_SEPARATE: /* req. for GL 1.5 */
return 1;
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_VERTEX_COLOR_UNCLAMPED:
return 1; /* The color outputs of vertex shaders are not clamped */
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
return 0; /* The driver can't clamp vertex colors */
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
return 0; /* The driver can't clamp fragment colors */
case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
return 1; /* expected for GL_ARB_framebuffer_object */
case PIPE_CAP_GLSL_FEATURE_LEVEL:
return 120;
/* Unsupported features */
case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
case PIPE_CAP_SM3:
case PIPE_CAP_SHADER_STENCIL_EXPORT:
case PIPE_CAP_DEPTH_CLIP_DISABLE:
case PIPE_CAP_SEAMLESS_CUBE_MAP:
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_INDEP_BLEND_ENABLE:
case PIPE_CAP_INDEP_BLEND_FUNC:
case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
case PIPE_CAP_PRIMITIVE_RESTART:
case PIPE_CAP_TGSI_INSTANCEID:
case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
case PIPE_CAP_SCALED_RESOLVE:
case PIPE_CAP_MIN_TEXEL_OFFSET:
case PIPE_CAP_MAX_TEXEL_OFFSET:
case PIPE_CAP_CONDITIONAL_RENDER:
case PIPE_CAP_TEXTURE_BARRIER:
case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
case PIPE_CAP_TGSI_CAN_COMPACT_VARYINGS:
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_COMPUTE:
case PIPE_CAP_START_INSTANCE:
case PIPE_CAP_QUERY_TIMESTAMP:
case PIPE_CAP_TEXTURE_MULTISAMPLE:
case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
case PIPE_CAP_CUBE_MAP_ARRAY:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
return 0;
case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
return 1;
}
debug_printf("Unexpected PIPE_CAP_ query %u\n", param);
return 0;
}
static void
nv30_transfer_rect_sifm(XFER_ARGS)
{
struct nouveau_pushbuf *push = nv30->base.pushbuf;
struct nouveau_pushbuf_refn refs[] = {
{ src->bo, src->domain | NOUVEAU_BO_RD },
{ dst->bo, dst->domain | NOUVEAU_BO_WR },
};
struct nv04_fifo *fifo = push->channel->data;
unsigned si_fmt, si_arg;
unsigned ss_fmt;
switch (dst->cpp) {
case 4: ss_fmt = NV04_SURFACE_SWZ_FORMAT_COLOR_A8R8G8B8; break;
case 2: ss_fmt = NV04_SURFACE_SWZ_FORMAT_COLOR_R5G6B5; break;
default:
ss_fmt = NV04_SURFACE_SWZ_FORMAT_COLOR_Y8;
break;
}
switch (src->cpp) {
case 4: si_fmt = NV03_SIFM_COLOR_FORMAT_A8R8G8B8; break;
case 2: si_fmt = NV03_SIFM_COLOR_FORMAT_R5G6B5; break;
default:
si_fmt = NV03_SIFM_COLOR_FORMAT_AY8;
break;
}
if (filter == NEAREST) {
si_arg = NV03_SIFM_FORMAT_ORIGIN_CENTER;
si_arg |= NV03_SIFM_FORMAT_FILTER_POINT_SAMPLE;
} else {
si_arg = NV03_SIFM_FORMAT_ORIGIN_CORNER;
si_arg |= NV03_SIFM_FORMAT_FILTER_BILINEAR;
}
if (nouveau_pushbuf_space(push, 32, 6, 0) ||
nouveau_pushbuf_refn (push, refs, 2))
return;
if (dst->pitch) {
BEGIN_NV04(push, NV04_SF2D(DMA_IMAGE_SOURCE), 2);
PUSH_RELOC(push, dst->bo, 0, NOUVEAU_BO_OR, fifo->vram, fifo->gart);
PUSH_RELOC(push, dst->bo, 0, NOUVEAU_BO_OR, fifo->vram, fifo->gart);
BEGIN_NV04(push, NV04_SF2D(FORMAT), 4);
PUSH_DATA (push, ss_fmt);
PUSH_DATA (push, dst->pitch << 16 | dst->pitch);
PUSH_RELOC(push, dst->bo, dst->offset, NOUVEAU_BO_LOW, 0, 0);
PUSH_RELOC(push, dst->bo, dst->offset, NOUVEAU_BO_LOW, 0, 0);
BEGIN_NV04(push, NV05_SIFM(SURFACE), 1);
PUSH_DATA (push, nv30->screen->surf2d->handle);
} else {
