void lru_put_into_cache(char *name,char *ip){
int index,temp;
node *ptr=get(name);
if(ptr){
printf("strange,it should not exist\n");
return;
}
if(ehead){
ptr=ehead;
if(ehead==etail)
ehead=etail=NULL;
else{
ehead=ehead->next;
ptr->next=NULL;
}
index=cache_num++;
}
else{
ptr=tail;
tail=tail->prev;
tail->next=NULL;
ptr->prev=NULL;
temp=ptr->hash_index;
removebykey(temp);
index=ptr->cache_index;
}
cache[index]=ip;
ptr->cache_index=index;
temp=put(name,ptr);
ptr->hash_index=temp;
move_to_head(ptr);
}
/**
* Update fragment state. This is called just prior to drawing
* something when some fragment-related state has changed.
*/
void
llvmpipe_update_fs(struct llvmpipe_context *lp)
{
struct lp_fragment_shader *shader = lp->fs;
struct lp_fragment_shader_variant_key key;
struct lp_fragment_shader_variant *variant = NULL;
struct lp_fs_variant_list_item *li;
make_variant_key(lp, shader, &key);
li = first_elem(&shader->variants);
while(!at_end(&shader->variants, li)) {
if(memcmp(&li->base->key, &key, shader->variant_key_size) == 0) {
variant = li->base;
break;
}
li = next_elem(li);
}
if (variant) {
move_to_head(&lp->fs_variants_list, &variant->list_item_global);
}
else {
int64_t t0, t1;
int64_t dt;
unsigned i;
if (lp->nr_fs_variants >= LP_MAX_SHADER_VARIANTS) {
struct pipe_context *pipe = &lp->pipe;
/*
* XXX: we need to flush the context until we have some sort of reference
* counting in fragment shaders as they may still be binned
* Flushing alone might not be sufficient we need to wait on it too.
*/
llvmpipe_finish(pipe, __FUNCTION__);
for (i = 0; i < LP_MAX_SHADER_VARIANTS / 4; i++) {
struct lp_fs_variant_list_item *item = last_elem(&lp->fs_variants_list);
remove_shader_variant(lp, item->base);
}
}
t0 = os_time_get();
variant = generate_variant(lp, shader, &key);
t1 = os_time_get();
dt = t1 - t0;
LP_COUNT_ADD(llvm_compile_time, dt);
LP_COUNT_ADD(nr_llvm_compiles, 2); /* emit vs. omit in/out test */
if (variant) {
insert_at_head(&shader->variants, &variant->list_item_local);
insert_at_head(&lp->fs_variants_list, &variant->list_item_global);
lp->nr_fs_variants++;
shader->variants_cached++;
}
}
lp_setup_set_fs_variant(lp->setup, variant);
}
char* lru_search_cache(char *name){
int index;
char *ip;
node *result=get(name);
if(!result)
return NULL;
index=result->cache_index;
ip=cache[index];
move_to_head(result);
return ip;
}
RingIterator<C> Ring<C>::sort_min(){
RingIterator<C> p1(*this);
RingIterator<C> p2(*this);
++p1;++p1;
++p2;
while ( ! p1.atOrigin() ){
if( ( (*p1) < (*p2) ) || ((*p1).V_x == (*p2).V_x ) ){
p2 = p1;
++p1;
move_to_head(p2) ;
}else{
++p1;
}
}
return p2;
}
void *
util_cache_get(struct util_cache *cache,
const void *key)
{
struct util_cache_entry *entry;
uint32_t hash = cache->hash(key);
assert(cache);
if (!cache)
return NULL;
entry = util_cache_entry_get(cache, hash, key);
if (!entry)
return NULL;
if (entry->state == FILLED)
move_to_head(&cache->lru, entry);
return entry->value;
}
static void
llvm_middle_end_prepare_gs(struct llvm_middle_end *fpme)
{
struct draw_context *draw = fpme->draw;
struct draw_geometry_shader *gs = draw->gs.geometry_shader;
struct draw_gs_llvm_variant_key *key;
struct draw_gs_llvm_variant *variant = NULL;
struct draw_gs_llvm_variant_list_item *li;
struct llvm_geometry_shader *shader = llvm_geometry_shader(gs);
char store[DRAW_GS_LLVM_MAX_VARIANT_KEY_SIZE];
unsigned i;
key = draw_gs_llvm_make_variant_key(fpme->llvm, store);
/* Search shader's list of variants for the key */
li = first_elem(&shader->variants);
while (!at_end(&shader->variants, li)) {
if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
variant = li->base;
break;
}
li = next_elem(li);
}
if (variant) {
/* found the variant, move to head of global list (for LRU) */
move_to_head(&fpme->llvm->gs_variants_list,
&variant->list_item_global);
}
else {
/* Need to create new variant */
/* First check if we've created too many variants. If so, free
* 25% of the LRU to avoid using too much memory.
