/**
* Define a vgpu10 blend state object for the given
* svga blend state.
*/
static void
define_blend_state_object(struct svga_context *svga,
struct svga_blend_state *bs)
{
SVGA3dDXBlendStatePerRT perRT[SVGA3D_MAX_RENDER_TARGETS];
unsigned try;
int i;
assert(svga_have_vgpu10(svga));
bs->id = util_bitmask_add(svga->blend_object_id_bm);
for (i = 0; i < SVGA3D_DX_MAX_RENDER_TARGETS; i++) {
perRT[i].blendEnable = bs->rt[i].blend_enable;
perRT[i].srcBlend = bs->rt[i].srcblend;
perRT[i].destBlend = bs->rt[i].dstblend;
perRT[i].blendOp = bs->rt[i].blendeq;
perRT[i].srcBlendAlpha = bs->rt[i].srcblend_alpha;
perRT[i].destBlendAlpha = bs->rt[i].dstblend_alpha;
perRT[i].blendOpAlpha = bs->rt[i].blendeq_alpha;
perRT[i].renderTargetWriteMask = bs->rt[i].writemask;
perRT[i].logicOpEnable = 0;
perRT[i].logicOp = SVGA3D_LOGICOP_COPY;
assert(perRT[i].srcBlend == perRT[0].srcBlend);
}
/* Loop in case command buffer is full and we need to flush and retry */
for (try = 0; try < 2; try++) {
enum pipe_error ret;
ret = SVGA3D_vgpu10_DefineBlendState(svga->swc,
bs->id,
bs->alpha_to_coverage,
bs->independent_blend_enable,
perRT);
if (ret == PIPE_OK)
return;
svga_context_flush(svga, NULL);
}
}
static void *
svga_create_blend_state(struct pipe_context *pipe,
const struct pipe_blend_state *templ)
{
struct svga_context *svga = svga_context(pipe);
struct svga_blend_state *blend = CALLOC_STRUCT( svga_blend_state );
unsigned i;
if (!blend)
return NULL;
/* Fill in the per-rendertarget blend state. We currently only
* support independent blend enable and colormask per render target.
*/
for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
/* No way to set this in SVGA3D, and no way to correctly implement it on
* top of D3D9 API. Instead we try to simulate with various blend modes.
*/
if (templ->logicop_enable) {
switch (templ->logicop_func) {
case PIPE_LOGICOP_XOR:
case PIPE_LOGICOP_INVERT:
blend->need_white_fragments = TRUE;
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_ONE;
blend->rt[i].dstblend = SVGA3D_BLENDOP_ONE;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_SUBTRACT;
break;
case PIPE_LOGICOP_CLEAR:
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_ZERO;
blend->rt[i].dstblend = SVGA3D_BLENDOP_ZERO;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_MINIMUM;
break;
case PIPE_LOGICOP_COPY:
blend->rt[i].blend_enable = FALSE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_ONE;
blend->rt[i].dstblend = SVGA3D_BLENDOP_ZERO;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_ADD;
break;
case PIPE_LOGICOP_COPY_INVERTED:
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_INVSRCCOLOR;
blend->rt[i].dstblend = SVGA3D_BLENDOP_ZERO;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_ADD;
break;
case PIPE_LOGICOP_NOOP:
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_ZERO;
blend->rt[i].dstblend = SVGA3D_BLENDOP_DESTCOLOR;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_ADD;
break;
case PIPE_LOGICOP_SET:
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_ONE;
blend->rt[i].dstblend = SVGA3D_BLENDOP_ONE;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_MAXIMUM;
break;
case PIPE_LOGICOP_AND:
/* Approximate with minimum - works for the 0 & anything case: */
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_SRCCOLOR;
blend->rt[i].dstblend = SVGA3D_BLENDOP_DESTCOLOR;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_MINIMUM;
break;
case PIPE_LOGICOP_AND_REVERSE:
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_SRCCOLOR;
blend->rt[i].dstblend = SVGA3D_BLENDOP_INVDESTCOLOR;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_MINIMUM;
break;
case PIPE_LOGICOP_AND_INVERTED:
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_INVSRCCOLOR;
blend->rt[i].dstblend = SVGA3D_BLENDOP_DESTCOLOR;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_MINIMUM;
break;
case PIPE_LOGICOP_OR:
/* Approximate with maximum - works for the 1 | anything case: */
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_SRCCOLOR;
blend->rt[i].dstblend = SVGA3D_BLENDOP_DESTCOLOR;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_MAXIMUM;
break;
case PIPE_LOGICOP_OR_REVERSE:
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_SRCCOLOR;
blend->rt[i].dstblend = SVGA3D_BLENDOP_INVDESTCOLOR;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_MAXIMUM;
break;
case PIPE_LOGICOP_OR_INVERTED:
blend->rt[i].blend_enable = TRUE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_INVSRCCOLOR;
blend->rt[i].dstblend = SVGA3D_BLENDOP_DESTCOLOR;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_MAXIMUM;
break;
case PIPE_LOGICOP_NAND:
case PIPE_LOGICOP_NOR:
case PIPE_LOGICOP_EQUIV:
/* Fill these in with plausible values */
blend->rt[i].blend_enable = FALSE;
blend->rt[i].srcblend = SVGA3D_BLENDOP_ONE;
blend->rt[i].dstblend = SVGA3D_BLENDOP_ZERO;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_ADD;
break;
default:
assert(0);
break;
}
blend->rt[i].srcblend_alpha = blend->rt[i].srcblend;
blend->rt[i].dstblend_alpha = blend->rt[i].dstblend;
blend->rt[i].blendeq_alpha = blend->rt[i].blendeq;
if (templ->logicop_func == PIPE_LOGICOP_XOR) {
pipe_debug_message(&svga->debug.callback, CONFORMANCE,
"XOR logicop mode has limited support");
}
else if (templ->logicop_func != PIPE_LOGICOP_COPY) {
pipe_debug_message(&svga->debug.callback, CONFORMANCE,
"general logicops are not supported");
}
}
else {
/* Note: the vgpu10 device does not yet support independent
* blend terms per render target. Target[0] always specifies the
* blending terms.