BEGIN_NV04(push, NV04_SSWZ(DMA_IMAGE), 1);
PUSH_RELOC(push, dst->bo, 0, NOUVEAU_BO_OR, fifo->vram, fifo->gart);
BEGIN_NV04(push, NV04_SSWZ(FORMAT), 2);
PUSH_DATA (push, ss_fmt | (util_logbase2(dst->w) << 16) |
(util_logbase2(dst->h) << 24));
PUSH_RELOC(push, dst->bo, dst->offset, NOUVEAU_BO_LOW, 0, 0);
BEGIN_NV04(push, NV05_SIFM(SURFACE), 1);
PUSH_DATA (push, nv30->screen->swzsurf->handle);
}
BEGIN_NV04(push, NV03_SIFM(DMA_IMAGE), 1);
PUSH_RELOC(push, src->bo, 0, NOUVEAU_BO_OR, fifo->vram, fifo->gart);
BEGIN_NV04(push, NV03_SIFM(COLOR_FORMAT), 8);
PUSH_DATA (push, si_fmt);
PUSH_DATA (push, NV03_SIFM_OPERATION_SRCCOPY);
PUSH_DATA (push, ( dst->y0 << 16) | dst->x0);
PUSH_DATA (push, ((dst->y1 - dst->y0) << 16) | (dst->x1 - dst->x0));
PUSH_DATA (push, ( dst->y0 << 16) | dst->x0);
PUSH_DATA (push, ((dst->y1 - dst->y0) << 16) | (dst->x1 - dst->x0));
PUSH_DATA (push, ((src->x1 - src->x0) << 20) / (dst->x1 - dst->x0));
PUSH_DATA (push, ((src->y1 - src->y0) << 20) / (dst->y1 - dst->y0));
BEGIN_NV04(push, NV03_SIFM(SIZE), 4);
PUSH_DATA (push, align(src->h, 2) << 16 | align(src->w, 2));
PUSH_DATA (push, src->pitch | si_arg);
PUSH_RELOC(push, src->bo, src->offset, NOUVEAU_BO_LOW, 0, 0);
PUSH_DATA (push, (src->y0 << 20) | src->x0 << 4);
}
bool r600_common_screen_init(struct r600_common_screen *rscreen,
struct radeon_winsys *ws)
{
char llvm_string[32] = {}, kernel_version[128] = {};
struct utsname uname_data;
ws->query_info(ws, &rscreen->info);
if (uname(&uname_data) == 0)
snprintf(kernel_version, sizeof(kernel_version),
" / %s", uname_data.release);
#if HAVE_LLVM
snprintf(llvm_string, sizeof(llvm_string),
", LLVM %i.%i.%i", (HAVE_LLVM >> 8) & 0xff,
HAVE_LLVM & 0xff, MESA_LLVM_VERSION_PATCH);
#endif
snprintf(rscreen->renderer_string, sizeof(rscreen->renderer_string),
"%s (DRM %i.%i.%i%s%s)",
r600_get_chip_name(rscreen), rscreen->info.drm_major,
rscreen->info.drm_minor, rscreen->info.drm_patchlevel,
kernel_version, llvm_string);
rscreen->b.get_name = r600_get_name;
rscreen->b.get_vendor = r600_get_vendor;
rscreen->b.get_device_vendor = r600_get_device_vendor;
rscreen->b.get_compute_param = r600_get_compute_param;
rscreen->b.get_paramf = r600_get_paramf;
rscreen->b.get_timestamp = r600_get_timestamp;
rscreen->b.fence_finish = r600_fence_finish;
rscreen->b.fence_reference = r600_fence_reference;
rscreen->b.resource_destroy = u_resource_destroy_vtbl;
rscreen->b.resource_from_user_memory = r600_buffer_from_user_memory;
rscreen->b.query_memory_info = r600_query_memory_info;
if (rscreen->info.has_uvd) {
rscreen->b.get_video_param = rvid_get_video_param;
rscreen->b.is_video_format_supported = rvid_is_format_supported;
} else {
rscreen->b.get_video_param = r600_get_video_param;
rscreen->b.is_video_format_supported = vl_video_buffer_is_format_supported;
}
r600_init_screen_texture_functions(rscreen);
r600_init_screen_query_functions(rscreen);
rscreen->ws = ws;
rscreen->family = rscreen->info.family;
rscreen->chip_class = rscreen->info.chip_class;
rscreen->debug_flags = debug_get_flags_option("R600_DEBUG", common_debug_options, 0);
rscreen->force_aniso = MIN2(16, debug_get_num_option("R600_TEX_ANISO", -1));
if (rscreen->force_aniso >= 0) {
printf("radeon: Forcing anisotropy filter to %ix\n",
/* round down to a power of two */
1 << util_logbase2(rscreen->force_aniso));
}
util_format_s3tc_init();