*/
if (fpme->llvm->nr_gs_variants >= DRAW_MAX_SHADER_VARIANTS) {
/*
* XXX: should we flush here ?
*/
for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) {
struct draw_gs_llvm_variant_list_item *item;
if (is_empty_list(&fpme->llvm->gs_variants_list)) {
break;
}
item = last_elem(&fpme->llvm->gs_variants_list);
assert(item);
assert(item->base);
draw_gs_llvm_destroy_variant(item->base);
}
}
variant = draw_gs_llvm_create_variant(fpme->llvm, gs->info.num_outputs, key);
if (variant) {
insert_at_head(&shader->variants, &variant->list_item_local);
insert_at_head(&fpme->llvm->gs_variants_list,
&variant->list_item_global);
fpme->llvm->nr_gs_variants++;
shader->variants_cached++;
}
}
gs->current_variant = variant;
}
/**
* Prepare/validate middle part of the vertex pipeline.
* NOTE: if you change this function, also look at the non-LLVM
* function fetch_pipeline_prepare() for similar changes.
*/
static void
llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
unsigned in_prim,
unsigned opt,
unsigned *max_vertices )
{
struct llvm_middle_end *fpme = llvm_middle_end(middle);
struct draw_context *draw = fpme->draw;
struct draw_vertex_shader *vs = draw->vs.vertex_shader;
struct draw_geometry_shader *gs = draw->gs.geometry_shader;
const unsigned out_prim = gs ? gs->output_primitive :
u_assembled_prim(in_prim);
unsigned point_clip = draw->rasterizer->fill_front == PIPE_POLYGON_MODE_POINT ||
out_prim == PIPE_PRIM_POINTS;
unsigned nr;
fpme->input_prim = in_prim;
fpme->opt = opt;
draw_pt_post_vs_prepare( fpme->post_vs,
draw->clip_xy,
draw->clip_z,
draw->clip_user,
point_clip ? draw->guard_band_points_xy :
draw->guard_band_xy,
draw->identity_viewport,
draw->rasterizer->clip_halfz,
(draw->vs.edgeflag_output ? TRUE : FALSE) );
draw_pt_so_emit_prepare( fpme->so_emit, gs == NULL );
if (!(opt & PT_PIPELINE)) {
draw_pt_emit_prepare( fpme->emit,
out_prim,
max_vertices );
*max_vertices = MAX2( *max_vertices, 4096 );
}
else {
/* limit max fetches by limiting max_vertices */
*max_vertices = 4096;
}
/* Get the number of float[4] attributes per vertex.
* Note: this must be done after draw_pt_emit_prepare() since that
* can effect the vertex size.
*/
nr = MAX2(vs->info.num_inputs, draw_total_vs_outputs(draw));
/* Always leave room for the vertex header whether we need it or
* not. It's hard to get rid of it in particular because of the
* viewport code in draw_pt_post_vs.c.
*/
fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
/* Get the number of float[4] attributes per vertex.
* Note: this must be done after draw_pt_emit_prepare() since that
* can effect the vertex size.
*/
nr = MAX2(vs->info.num_inputs, draw_total_vs_outputs(draw));
/* Always leave room for the vertex header whether we need it or
* not. It's hard to get rid of it in particular because of the
* viewport code in draw_pt_post_vs.c.
*/
fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
/* return even number */
*max_vertices = *max_vertices & ~1;
/* Find/create the vertex shader variant */
{
struct draw_llvm_variant_key *key;
struct draw_llvm_variant *variant = NULL;
struct draw_llvm_variant_list_item *li;
struct llvm_vertex_shader *shader = llvm_vertex_shader(vs);
char store[DRAW_LLVM_MAX_VARIANT_KEY_SIZE];
unsigned i;
key = draw_llvm_make_variant_key(fpme->llvm, store);
/* Search shader's list of variants for the key */
li = first_elem(&shader->variants);
while (!at_end(&shader->variants, li)) {
if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
variant = li->base;
break;
}
li = next_elem(li);
}
if (variant) {
/* found the variant, move to head of global list (for LRU) */
move_to_head(&fpme->llvm->vs_variants_list,
&variant->list_item_global);
}
else {
/* Need to create new variant */
/* First check if we've created too many variants. If so, free
* 25% of the LRU to avoid using too much memory.