*/
if (templ->independent_blend_enable || templ->rt[0].blend_enable) {
/* always use the 0th target's blending terms for now */
blend->rt[i].srcblend =
svga_translate_blend_factor(svga, templ->rt[0].rgb_src_factor);
blend->rt[i].dstblend =
svga_translate_blend_factor(svga, templ->rt[0].rgb_dst_factor);
blend->rt[i].blendeq =
svga_translate_blend_func(templ->rt[0].rgb_func);
blend->rt[i].srcblend_alpha =
svga_translate_blend_factor(svga, templ->rt[0].alpha_src_factor);
blend->rt[i].dstblend_alpha =
svga_translate_blend_factor(svga, templ->rt[0].alpha_dst_factor);
blend->rt[i].blendeq_alpha =
svga_translate_blend_func(templ->rt[0].alpha_func);
if (blend->rt[i].srcblend_alpha != blend->rt[i].srcblend ||
blend->rt[i].dstblend_alpha != blend->rt[i].dstblend ||
blend->rt[i].blendeq_alpha != blend->rt[i].blendeq) {
blend->rt[i].separate_alpha_blend_enable = TRUE;
}
}
else {
/* disabled - default blend terms */
blend->rt[i].srcblend = SVGA3D_BLENDOP_ONE;
blend->rt[i].dstblend = SVGA3D_BLENDOP_ZERO;
blend->rt[i].blendeq = SVGA3D_BLENDEQ_ADD;
blend->rt[i].srcblend_alpha = SVGA3D_BLENDOP_ONE;
blend->rt[i].dstblend_alpha = SVGA3D_BLENDOP_ZERO;
blend->rt[i].blendeq_alpha = SVGA3D_BLENDEQ_ADD;
}
if (templ->independent_blend_enable) {
blend->rt[i].blend_enable = templ->rt[i].blend_enable;
}
else {
blend->rt[i].blend_enable = templ->rt[0].blend_enable;
}
}
/* Some GL blend modes are not supported by the VGPU9 device (there's
* no equivalent of PIPE_BLENDFACTOR_[INV_]CONST_ALPHA).
* When we set this flag, we copy the constant blend alpha value
* to the R, G, B components.
* This works as long as the src/dst RGB blend factors doesn't use
* PIPE_BLENDFACTOR_CONST_COLOR and PIPE_BLENDFACTOR_CONST_ALPHA
* at the same time. There's no work-around for that.
*/
if (!svga_have_vgpu10(svga)) {
if (templ->rt[0].rgb_src_factor == PIPE_BLENDFACTOR_CONST_ALPHA ||
templ->rt[0].rgb_dst_factor == PIPE_BLENDFACTOR_CONST_ALPHA ||
templ->rt[0].rgb_src_factor == PIPE_BLENDFACTOR_INV_CONST_ALPHA ||
templ->rt[0].rgb_dst_factor == PIPE_BLENDFACTOR_INV_CONST_ALPHA) {
blend->blend_color_alpha = TRUE;
}
}
if (templ->independent_blend_enable) {
blend->rt[i].writemask = templ->rt[i].colormask;
}
else {
blend->rt[i].writemask = templ->rt[0].colormask;
}
}
blend->independent_blend_enable = templ->independent_blend_enable;
blend->alpha_to_coverage = templ->alpha_to_coverage;
if (svga_have_vgpu10(svga)) {
define_blend_state_object(svga, blend);
}
svga->hud.num_blend_objects++;
SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
SVGA_STATS_COUNT_BLENDSTATE);
return blend;
}
static void svga_bind_blend_state(struct pipe_context *pipe,
void *blend)
{
struct svga_context *svga = svga_context(pipe);
svga->curr.blend = (struct svga_blend_state*)blend;
svga->dirty |= SVGA_NEW_BLEND;
}
static void svga_delete_blend_state(struct pipe_context *pipe,
void *blend)
{
struct svga_context *svga = svga_context(pipe);
struct svga_blend_state *bs =
(struct svga_blend_state *) blend;
if (bs->id != SVGA3D_INVALID_ID) {
enum pipe_error ret;
ret = SVGA3D_vgpu10_DestroyBlendState(svga->swc, bs->id);
if (ret != PIPE_OK) {
svga_context_flush(svga, NULL);
ret = SVGA3D_vgpu10_DestroyBlendState(svga->swc, bs->id);
assert(ret == PIPE_OK);
}
if (bs->id == svga->state.hw_draw.blend_id)
svga->state.hw_draw.blend_id = SVGA3D_INVALID_ID;
util_bitmask_clear(svga->blend_object_id_bm, bs->id);
bs->id = SVGA3D_INVALID_ID;
}
FREE(blend);
svga->hud.num_blend_objects--;
}
static void svga_set_blend_color( struct pipe_context *pipe,
const struct pipe_blend_color *blend_color )
{
struct svga_context *svga = svga_context(pipe);
svga->curr.blend_color = *blend_color;
svga->dirty |= SVGA_NEW_BLEND_COLOR;
}
void svga_init_blend_functions( struct svga_context *svga )
{
svga->pipe.create_blend_state = svga_create_blend_state;
svga->pipe.bind_blend_state = svga_bind_blend_state;
svga->pipe.delete_blend_state = svga_delete_blend_state;
svga->pipe.set_blend_color = svga_set_blend_color;
}
/**
* Define a vgpu10 sampler state.