pipe_mutex_init(rscreen->aux_context_lock);
pipe_mutex_init(rscreen->gpu_load_mutex);
if (rscreen->debug_flags & DBG_INFO) {
printf("pci_id = 0x%x\n", rscreen->info.pci_id);
printf("family = %i (%s)\n", rscreen->info.family,
r600_get_chip_name(rscreen));
printf("chip_class = %i\n", rscreen->info.chip_class);
printf("gart_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.gart_size, 1024*1024));
printf("vram_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.vram_size, 1024*1024));
printf("max_alloc_size = %i MB\n",
(int)DIV_ROUND_UP(rscreen->info.max_alloc_size, 1024*1024));
printf("has_virtual_memory = %i\n", rscreen->info.has_virtual_memory);
printf("gfx_ib_pad_with_type2 = %i\n", rscreen->info.gfx_ib_pad_with_type2);
printf("has_sdma = %i\n", rscreen->info.has_sdma);
printf("has_uvd = %i\n", rscreen->info.has_uvd);
printf("vce_fw_version = %i\n", rscreen->info.vce_fw_version);
printf("vce_harvest_config = %i\n", rscreen->info.vce_harvest_config);
printf("clock_crystal_freq = %i\n", rscreen->info.clock_crystal_freq);
printf("drm = %i.%i.%i\n", rscreen->info.drm_major,
rscreen->info.drm_minor, rscreen->info.drm_patchlevel);
printf("has_userptr = %i\n", rscreen->info.has_userptr);
printf("r600_max_quad_pipes = %i\n", rscreen->info.r600_max_quad_pipes);
printf("max_shader_clock = %i\n", rscreen->info.max_shader_clock);
printf("num_good_compute_units = %i\n", rscreen->info.num_good_compute_units);
printf("max_se = %i\n", rscreen->info.max_se);
printf("max_sh_per_se = %i\n", rscreen->info.max_sh_per_se);
printf("r600_gb_backend_map = %i\n", rscreen->info.r600_gb_backend_map);
printf("r600_gb_backend_map_valid = %i\n", rscreen->info.r600_gb_backend_map_valid);
printf("r600_num_banks = %i\n", rscreen->info.r600_num_banks);
printf("num_render_backends = %i\n", rscreen->info.num_render_backends);
printf("num_tile_pipes = %i\n", rscreen->info.num_tile_pipes);
printf("pipe_interleave_bytes = %i\n", rscreen->info.pipe_interleave_bytes);
}
return true;
}
static HRESULT
NineVolumeTexture9_ctor( struct NineVolumeTexture9 *This,
struct NineUnknownParams *pParams,
UINT Width, UINT Height, UINT Depth, UINT Levels,
DWORD Usage,
D3DFORMAT Format,
D3DPOOL Pool,
HANDLE *pSharedHandle )
{
struct pipe_resource *info = &This->base.base.info;
struct pipe_screen *screen = pParams->device->screen;
enum pipe_format pf;
unsigned l;
D3DVOLUME_DESC voldesc;
HRESULT hr;
DBG("This=%p pParams=%p Width=%u Height=%u Depth=%u Levels=%u "
"Usage=%d Format=%d Pool=%d pSharedHandle=%p\n",
This, pParams, Width, Height, Depth, Levels,
Usage, Format, Pool, pSharedHandle);
user_assert(Width && Height && Depth, D3DERR_INVALIDCALL);
/* user_assert(!pSharedHandle || Pool == D3DPOOL_DEFAULT, D3DERR_INVALIDCALL); */
user_assert(!pSharedHandle, D3DERR_INVALIDCALL); /* TODO */
/* An IDirect3DVolume9 cannot be bound as a render target can it ? */
user_assert(!(Usage & (D3DUSAGE_RENDERTARGET | D3DUSAGE_DEPTHSTENCIL)),
D3DERR_INVALIDCALL);
user_assert(!(Usage & D3DUSAGE_AUTOGENMIPMAP), D3DERR_INVALIDCALL);
pf = d3d9_to_pipe_format_checked(screen, Format, PIPE_TEXTURE_3D, 0,
PIPE_BIND_SAMPLER_VIEW, FALSE,
Pool == D3DPOOL_SCRATCH);
if (pf == PIPE_FORMAT_NONE)
return D3DERR_INVALIDCALL;
/* We support ATI1 and ATI2 hacks only for 2D and Cube textures */
if (Format == D3DFMT_ATI1 || Format == D3DFMT_ATI2)
return D3DERR_INVALIDCALL;