*/
if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
/*
* XXX: should we flush here ?
*/
for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) {
struct draw_llvm_variant_list_item *item;
if (is_empty_list(&fpme->llvm->vs_variants_list)) {
break;
}
item = last_elem(&fpme->llvm->vs_variants_list);
assert(item);
assert(item->base);
draw_llvm_destroy_variant(item->base);
}
}
variant = draw_llvm_create_variant(fpme->llvm, nr, key);
if (variant) {
insert_at_head(&shader->variants, &variant->list_item_local);
insert_at_head(&fpme->llvm->vs_variants_list,
&variant->list_item_global);
fpme->llvm->nr_variants++;
shader->variants_cached++;
}
}
fpme->current_variant = variant;
}
if (gs) {
llvm_middle_end_prepare_gs(fpme);
}
}
static void
llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
unsigned in_prim,
unsigned opt,
unsigned *max_vertices )
{
struct llvm_middle_end *fpme = (struct llvm_middle_end *)middle;
struct draw_context *draw = fpme->draw;
struct llvm_vertex_shader *shader =
llvm_vertex_shader(draw->vs.vertex_shader);
char store[DRAW_LLVM_MAX_VARIANT_KEY_SIZE];
struct draw_llvm_variant_key *key;
struct draw_llvm_variant *variant = NULL;
struct draw_llvm_variant_list_item *li;
unsigned i;
unsigned instance_id_index = ~0;
const unsigned out_prim = (draw->gs.geometry_shader ?
draw->gs.geometry_shader->output_primitive :
in_prim);
/* Add one to num_outputs because the pipeline occasionally tags on
* an additional texcoord, eg for AA lines.
*/
const unsigned nr = MAX2( shader->base.info.num_inputs,
shader->base.info.num_outputs + 1 );
/* Scan for instanceID system value.
* XXX but we never use instance_id_index?!
*/
for (i = 0; i < shader->base.info.num_inputs; i++) {
if (shader->base.info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
instance_id_index = i;
break;
}
}
fpme->input_prim = in_prim;
fpme->opt = opt;
/* Always leave room for the vertex header whether we need it or
* not. It's hard to get rid of it in particular because of the
* viewport code in draw_pt_post_vs.c.
*/
fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
/* XXX: it's not really gl rasterization rules we care about here,
* but gl vs dx9 clip spaces.
*/
draw_pt_post_vs_prepare( fpme->post_vs,
draw->clip_xy,
draw->clip_z,
draw->clip_user,
draw->identity_viewport,
(boolean)draw->rasterizer->gl_rasterization_rules,
(draw->vs.edgeflag_output ? TRUE : FALSE) );
draw_pt_so_emit_prepare( fpme->so_emit );
if (!(opt & PT_PIPELINE)) {
draw_pt_emit_prepare( fpme->emit,
out_prim,
max_vertices );
*max_vertices = MAX2( *max_vertices, 4096 );
}
else {
/* limit max fetches by limiting max_vertices */
*max_vertices = 4096;
}
/* return even number */
*max_vertices = *max_vertices & ~1;
key = draw_llvm_make_variant_key(fpme->llvm, store);
/* Search shader's list of variants for the key */
li = first_elem(&shader->variants);
while (!at_end(&shader->variants, li)) {
if (memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
variant = li->base;
break;
}
li = next_elem(li);
}
if (variant) {
/* found the variant, move to head of global list (for LRU) */
move_to_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
}
else {
/* Need to create new variant */
unsigned i;
/* First check if we've created too many variants. If so, free
* 25% of the LRU to avoid using too much memory.
*/
if (fpme->llvm->nr_variants >= DRAW_MAX_SHADER_VARIANTS) {
/*
* XXX: should we flush here ?