*/
static void
define_sampler_state_object(struct svga_context *svga,
struct svga_sampler_state *ss,
const struct pipe_sampler_state *ps)
{
uint8_t max_aniso = (uint8_t) 255; /* XXX fix me */
boolean anisotropic;
uint8 compare_func;
SVGA3dFilter filter;
SVGA3dRGBAFloat bcolor;
unsigned try;
float min_lod, max_lod;
assert(svga_have_vgpu10(svga));
anisotropic = ss->aniso_level > 1.0f;
filter = translate_filter_mode(ps->min_mip_filter,
ps->min_img_filter,
ps->mag_img_filter,
anisotropic,
ss->compare_mode);
compare_func = translate_comparison_func(ss->compare_func);
COPY_4V(bcolor.value, ps->border_color.f);
assert(ps->min_lod <= ps->max_lod);
if (ps->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
/* just use the base level image */
min_lod = max_lod = 0.0f;
}
else {
min_lod = ps->min_lod;
max_lod = ps->max_lod;
}
/* If shadow comparisons are enabled, create two sampler states: one
* with the given shadow compare mode, another with shadow comparison off.
* We need the later because in some cases, we have to do the shadow
* compare in the shader. So, we don't want to do it twice.
*/
STATIC_ASSERT(PIPE_TEX_COMPARE_NONE == 0);
STATIC_ASSERT(PIPE_TEX_COMPARE_R_TO_TEXTURE == 1);
ss->id[1] = SVGA3D_INVALID_ID;
unsigned i;
for (i = 0; i <= ss->compare_mode; i++) {
ss->id[i] = util_bitmask_add(svga->sampler_object_id_bm);
/* Loop in case command buffer is full and we need to flush and retry */
for (try = 0; try < 2; try++) {
enum pipe_error ret =
SVGA3D_vgpu10_DefineSamplerState(svga->swc,
ss->id[i],
filter,
ss->addressu,
ss->addressv,
ss->addressw,
ss->lod_bias, /* float */
max_aniso,
compare_func,
bcolor,
min_lod, /* float */
max_lod); /* float */
if (ret == PIPE_OK)
break;
svga_context_flush(svga, NULL);
}
/* turn off the shadow compare option for second iteration */
filter &= ~SVGA3D_FILTER_COMPARE;
}
}
static void *
svga_create_sampler_state(struct pipe_context *pipe,
const struct pipe_sampler_state *sampler)
{
struct svga_context *svga = svga_context(pipe);
struct svga_sampler_state *cso = CALLOC_STRUCT( svga_sampler_state );
if (!cso)
return NULL;
cso->mipfilter = translate_mip_filter(sampler->min_mip_filter);
cso->magfilter = translate_img_filter( sampler->mag_img_filter );
cso->minfilter = translate_img_filter( sampler->min_img_filter );
cso->aniso_level = MAX2( sampler->max_anisotropy, 1 );
if (sampler->max_anisotropy)
cso->magfilter = cso->minfilter = SVGA3D_TEX_FILTER_ANISOTROPIC;
cso->lod_bias = sampler->lod_bias;
cso->addressu = translate_wrap_mode(sampler->wrap_s);
cso->addressv = translate_wrap_mode(sampler->wrap_t);
cso->addressw = translate_wrap_mode(sampler->wrap_r);
cso->normalized_coords = sampler->normalized_coords;
cso->compare_mode = sampler->compare_mode;
cso->compare_func = sampler->compare_func;
{
uint32 r = float_to_ubyte(sampler->border_color.f[0]);
uint32 g = float_to_ubyte(sampler->border_color.f[1]);
uint32 b = float_to_ubyte(sampler->border_color.f[2]);
uint32 a = float_to_ubyte(sampler->border_color.f[3]);
cso->bordercolor = (a << 24) | (r << 16) | (g << 8) | b;
}
/* No SVGA3D support for:
* - min/max LOD clamping
*/
cso->min_lod = 0;
cso->view_min_lod = MAX2((int) (sampler->min_lod + 0.5), 0);
cso->view_max_lod = MAX2((int) (sampler->max_lod + 0.5), 0);
/* Use min_mipmap */
if (svga->debug.use_min_mipmap) {
if (cso->view_min_lod == cso->view_max_lod) {