if (compressed_format(Format)) {
const unsigned w = util_format_get_blockwidth(pf);
const unsigned h = util_format_get_blockheight(pf);
/* Compressed formats are not compressed on depth component */
user_assert(!(Width % w) && !(Height % h), D3DERR_INVALIDCALL);
}
info->screen = pParams->device->screen;
info->target = PIPE_TEXTURE_3D;
info->format = pf;
info->width0 = Width;
info->height0 = Height;
info->depth0 = Depth;
if (Levels)
info->last_level = Levels - 1;
else
info->last_level = util_logbase2(MAX2(MAX2(Width, Height), Depth));
info->array_size = 1;
info->nr_samples = 0;
info->bind = PIPE_BIND_SAMPLER_VIEW;
info->usage = PIPE_USAGE_DEFAULT;
info->flags = 0;
if (Usage & D3DUSAGE_DYNAMIC) {
info->usage = PIPE_USAGE_DYNAMIC;
info->bind |=
PIPE_BIND_TRANSFER_READ |
PIPE_BIND_TRANSFER_WRITE;
}
if (Usage & D3DUSAGE_SOFTWAREPROCESSING)
DBG("Application asked for Software Vertex Processing, "
"but this is unimplemented\n");
This->volumes = CALLOC(info->last_level + 1, sizeof(*This->volumes));
if (!This->volumes)
return E_OUTOFMEMORY;
This->base.pstype = 3;
hr = NineBaseTexture9_ctor(&This->base, pParams, NULL,
D3DRTYPE_VOLUMETEXTURE, Format, Pool, Usage);
if (FAILED(hr))
return hr;
voldesc.Format = Format;
voldesc.Type = D3DRTYPE_VOLUME;
voldesc.Usage = Usage;
voldesc.Pool = Pool;
for (l = 0; l <= info->last_level; ++l) {
voldesc.Width = u_minify(Width, l);
voldesc.Height = u_minify(Height, l);
voldesc.Depth = u_minify(Depth, l);
hr = NineVolume9_new(This->base.base.base.device, NineUnknown(This),
This->base.base.resource, l,
&voldesc, &This->volumes[l]);
if (FAILED(hr))
return hr;
}
/* Textures start initially dirty */
NineVolumeTexture9_AddDirtyBox(This, NULL);
return D3D_OK;
}
void r300_texture_setup_format_state(struct r300_screen *screen,
struct r300_resource *tex,
enum pipe_format format,
unsigned level,
unsigned width0_override,
unsigned height0_override,
struct r300_texture_format_state *out)
{
struct pipe_resource *pt = &tex->b.b;
struct r300_texture_desc *desc = &tex->tex;
boolean is_r500 = screen->caps.is_r500;
unsigned width, height, depth;
unsigned txwidth, txheight, txdepth;
width = u_minify(width0_override, level);
height = u_minify(height0_override, level);
depth = u_minify(desc->depth0, level);
txwidth = (width - 1) & 0x7ff;
txheight = (height - 1) & 0x7ff;
txdepth = util_logbase2(depth) & 0xf;
/* Mask out all the fields we change. */
out->format0 = 0;
out->format1 &= ~R300_TX_FORMAT_TEX_COORD_TYPE_MASK;
out->format2 &= R500_TXFORMAT_MSB;
out->tile_config = 0;
/* Set sampler state. */
out->format0 =
R300_TX_WIDTH(txwidth) |
R300_TX_HEIGHT(txheight) |
R300_TX_DEPTH(txdepth);
if (desc->uses_stride_addressing) {
unsigned stride =
r300_stride_to_width(format, desc->stride_in_bytes[level]);
/* rectangles love this */
out->format0 |= R300_TX_PITCH_EN;
out->format2 = (stride - 1) & 0x1fff;
}
if (pt->target == PIPE_TEXTURE_CUBE) {
out->format1 |= R300_TX_FORMAT_CUBIC_MAP;
}
if (pt->target == PIPE_TEXTURE_3D) {
out->format1 |= R300_TX_FORMAT_3D;
}
/* large textures on r500 */
if (is_r500)
{
unsigned us_width = txwidth;
unsigned us_height = txheight;
unsigned us_depth = txdepth;
if (width > 2048) {
out->format2 |= R500_TXWIDTH_BIT11;
}
if (height > 2048) {
out->format2 |= R500_TXHEIGHT_BIT11;
}
/* The US_FORMAT register fixes an R500 TX addressing bug.