*/
for (i = 0; i < DRAW_MAX_SHADER_VARIANTS / 4; i++) {
struct draw_llvm_variant_list_item *item =
last_elem(&fpme->llvm->vs_variants_list);
draw_llvm_destroy_variant(item->base);
}
}
variant = draw_llvm_create_variant(fpme->llvm, nr, key);
if (variant) {
insert_at_head(&shader->variants, &variant->list_item_local);
insert_at_head(&fpme->llvm->vs_variants_list, &variant->list_item_global);
fpme->llvm->nr_variants++;
shader->variants_cached++;
}
}
fpme->current_variant = variant;
/*XXX we only support one constant buffer */
fpme->llvm->jit_context.vs_constants =
draw->pt.user.vs_constants[0];
fpme->llvm->jit_context.gs_constants =
draw->pt.user.gs_constants[0];
fpme->llvm->jit_context.planes =
(float (*) [12][4]) draw->pt.user.planes[0];
fpme->llvm->jit_context.viewport =
(float *)draw->viewport.scale;
}
void radeonUpdateTextureState( GLcontext *ctx )
{
radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
struct gl_texture_unit *texUnit = &ctx->Texture.Unit[0];
if ( texUnit->_ReallyEnabled & (TEXTURE_1D_BIT | TEXTURE_2D_BIT) ) {
struct gl_texture_object *tObj = texUnit->_Current;
radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
/* Upload teximages (not pipelined)
*/
if ( t->dirty_images ) {
RADEON_FIREVERTICES( rmesa );
radeonSetTexImages( rmesa, tObj );
}
/* Update state if this is a different texture object to last
* time.
*/
if ( rmesa->state.texture.unit[0].texobj != t ) {
rmesa->state.texture.unit[0].texobj = t;
t->dirty_state |= 1<<0;
move_to_head( &rmesa->texture.objects[0], t );
}
if (t->dirty_state) {
GLuint *cmd = RADEON_DB_STATE( tex[0] );
cmd[TEX_PP_TXFILTER] &= ~TEXOBJ_TXFILTER_MASK;
cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
cmd[TEX_PP_TXFILTER] |= t->pp_txfilter & TEXOBJ_TXFILTER_MASK;
cmd[TEX_PP_TXFORMAT] |= t->pp_txformat & TEXOBJ_TXFORMAT_MASK;
cmd[TEX_PP_TXOFFSET] = t->pp_txoffset;
cmd[TEX_PP_BORDER_COLOR] = t->pp_border_color;
RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.tex[0] );
t->dirty_state = 0;
}
/* Newly enabled?
*/
if (!(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & RADEON_TEX_0_ENABLE)) {
RADEON_STATECHANGE( rmesa, ctx );
rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= (RADEON_TEX_0_ENABLE |
RADEON_TEX_BLEND_0_ENABLE);
RADEON_STATECHANGE( rmesa, tcl );
rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_ST0;
}
radeonUpdateTextureEnv( ctx, 0 );
}
else if (rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (RADEON_TEX_0_ENABLE<<0)) {
/* Texture unit disabled */
rmesa->state.texture.unit[0].texobj = 0;
RADEON_STATECHANGE( rmesa, ctx );
rmesa->hw.ctx.cmd[CTX_PP_CNTL] &=
~((RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << 0);
RADEON_STATECHANGE( rmesa, tcl );
rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~(RADEON_TCL_VTX_ST0 |
RADEON_TCL_VTX_Q0);
}
}
/**
* \brief Upload texture images.
*
* This might require removing our own and/or other client's texture objects to
* make room for these images.
*
* \param rmesa Radeon context.
* \param tObj texture object to upload.
*
* Sets the matching hardware texture format. Calculates which mipmap levels to
* send, depending of the base image size, GL_TEXTURE_MIN_LOD,
* GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL and the
* Radeon offset rules. Kicks out textures until the requested texture fits,
* sets the texture hardware state and, while holding the hardware lock,
* uploads any images that are new.
*/
static void radeonSetTexImages( radeonContextPtr rmesa,
struct gl_texture_object *tObj )
{
radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData;
const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
GLint totalSize;
GLint texelsPerDword = 0, blitWidth = 0, blitPitch = 0;
GLint x, y, width, height;
GLint i;
GLint firstLevel, lastLevel, numLevels;
GLint log2Width, log2Height;
GLuint txformat = 0;
/* This code cannot be reached once we have lost focus
*/
assert(rmesa->radeonScreen->buffers);
/* Set the hardware texture format
*/
switch (baseImage->TexFormat->MesaFormat) {
case MESA_FORMAT_I8:
txformat = RADEON_TXFORMAT_I8;
texelsPerDword = 4;
blitPitch = 64;
break;
case MESA_FORMAT_RGBA8888:
txformat = RADEON_TXFORMAT_RGBA8888 | RADEON_TXFORMAT_ALPHA_IN_MAP;
texelsPerDword = 1;
blitPitch = 16;
break;
case MESA_FORMAT_RGB565:
txformat = RADEON_TXFORMAT_RGB565;
texelsPerDword = 2;
blitPitch = 32;
break;
default:
_mesa_problem(NULL, "unexpected texture format in radeonTexImage2D");
return;
}
t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
RADEON_TXFORMAT_ALPHA_IN_MAP);
t->pp_txformat |= txformat;
/* Select the larger of the two widths for our global texture image
* coordinate space. As the Radeon has very strict offset rules, we
* can't upload mipmaps directly and have to reference their location
* from the aligned start of the whole image.