cso->min_lod = cso->view_min_lod;
cso->view_min_lod = 0;
cso->view_max_lod = 1000; /* Just a high number */
cso->mipfilter = SVGA3D_TEX_FILTER_NONE;
}
}
if (svga_have_vgpu10(svga)) {
define_sampler_state_object(svga, cso, sampler);
}
SVGA_DBG(DEBUG_SAMPLERS,
"New sampler: min %u, view(min %u, max %u) lod, mipfilter %s\n",
cso->min_lod, cso->view_min_lod, cso->view_max_lod,
cso->mipfilter == SVGA3D_TEX_FILTER_NONE ? "SVGA3D_TEX_FILTER_NONE" : "SOMETHING");
svga->hud.num_sampler_objects++;
SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
SVGA_STATS_COUNT_SAMPLER);
return cso;
}
static void
svga_bind_sampler_states(struct pipe_context *pipe,
enum pipe_shader_type shader,
unsigned start,
unsigned num,
void **samplers)
{
struct svga_context *svga = svga_context(pipe);
unsigned i;
boolean any_change = FALSE;
assert(shader < PIPE_SHADER_TYPES);
assert(start + num <= PIPE_MAX_SAMPLERS);
/* Pre-VGPU10 only supports FS textures */
if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
return;
for (i = 0; i < num; i++) {
if (svga->curr.sampler[shader][start + i] != samplers[i])
any_change = TRUE;
svga->curr.sampler[shader][start + i] = samplers[i];
}
if (!any_change) {
return;
}
/* find highest non-null sampler[] entry */
{
unsigned j = MAX2(svga->curr.num_samplers[shader], start + num);
while (j > 0 && svga->curr.sampler[shader][j - 1] == NULL)
j--;
svga->curr.num_samplers[shader] = j;
}
svga->dirty |= SVGA_NEW_SAMPLER;
}
static void
svga_delete_sampler_state(struct pipe_context *pipe, void *sampler)
{
struct svga_sampler_state *ss = (struct svga_sampler_state *) sampler;
struct svga_context *svga = svga_context(pipe);
if (svga_have_vgpu10(svga)) {
unsigned i;
for (i = 0; i < 2; i++) {
enum pipe_error ret;
if (ss->id[i] != SVGA3D_INVALID_ID) {
svga_hwtnl_flush_retry(svga);
ret = SVGA3D_vgpu10_DestroySamplerState(svga->swc, ss->id[i]);
if (ret != PIPE_OK) {
svga_context_flush(svga, NULL);
ret = SVGA3D_vgpu10_DestroySamplerState(svga->swc, ss->id[i]);
}
util_bitmask_clear(svga->sampler_object_id_bm, ss->id[i]);
}
}
}
FREE(sampler);
svga->hud.num_sampler_objects--;
}
static struct pipe_sampler_view *
svga_create_sampler_view(struct pipe_context *pipe,
struct pipe_resource *texture,
const struct pipe_sampler_view *templ)
{
struct svga_context *svga = svga_context(pipe);
struct svga_pipe_sampler_view *sv = CALLOC_STRUCT(svga_pipe_sampler_view);
if (!sv) {
return NULL;
}
sv->base = *templ;
sv->base.reference.count = 1;
sv->base.texture = NULL;
pipe_resource_reference(&sv->base.texture, texture);
sv->base.context = pipe;
sv->id = SVGA3D_INVALID_ID;
svga->hud.num_samplerview_objects++;
SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
SVGA_STATS_COUNT_SAMPLERVIEW);
return &sv->base;
}
static void
svga_sampler_view_destroy(struct pipe_context *pipe,
struct pipe_sampler_view *view)
{
struct svga_context *svga = svga_context(pipe);
struct svga_pipe_sampler_view *sv = svga_pipe_sampler_view(view);
if (svga_have_vgpu10(svga) && sv->id != SVGA3D_INVALID_ID) {
if (view->context != pipe) {
/* The SVGA3D device will generate an error (and on Linux, cause
* us to abort) if we try to destroy a shader resource view from
* a context other than the one it was created with. Skip the
* SVGA3D_vgpu10_DestroyShaderResourceView() and leak the sampler
* view for now. This should only sometimes happen when a shared
* texture is deleted.