* Don't ask why it must be set like this. I don't know it either. */
if (width > 2048) {
us_width = (0x000007FF + us_width) >> 1;
us_depth |= 0x0000000D;
}
if (height > 2048) {
us_height = (0x000007FF + us_height) >> 1;
us_depth |= 0x0000000E;
}
static HRESULT
NineTexture9_ctor( struct NineTexture9 *This,
struct NineUnknownParams *pParams,
UINT Width, UINT Height, UINT Levels,
DWORD Usage,
D3DFORMAT Format,
D3DPOOL Pool,
HANDLE *pSharedHandle )
{
struct pipe_screen *screen = pParams->device->screen;
struct pipe_resource *info = &This->base.base.info;
enum pipe_format pf;
unsigned *level_offsets;
unsigned l;
D3DSURFACE_DESC sfdesc;
HRESULT hr;
void *user_buffer = NULL, *user_buffer_for_level;
DBG("(%p) Width=%u Height=%u Levels=%u Usage=%s Format=%s Pool=%s "
"pSharedHandle=%p\n", This, Width, Height, Levels,
nine_D3DUSAGE_to_str(Usage),
d3dformat_to_string(Format), nine_D3DPOOL_to_str(Pool), pSharedHandle);
user_assert(Width && Height, D3DERR_INVALIDCALL);
/* pSharedHandle: can be non-null for ex only.
* D3DPOOL_SYSTEMMEM: Levels must be 1
* D3DPOOL_DEFAULT: no restriction for Levels
* Other Pools are forbidden. */
user_assert(!pSharedHandle || pParams->device->ex, D3DERR_INVALIDCALL);
user_assert(!pSharedHandle ||
(Pool == D3DPOOL_SYSTEMMEM && Levels == 1) ||
Pool == D3DPOOL_DEFAULT, D3DERR_INVALIDCALL);
user_assert(!(Usage & D3DUSAGE_AUTOGENMIPMAP) ||
(Pool != D3DPOOL_SYSTEMMEM && Pool != D3DPOOL_SCRATCH && Levels <= 1),
D3DERR_INVALIDCALL);
/* TODO: implement pSharedHandle for D3DPOOL_DEFAULT (cross process
* buffer sharing).
*
* Gem names may have fit but they're depreciated and won't work on render-nodes.
* One solution is to use shm buffers. We would use a /dev/shm file, fill the first
* values to tell it is a nine buffer, the size, which function created it, etc,
* and then it would contain the data. The handle would be a number, corresponding to
* the file to read (/dev/shm/nine-share-4 for example would be 4).
*
* Wine just ignores the argument, which works only if the app creates the handle
* and won't use it. Instead of failing, we support that situation by putting an
* invalid handle, that we would fail to import. Please note that we don't advertise
* the flag indicating the support for that feature, but apps seem to not care.