*/
blitWidth = MAX2( baseImage->Width, blitPitch );
/* Calculate mipmap offsets and dimensions.
*/
totalSize = 0;
x = 0;
y = 0;
/* Compute which mipmap levels we really want to send to the hardware.
* This depends on the base image size, GL_TEXTURE_MIN_LOD,
* GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL.
* Yes, this looks overly complicated, but it's all needed.
*/
firstLevel = tObj->BaseLevel + (GLint) (tObj->MinLod + 0.5);
firstLevel = MAX2(firstLevel, tObj->BaseLevel);
lastLevel = tObj->BaseLevel + (GLint) (tObj->MaxLod + 0.5);
lastLevel = MAX2(lastLevel, tObj->BaseLevel);
lastLevel = MIN2(lastLevel, tObj->BaseLevel + baseImage->MaxLog2);
lastLevel = MIN2(lastLevel, tObj->MaxLevel);
lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */
/* save these values */
t->firstLevel = firstLevel;
t->lastLevel = lastLevel;
numLevels = lastLevel - firstLevel + 1;
log2Width = tObj->Image[0][firstLevel]->WidthLog2;
log2Height = tObj->Image[0][firstLevel]->HeightLog2;
for ( i = 0 ; i < numLevels ; i++ ) {
const struct gl_texture_image *texImage = tObj->Image[0][i + firstLevel];
if ( !texImage )
break;
width = texImage->Width;
height = texImage->Height;
/* Texture images have a minimum pitch of 32 bytes (half of the
* 64-byte minimum pitch for blits). For images that have a
* width smaller than this, we must pad each texture image
* scanline out to this amount.
*/
if ( width < blitPitch / 2 ) {
width = blitPitch / 2;
}
totalSize += width * height * baseImage->TexFormat->TexelBytes;
ASSERT( (totalSize & 31) == 0 );
while ( width < blitWidth && height > 1 ) {
width *= 2;
height /= 2;
}
ASSERT(i < RADEON_MAX_TEXTURE_LEVELS);
t->image[i].x = x;
t->image[i].y = y;
t->image[i].width = width;
t->image[i].height = height;
/* While blits must have a pitch of at least 64 bytes, mipmaps
* must be aligned on a 32-byte boundary (just like each texture
* image scanline).
*/
if ( width >= blitWidth ) {
y += height;
} else {
x += width;
if ( x >= blitWidth ) {
x = 0;
y++;
}
}
}
/* Align the total size of texture memory block.
*/
t->totalSize = (totalSize + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
/* Hardware state:
*/
t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK;
t->pp_txfilter |= (numLevels - 1) << RADEON_MAX_MIP_LEVEL_SHIFT;
t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK |
RADEON_TXFORMAT_HEIGHT_MASK);
t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) |
(log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT));
t->dirty_state = TEX_ALL;
/* Update the local texture LRU.
*/
move_to_head( &rmesa->texture.objects[0], t );
LOCK_HARDWARE( rmesa );
/* Kick out textures until the requested texture fits */
while ( !t->memBlock ) {
t->memBlock = mmAllocMem( rmesa->texture.heap[0], t->totalSize, 12, 0);
if (!t->memBlock)
radeonSwapOutTexObj( rmesa, rmesa->texture.objects[0].prev );
}
/* Set the base offset of the texture image */
t->bufAddr = rmesa->radeonScreen->texOffset[0] + t->memBlock->ofs;
t->pp_txoffset = t->bufAddr;
/* Upload any images that are new
*/
for ( i = 0 ; i < numLevels ; i++ ) {
if ( t->dirty_images & (1 << i) ) {
radeonUploadSubImage( rmesa, t, i, 0, 0,
t->image[i].width, t->image[i].height );
}
}
rmesa->texture.age[0] = ++rmesa->sarea->texAge[0];
UNLOCK_HARDWARE( rmesa );
t->dirty_images = 0;
}