*/
_debug_printf("context mismatch in %s\n", __func__);
}
else {
enum pipe_error ret;
svga_hwtnl_flush_retry(svga); /* XXX is this needed? */
ret = SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc, sv->id);
if (ret != PIPE_OK) {
svga_context_flush(svga, NULL);
ret = SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc, sv->id);
}
util_bitmask_clear(svga->sampler_view_id_bm, sv->id);
}
}
pipe_resource_reference(&sv->base.texture, NULL);
FREE(sv);
svga->hud.num_samplerview_objects--;
}
static void
svga_set_sampler_views(struct pipe_context *pipe,
enum pipe_shader_type shader,
unsigned start,
unsigned num,
struct pipe_sampler_view **views)
{
struct svga_context *svga = svga_context(pipe);
unsigned flag_1d = 0;
unsigned flag_srgb = 0;
uint i;
boolean any_change = FALSE;
assert(shader < PIPE_SHADER_TYPES);
assert(start + num <= ARRAY_SIZE(svga->curr.sampler_views[shader]));
/* Pre-VGPU10 only supports FS textures */
if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
return;
SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_SETSAMPLERVIEWS);
/* This bit of code works around a quirk in the CSO module.
* If start=num=0 it means all sampler views should be released.
* Note that the CSO module treats sampler views for fragment shaders
* differently than other shader types.
*/
if (start == 0 && num == 0 && svga->curr.num_sampler_views[shader] > 0) {
for (i = 0; i < svga->curr.num_sampler_views[shader]; i++) {
pipe_sampler_view_release(pipe, &svga->curr.sampler_views[shader][i]);
}
any_change = TRUE;
}
for (i = 0; i < num; i++) {
enum pipe_texture_target target;
if (svga->curr.sampler_views[shader][start + i] != views[i]) {
/* Note: we're using pipe_sampler_view_release() here to work around
* a possible crash when the old view belongs to another context that
* was already destroyed.
*/
pipe_sampler_view_release(pipe, &svga->curr.sampler_views[shader][start + i]);
pipe_sampler_view_reference(&svga->curr.sampler_views[shader][start + i],
views[i]);
any_change = TRUE;
}
if (!views[i])
continue;
if (util_format_is_srgb(views[i]->format))
flag_srgb |= 1 << (start + i);
target = views[i]->target;
if (target == PIPE_TEXTURE_1D) {
flag_1d |= 1 << (start + i);
} else if (target == PIPE_TEXTURE_RECT) {
/* If the size of the bound texture changes, we need to emit new
* const buffer values.
*/
svga->dirty |= SVGA_NEW_TEXTURE_CONSTS;
} else if (target == PIPE_BUFFER) {
/* If the size of the bound buffer changes, we need to emit new
* const buffer values.
*/
svga->dirty |= SVGA_NEW_TEXTURE_CONSTS;
}
}
if (!any_change) {
goto done;
}
/* find highest non-null sampler_views[] entry */
{
unsigned j = MAX2(svga->curr.num_sampler_views[shader], start + num);
while (j > 0 && svga->curr.sampler_views[shader][j - 1] == NULL)
j--;
svga->curr.num_sampler_views[shader] = j;
}
svga->dirty |= SVGA_NEW_TEXTURE_BINDING;
if (flag_srgb != svga->curr.tex_flags.flag_srgb ||
flag_1d != svga->curr.tex_flags.flag_1d) {
svga->dirty |= SVGA_NEW_TEXTURE_FLAGS;
svga->curr.tex_flags.flag_1d = flag_1d;
svga->curr.tex_flags.flag_srgb = flag_srgb;
}
/* Check if any of the sampler view resources collide with the framebuffer
* color buffers or depth stencil resource. If so, set the NEW_FRAME_BUFFER
* dirty bit so that emit_framebuffer can be invoked to create backed view
* for the conflicted surface view.
*/
if (svga_check_sampler_framebuffer_resource_collision(svga, shader)) {
svga->dirty |= SVGA_NEW_FRAME_BUFFER;
}
done:
SVGA_STATS_TIME_POP(svga_sws(svga));
}
/**
* Clean up sampler, sampler view state at context destruction time
*/
void
svga_cleanup_sampler_state(struct svga_context *svga)
{
enum pipe_shader_type shader;
for (shader = 0; shader <= PIPE_SHADER_GEOMETRY; shader++) {
unsigned i;
for (i = 0; i < svga->state.hw_draw.num_sampler_views[shader]; i++) {
pipe_sampler_view_release(&svga->pipe,
&svga->state.hw_draw.sampler_views[shader][i]);
}
}
/* free polygon stipple state */
if (svga->polygon_stipple.sampler) {
svga->pipe.delete_sampler_state(&svga->pipe, svga->polygon_stipple.sampler);
}
if (svga->polygon_stipple.sampler_view) {
svga->pipe.sampler_view_destroy(&svga->pipe,
&svga->polygon_stipple.sampler_view->base);
}
pipe_resource_reference(&svga->polygon_stipple.texture, NULL);
}
void
svga_init_sampler_functions( struct svga_context *svga )
{
svga->pipe.create_sampler_state = svga_create_sampler_state;
svga->pipe.bind_sampler_states = svga_bind_sampler_states;
svga->pipe.delete_sampler_state = svga_delete_sampler_state;
svga->pipe.set_sampler_views = svga_set_sampler_views;
svga->pipe.create_sampler_view = svga_create_sampler_view;
svga->pipe.sampler_view_destroy = svga_sampler_view_destroy;
}
/**
* Use direct map for the transfer request
*/
static void *
svga_texture_transfer_map_direct(struct svga_context *svga,
struct svga_transfer *st)
{
struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
struct pipe_transfer *transfer = &st->base;
struct pipe_resource *texture = transfer->resource;
struct svga_texture *tex = svga_texture(texture);
struct svga_winsys_surface *surf = tex->handle;
unsigned level = st->base.level;
unsigned w, h, nblocksx, nblocksy, i;
unsigned usage = st->base.usage;
if (need_tex_readback(st)) {
enum pipe_error ret;
svga_surfaces_flush(svga);
for (i = 0; i < st->box.d; i++) {
if (svga_have_vgpu10(svga)) {
ret = readback_image_vgpu10(svga, surf, st->slice + i, level,
tex->b.b.last_level + 1);
} else {
ret = readback_image_vgpu9(svga, surf, st->slice + i, level);
}
}
svga->hud.num_readbacks++;
SVGA_STATS_COUNT_INC(sws, SVGA_STATS_COUNT_TEXREADBACK);
assert(ret == PIPE_OK);
(void) ret;
svga_context_flush(svga, NULL);
/*
* Note: if PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE were specified
* we could potentially clear the flag for all faces/layers/mips.