*/
if (pSharedHandle && Pool == D3DPOOL_DEFAULT) {
if (!*pSharedHandle) {
DBG("Creating Texture with invalid handle. Importing will fail\n.");
*pSharedHandle = (HANDLE)1; /* Wine would keep it NULL */
pSharedHandle = NULL;
} else {
ERR("Application tries to use cross-process sharing feature. Nine "
"doesn't support it");
return D3DERR_INVALIDCALL;
}
}
if (Usage & D3DUSAGE_AUTOGENMIPMAP)
Levels = 0;
pf = d3d9_to_pipe_format_checked(screen, Format, PIPE_TEXTURE_2D, 0,
PIPE_BIND_SAMPLER_VIEW, FALSE,
Pool == D3DPOOL_SCRATCH);
if (Format != D3DFMT_NULL && pf == PIPE_FORMAT_NONE)
return D3DERR_INVALIDCALL;
if (compressed_format(Format)) {
const unsigned w = util_format_get_blockwidth(pf);
const unsigned h = util_format_get_blockheight(pf);
user_assert(!(Width % w) && !(Height % h), D3DERR_INVALIDCALL);
}
info->screen = screen;
info->target = PIPE_TEXTURE_2D;
info->format = pf;
info->width0 = Width;
info->height0 = Height;
info->depth0 = 1;
if (Levels)
info->last_level = Levels - 1;
else
info->last_level = util_logbase2(MAX2(Width, Height));
info->array_size = 1;
info->nr_samples = 0;
info->nr_storage_samples = 0;
info->bind = PIPE_BIND_SAMPLER_VIEW;
info->usage = PIPE_USAGE_DEFAULT;
info->flags = 0;
if (Usage & D3DUSAGE_RENDERTARGET)
info->bind |= PIPE_BIND_RENDER_TARGET;
if (Usage & D3DUSAGE_DEPTHSTENCIL)
info->bind |= PIPE_BIND_DEPTH_STENCIL;
if (Usage & D3DUSAGE_DYNAMIC) {
info->usage = PIPE_USAGE_DYNAMIC;
}
if (Usage & D3DUSAGE_SOFTWAREPROCESSING)
DBG("Application asked for Software Vertex Processing, "
"but this is unimplemented\n");
if (pSharedHandle && *pSharedHandle) { /* Pool == D3DPOOL_SYSTEMMEM */
user_buffer = (void *)*pSharedHandle;
level_offsets = alloca(sizeof(unsigned) * (info->last_level + 1));
(void) nine_format_get_size_and_offsets(pf, level_offsets,
Width, Height,
info->last_level);
} else if (Pool != D3DPOOL_DEFAULT) {
/* TODO: For D3DUSAGE_AUTOGENMIPMAP, it is likely we only have to
* allocate only for the first level, since it is the only lockable
* level. Check apps don't crash if we allocate smaller buffer (some
* apps access sublevels of texture even if they locked only first
* level) */
level_offsets = alloca(sizeof(unsigned) * (info->last_level + 1));
user_buffer = align_calloc(
nine_format_get_size_and_offsets(pf, level_offsets,
Width, Height,
info->last_level), 32);
This->managed_buffer = user_buffer;
if (!This->managed_buffer)
return E_OUTOFMEMORY;
}
This->surfaces = CALLOC(info->last_level + 1, sizeof(*This->surfaces));
if (!This->surfaces)
return E_OUTOFMEMORY;
hr = NineBaseTexture9_ctor(&This->base, pParams, NULL, D3DRTYPE_TEXTURE, Format, Pool, Usage);
if (FAILED(hr))
return hr;
This->base.pstype = (Height == 1) ? 1 : 0;
/* Create all the surfaces right away.
* They manage backing storage, and transfers (LockRect) are deferred
* to them.
*/
sfdesc.Format = Format;
sfdesc.Type = D3DRTYPE_SURFACE;
sfdesc.Usage = Usage;
sfdesc.Pool = Pool;
sfdesc.MultiSampleType = D3DMULTISAMPLE_NONE;
sfdesc.MultiSampleQuality = 0;
for (l = 0; l <= info->last_level; ++l) {
sfdesc.Width = u_minify(Width, l);
sfdesc.Height = u_minify(Height, l);
/* Some apps expect the memory to be allocated in
* continous blocks */
user_buffer_for_level = user_buffer ? user_buffer +
level_offsets[l] : NULL;
hr = NineSurface9_new(This->base.base.base.device, NineUnknown(This),
This->base.base.resource, user_buffer_for_level,
D3DRTYPE_TEXTURE, l, 0,
&sfdesc, &This->surfaces[l]);
if (FAILED(hr))
return hr;
}
/* Textures start initially dirty */
This->dirty_rect.width = Width;