*/
svga_clear_texture_rendered_to(tex, st->slice, level);
}
else {
assert(usage & PIPE_TRANSFER_WRITE);
if ((usage & PIPE_TRANSFER_UNSYNCHRONIZED) == 0) {
if (svga_is_texture_dirty(tex, st->slice, level)) {
/*
* do a surface flush if the subresource has been modified
* in this command buffer.
*/
svga_surfaces_flush(svga);
if (!sws->surface_is_flushed(sws, surf)) {
svga->hud.surface_write_flushes++;
SVGA_STATS_COUNT_INC(sws, SVGA_STATS_COUNT_SURFACEWRITEFLUSH);
svga_context_flush(svga, NULL);
}
}
}
}
/* we'll directly access the guest-backed surface */
w = u_minify(texture->width0, level);
h = u_minify(texture->height0, level);
nblocksx = util_format_get_nblocksx(texture->format, w);
nblocksy = util_format_get_nblocksy(texture->format, h);
st->hw_nblocksy = nblocksy;
st->base.stride = nblocksx*util_format_get_blocksize(texture->format);
st->base.layer_stride = st->base.stride * nblocksy;
/*
* Begin mapping code
*/
{
SVGA3dSize baseLevelSize;
uint8_t *map;
boolean retry;
unsigned offset, mip_width, mip_height;
map = svga->swc->surface_map(svga->swc, surf, usage, &retry);
if (map == NULL && retry) {
/*
* At this point, the svga_surfaces_flush() should already have
* called in svga_texture_get_transfer().
*/
svga->hud.surface_write_flushes++;
svga_context_flush(svga, NULL);
map = svga->swc->surface_map(svga->swc, surf, usage, &retry);
}
/*
* Make sure we return NULL if the map fails
*/
if (!map) {
return NULL;
}
/**
* Compute the offset to the specific texture slice in the buffer.
*/
baseLevelSize.width = tex->b.b.width0;
baseLevelSize.height = tex->b.b.height0;
baseLevelSize.depth = tex->b.b.depth0;
if ((tex->b.b.target == PIPE_TEXTURE_1D_ARRAY) ||
(tex->b.b.target == PIPE_TEXTURE_2D_ARRAY) ||
(tex->b.b.target == PIPE_TEXTURE_CUBE_ARRAY)) {
st->base.layer_stride =
svga3dsurface_get_image_offset(tex->key.format, baseLevelSize,
tex->b.b.last_level + 1, 1, 0);
}
offset = svga3dsurface_get_image_offset(tex->key.format, baseLevelSize,
tex->b.b.last_level + 1, /* numMips */
st->slice, level);
if (level > 0) {
assert(offset > 0);
}
mip_width = u_minify(tex->b.b.width0, level);
mip_height = u_minify(tex->b.b.height0, level);
offset += svga3dsurface_get_pixel_offset(tex->key.format,
mip_width, mip_height,
st->box.x,
st->box.y,
st->box.z);
return (void *) (map + offset);
}
}
static void
define_rasterizer_object(struct svga_context *svga,
struct svga_rasterizer_state *rast)
{
struct svga_screen *svgascreen = svga_screen(svga->pipe.screen);
unsigned fill_mode = translate_fill_mode(rast->templ.fill_front);
const unsigned cull_mode = translate_cull_mode(rast->templ.cull_face);
const int depth_bias = rast->templ.offset_units;
const float slope_scaled_depth_bias = rast->templ.offset_scale;
/* PIPE_CAP_POLYGON_OFFSET_CLAMP not supported: */
const float depth_bias_clamp = 0.0;
const float line_width = rast->templ.line_width > 0.0f ?
rast->templ.line_width : 1.0f;
const uint8 line_factor = rast->templ.line_stipple_enable ?
rast->templ.line_stipple_factor : 0;
const uint16 line_pattern = rast->templ.line_stipple_enable ?
rast->templ.line_stipple_pattern : 0;
unsigned try;
rast->id = util_bitmask_add(svga->rast_object_id_bm);
if (rast->templ.fill_front != rast->templ.fill_back) {
/* The VGPU10 device can't handle different front/back fill modes.
* We'll handle that with a swtnl/draw fallback. But we need to
* make sure we always fill triangles in that case.
*/
fill_mode = SVGA3D_FILLMODE_FILL;
}
for (try = 0; try < 2; try++) {
const uint8 pv_last = !rast->templ.flatshade_first &&
svgascreen->haveProvokingVertex;
enum pipe_error ret =
SVGA3D_vgpu10_DefineRasterizerState(svga->swc,
rast->id,
fill_mode,
cull_mode,
rast->templ.front_ccw,
depth_bias,
depth_bias_clamp,
slope_scaled_depth_bias,
rast->templ.depth_clip_near,
rast->templ.scissor,
rast->templ.multisample,
rast->templ.line_smooth,
line_width,
rast->templ.line_stipple_enable,
line_factor,
line_pattern,
pv_last);
if (ret == PIPE_OK)
return;
svga_context_flush(svga, NULL);
}
}
static void *
svga_create_rasterizer_state(struct pipe_context *pipe,
const struct pipe_rasterizer_state *templ)
{
struct svga_context *svga = svga_context(pipe);
struct svga_rasterizer_state *rast = CALLOC_STRUCT(svga_rasterizer_state);
struct svga_screen *screen = svga_screen(pipe->screen);
if (!rast)
return NULL;
/* need this for draw module. */
rast->templ = *templ;
rast->shademode = svga_translate_flatshade(templ->flatshade);
rast->cullmode = svga_translate_cullmode(templ->cull_face, templ->front_ccw);
rast->scissortestenable = templ->scissor;
rast->multisampleantialias = templ->multisample;
rast->antialiasedlineenable = templ->line_smooth;
rast->lastpixel = templ->line_last_pixel;
rast->pointsprite = templ->point_quad_rasterization;
if (rast->templ.multisample) {
/* The OpenGL 3.0 spec says points are always drawn as circles when
* MSAA is enabled. Note that our implementation isn't 100% correct,
* though. Our smooth point implementation involves drawing a square,
* computing fragment distance from point center, then attenuating
* the fragment alpha value. We should not attenuate alpha if msaa
* is enabled. We should kill fragments entirely outside the circle
* and let the GPU compute per-fragment coverage.
* But as-is, our implementation gives acceptable results and passes
* Piglit's MSAA point smooth test.
*/
rast->templ.point_smooth = TRUE;
}
if (templ->point_smooth) {
/* For smooth points we need to generate fragments for at least
* a 2x2 region. Otherwise the quad we draw may be too small and
* we may generate no fragments at all.
*/
rast->pointsize = MAX2(2.0f, templ->point_size);
}
else {
rast->pointsize = templ->point_size;
}
rast->hw_fillmode = PIPE_POLYGON_MODE_FILL;
/* Use swtnl + decomposition implement these:
*/
if (templ->line_width <= screen->maxLineWidth) {
/* pass line width to device */
rast->linewidth = MAX2(1.0F, templ->line_width);
}
else if (svga->debug.no_line_width) {
/* nothing */
}
else {
/* use 'draw' pipeline for wide line */
rast->need_pipeline |= SVGA_PIPELINE_FLAG_LINES;
rast->need_pipeline_lines_str = "line width";
}
if (templ->line_stipple_enable) {
if (screen->haveLineStipple || svga->debug.force_hw_line_stipple) {
SVGA3dLinePattern lp;
lp.repeat = templ->line_stipple_factor + 1;
lp.pattern = templ->line_stipple_pattern;
rast->linepattern = lp.uintValue;
}
else {
/* use 'draw' module to decompose into short line segments */
rast->need_pipeline |= SVGA_PIPELINE_FLAG_LINES;
rast->need_pipeline_lines_str = "line stipple";
}
}
if (!svga_have_vgpu10(svga) && templ->point_smooth) {
rast->need_pipeline |= SVGA_PIPELINE_FLAG_POINTS;
rast->need_pipeline_points_str = "smooth points";
}
if (templ->line_smooth && !screen->haveLineSmooth) {
/*
* XXX: Enabling the pipeline slows down performance immensely, so ignore
* line smooth state, where there is very little visual improvement.
* Smooth lines will still be drawn for wide lines.