This->dirty_rect.height = Height;
This->dirty_rect.depth = 1; /* widht == 0 means empty, depth stays 1 */
if (pSharedHandle && !*pSharedHandle) {/* Pool == D3DPOOL_SYSTEMMEM */
*pSharedHandle = This->surfaces[0]->data;
}
return D3D_OK;
}
static void
nv50_screen_init_hwctx(struct nv50_screen *screen)
{
struct nouveau_pushbuf *push = screen->base.pushbuf;
struct nv04_fifo *fifo;
unsigned i;
fifo = (struct nv04_fifo *)screen->base.channel->data;
BEGIN_NV04(push, SUBC_M2MF(NV01_SUBCHAN_OBJECT), 1);
PUSH_DATA (push, screen->m2mf->handle);
BEGIN_NV04(push, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY), 3);
PUSH_DATA (push, screen->sync->handle);
PUSH_DATA (push, fifo->vram);
PUSH_DATA (push, fifo->vram);
BEGIN_NV04(push, SUBC_2D(NV01_SUBCHAN_OBJECT), 1);
PUSH_DATA (push, screen->eng2d->handle);
BEGIN_NV04(push, NV50_2D(DMA_NOTIFY), 4);
PUSH_DATA (push, screen->sync->handle);
PUSH_DATA (push, fifo->vram);
PUSH_DATA (push, fifo->vram);
PUSH_DATA (push, fifo->vram);
BEGIN_NV04(push, NV50_2D(OPERATION), 1);
PUSH_DATA (push, NV50_2D_OPERATION_SRCCOPY);
BEGIN_NV04(push, NV50_2D(CLIP_ENABLE), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_2D(COLOR_KEY_ENABLE), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, SUBC_2D(0x0888), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, SUBC_3D(NV01_SUBCHAN_OBJECT), 1);
PUSH_DATA (push, screen->tesla->handle);
BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
BEGIN_NV04(push, NV50_3D(DMA_NOTIFY), 1);
PUSH_DATA (push, screen->sync->handle);
BEGIN_NV04(push, NV50_3D(DMA_ZETA), 11);
for (i = 0; i < 11; ++i)
PUSH_DATA(push, fifo->vram);
BEGIN_NV04(push, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN);
for (i = 0; i < NV50_3D_DMA_COLOR__LEN; ++i)
PUSH_DATA(push, fifo->vram);
BEGIN_NV04(push, NV50_3D(REG_MODE), 1);
PUSH_DATA (push, NV50_3D_REG_MODE_STRIPED);
BEGIN_NV04(push, NV50_3D(UNK1400_LANES), 1);
PUSH_DATA (push, 0xf);
if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE)) {
BEGIN_NV04(push, NV50_3D(WATCHDOG_TIMER), 1);
PUSH_DATA (push, 0x18);
}
BEGIN_NV04(push, NV50_3D(RT_CONTROL), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_3D(CSAA_ENABLE), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(MULTISAMPLE_ENABLE), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(MULTISAMPLE_MODE), 1);
PUSH_DATA (push, NV50_3D_MULTISAMPLE_MODE_MS1);
BEGIN_NV04(push, NV50_3D(MULTISAMPLE_CTRL), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(LINE_LAST_PIXEL), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(BLEND_SEPARATE_ALPHA), 1);
PUSH_DATA (push, 1);
if (screen->tesla->oclass >= NVA0_3D_CLASS) {
BEGIN_NV04(push, SUBC_3D(NVA0_3D_TEX_MISC), 1);
PUSH_DATA (push, NVA0_3D_TEX_MISC_SEAMLESS_CUBE_MAP);
}
BEGIN_NV04(push, NV50_3D(SCREEN_Y_CONTROL), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(WINDOW_OFFSET_X), 2);
PUSH_DATA (push, 0);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(ZCULL_REGION), 1);
PUSH_DATA (push, 0x3f);
BEGIN_NV04(push, NV50_3D(VP_ADDRESS_HIGH), 2);
PUSH_DATAh(push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
PUSH_DATA (push, screen->code->offset + (0 << NV50_CODE_BO_SIZE_LOG2));
BEGIN_NV04(push, NV50_3D(FP_ADDRESS_HIGH), 2);
PUSH_DATAh(push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
PUSH_DATA (push, screen->code->offset + (1 << NV50_CODE_BO_SIZE_LOG2));
BEGIN_NV04(push, NV50_3D(GP_ADDRESS_HIGH), 2);
PUSH_DATAh(push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
PUSH_DATA (push, screen->code->offset + (2 << NV50_CODE_BO_SIZE_LOG2));
BEGIN_NV04(push, NV50_3D(LOCAL_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->tls_bo->offset);
PUSH_DATA (push, screen->tls_bo->offset);
PUSH_DATA (push, util_logbase2(screen->cur_tls_space / 8));
BEGIN_NV04(push, NV50_3D(STACK_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->stack_bo->offset);
PUSH_DATA (push, screen->stack_bo->offset);
PUSH_DATA (push, 4);
BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->uniforms->offset + (0 << 16));
PUSH_DATA (push, screen->uniforms->offset + (0 << 16));
PUSH_DATA (push, (NV50_CB_PVP << 16) | 0x0000);
BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->uniforms->offset + (1 << 16));
PUSH_DATA (push, screen->uniforms->offset + (1 << 16));
PUSH_DATA (push, (NV50_CB_PGP << 16) | 0x0000);
BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->uniforms->offset + (2 << 16));
PUSH_DATA (push, screen->uniforms->offset + (2 << 16));
PUSH_DATA (push, (NV50_CB_PFP << 16) | 0x0000);
BEGIN_NV04(push, NV50_3D(CB_DEF_ADDRESS_HIGH), 3);
PUSH_DATAh(push, screen->uniforms->offset + (3 << 16));
PUSH_DATA (push, screen->uniforms->offset + (3 << 16));
PUSH_DATA (push, (NV50_CB_AUX << 16) | 0x0200);
BEGIN_NI04(push, NV50_3D(SET_PROGRAM_CB), 3);
PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf01);
PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf21);
PUSH_DATA (push, (NV50_CB_AUX << 12) | 0xf31);
/* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
BEGIN_NV04(push, NV50_3D(CB_ADDR), 1);
PUSH_DATA (push, ((1 << 9) << 6) | NV50_CB_AUX);
BEGIN_NI04(push, NV50_3D(CB_DATA(0)), 4);
PUSH_DATAf(push, 0.0f);
PUSH_DATAf(push, 0.0f);
PUSH_DATAf(push, 0.0f);
PUSH_DATAf(push, 0.0f);
BEGIN_NV04(push, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH), 2);
PUSH_DATAh(push, screen->uniforms->offset + (3 << 16) + (1 << 9));
PUSH_DATA (push, screen->uniforms->offset + (3 << 16) + (1 << 9));
/* max TIC (bits 4:8) & TSC bindings, per program type */
for (i = 0; i < 3; ++i) {
BEGIN_NV04(push, NV50_3D(TEX_LIMITS(i)), 1);
PUSH_DATA (push, 0x54);
}
BEGIN_NV04(push, NV50_3D(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_3D(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_3D(LINKED_TSC), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(CLIP_RECTS_EN), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(CLIP_RECTS_MODE), 1);
PUSH_DATA (push, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY);
BEGIN_NV04(push, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
for (i = 0; i < 8 * 2; ++i)
PUSH_DATA(push, 0);
BEGIN_NV04(push, NV50_3D(CLIPID_ENABLE), 1);
PUSH_DATA (push, 0);
BEGIN_NV04(push, NV50_3D(VIEWPORT_TRANSFORM_EN), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_3D(DEPTH_RANGE_NEAR(0)), 2);
PUSH_DATAf(push, 0.0f);
PUSH_DATAf(push, 1.0f);
BEGIN_NV04(push, NV50_3D(VIEW_VOLUME_CLIP_CTRL), 1);
#ifdef NV50_SCISSORS_CLIPPING
PUSH_DATA (push, 0x0000);
#else
PUSH_DATA (push, 0x1080);
#endif
BEGIN_NV04(push, NV50_3D(CLEAR_FLAGS), 1);
PUSH_DATA (push, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT);
/* We use scissors instead of exact view volume clipping,
* so they're always enabled.
*/
BEGIN_NV04(push, NV50_3D(SCISSOR_ENABLE(0)), 3);
PUSH_DATA (push, 1);
PUSH_DATA (push, 8192 << 16);
PUSH_DATA (push, 8192 << 16);
BEGIN_NV04(push, NV50_3D(RASTERIZE_ENABLE), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_3D(POINT_RASTER_RULES), 1);
PUSH_DATA (push, NV50_3D_POINT_RASTER_RULES_OGL);
BEGIN_NV04(push, NV50_3D(FRAG_COLOR_CLAMP_EN), 1);
PUSH_DATA (push, 0x11111111);
BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
PUSH_DATA (push, 1);
PUSH_KICK (push);
}