*/
#if 0
rast->need_pipeline |= SVGA_PIPELINE_FLAG_LINES;
rast->need_pipeline_lines_str = "smooth lines";
#endif
}
{
int fill_front = templ->fill_front;
int fill_back = templ->fill_back;
int fill = PIPE_POLYGON_MODE_FILL;
boolean offset_front = util_get_offset(templ, fill_front);
boolean offset_back = util_get_offset(templ, fill_back);
boolean offset = FALSE;
switch (templ->cull_face) {
case PIPE_FACE_FRONT_AND_BACK:
offset = FALSE;
fill = PIPE_POLYGON_MODE_FILL;
break;
case PIPE_FACE_FRONT:
offset = offset_back;
fill = fill_back;
break;
case PIPE_FACE_BACK:
offset = offset_front;
fill = fill_front;
break;
case PIPE_FACE_NONE:
if (fill_front != fill_back || offset_front != offset_back) {
/* Always need the draw module to work out different
* front/back fill modes:
*/
rast->need_pipeline |= SVGA_PIPELINE_FLAG_TRIS;
rast->need_pipeline_tris_str = "different front/back fillmodes";
fill = PIPE_POLYGON_MODE_FILL;
}
else {
offset = offset_front;
fill = fill_front;
}
break;
default:
assert(0);
break;
}
/* Unfilled primitive modes aren't implemented on all virtual
* hardware. We can do some unfilled processing with index
* translation, but otherwise need the draw module:
*/
if (fill != PIPE_POLYGON_MODE_FILL &&
(templ->flatshade ||
templ->light_twoside ||
offset)) {
fill = PIPE_POLYGON_MODE_FILL;
rast->need_pipeline |= SVGA_PIPELINE_FLAG_TRIS;
rast->need_pipeline_tris_str = "unfilled primitives with no index manipulation";
}
/* If we are decomposing to lines, and lines need the pipeline,
* then we also need the pipeline for tris.
*/
if (fill == PIPE_POLYGON_MODE_LINE &&
(rast->need_pipeline & SVGA_PIPELINE_FLAG_LINES)) {
fill = PIPE_POLYGON_MODE_FILL;
rast->need_pipeline |= SVGA_PIPELINE_FLAG_TRIS;
rast->need_pipeline_tris_str = "decomposing lines";
}
/* Similarly for points:
*/
if (fill == PIPE_POLYGON_MODE_POINT &&
(rast->need_pipeline & SVGA_PIPELINE_FLAG_POINTS)) {
fill = PIPE_POLYGON_MODE_FILL;
rast->need_pipeline |= SVGA_PIPELINE_FLAG_TRIS;
rast->need_pipeline_tris_str = "decomposing points";
}
if (offset) {
rast->slopescaledepthbias = templ->offset_scale;
rast->depthbias = templ->offset_units;
}
rast->hw_fillmode = fill;
}
if (rast->need_pipeline & SVGA_PIPELINE_FLAG_TRIS) {
/* Turn off stuff which will get done in the draw module:
*/
rast->hw_fillmode = PIPE_POLYGON_MODE_FILL;
rast->slopescaledepthbias = 0;
rast->depthbias = 0;
}
if (0 && rast->need_pipeline) {
debug_printf("svga: rast need_pipeline = 0x%x\n", rast->need_pipeline);
debug_printf(" pnts: %s \n", rast->need_pipeline_points_str);
debug_printf(" lins: %s \n", rast->need_pipeline_lines_str);
debug_printf(" tris: %s \n", rast->need_pipeline_tris_str);
}
if (svga_have_vgpu10(svga)) {
define_rasterizer_object(svga, rast);
}
if (templ->poly_smooth) {
pipe_debug_message(&svga->debug.callback, CONFORMANCE,
"GL_POLYGON_SMOOTH not supported");
}
svga->hud.num_rasterizer_objects++;
SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
SVGA_STATS_COUNT_RASTERIZERSTATE);
return rast;
}
static void
svga_bind_rasterizer_state(struct pipe_context *pipe, void *state)
{
struct svga_context *svga = svga_context(pipe);
struct svga_rasterizer_state *raster = (struct svga_rasterizer_state *)state;
if (!raster || !svga->curr.rast) {
svga->dirty |= SVGA_NEW_STIPPLE | SVGA_NEW_DEPTH_STENCIL_ALPHA;
}
else {
if (raster->templ.poly_stipple_enable !=
svga->curr.rast->templ.poly_stipple_enable) {
svga->dirty |= SVGA_NEW_STIPPLE;
}
if (raster->templ.rasterizer_discard !=
svga->curr.rast->templ.rasterizer_discard) {
svga->dirty |= SVGA_NEW_DEPTH_STENCIL_ALPHA;
}
}
svga->curr.rast = raster;
svga->dirty |= SVGA_NEW_RAST;
}
static void
svga_delete_rasterizer_state(struct pipe_context *pipe, void *state)
{
struct svga_context *svga = svga_context(pipe);
struct svga_rasterizer_state *raster =
(struct svga_rasterizer_state *) state;
if (svga_have_vgpu10(svga)) {
enum pipe_error ret =
SVGA3D_vgpu10_DestroyRasterizerState(svga->swc, raster->id);
if (ret != PIPE_OK) {
svga_context_flush(svga, NULL);
ret = SVGA3D_vgpu10_DestroyRasterizerState(svga->swc, raster->id);
}
if (raster->id == svga->state.hw_draw.rasterizer_id)
svga->state.hw_draw.rasterizer_id = SVGA3D_INVALID_ID;
util_bitmask_clear(svga->rast_object_id_bm, raster->id);
}
FREE(state);
svga->hud.num_rasterizer_objects--;
}
void
svga_init_rasterizer_functions(struct svga_context *svga)
{
svga->pipe.create_rasterizer_state = svga_create_rasterizer_state;
svga->pipe.bind_rasterizer_state = svga_bind_rasterizer_state;
svga->pipe.delete_rasterizer_state = svga_delete_rasterizer_state;
}