eliminate fallback key/hash pair cache in texture cache layer; we can rely on rewinding purely via the hash store layer

This commit is contained in:
Ryan Fleury
2024-03-26 11:50:44 -07:00
parent 4027338a2f
commit 7cf8da0b43
4 changed files with 46 additions and 115 deletions
+13 -3
View File
@@ -877,6 +877,17 @@ DF_GFX_VIEW_RULE_BLOCK_UI_FUNCTION_DEF(bitmap)
U64 expected_size = topology_info.width*topology_info.height*r_tex2d_format_bytes_per_pixel_table[topology_info.fmt];
Rng1U64 vaddr_range = r1u64(base_vaddr, base_vaddr+expected_size);
//- rjf: obtain key for this data range
U128 texture_key = ctrl_hash_store_key_from_process_vaddr_range(process->ctrl_machine_id, process->ctrl_handle, vaddr_range, 0);
// HACK(rjf): we should not need to explicitly inform the process memory cache layer here -
// want a joined "get me the key & load it" operation here
ctrl_stored_hash_from_process_vaddr_range(process->ctrl_machine_id, process->ctrl_handle, vaddr_range, 0, 0, 0);
//- rjf: hash & topology -> texture
TEX_Topology topology = tex_topology_make(v2s32((S32)topology_info.width, (S32)topology_info.height), topology_info.fmt);
R_Handle texture = tex_texture_from_key_topology(tex_scope, texture_key, topology);
#if 0
//- rjf: unique identifying info about this address -> unique key
U128 texture_key = {0};
{
@@ -892,10 +903,8 @@ DF_GFX_VIEW_RULE_BLOCK_UI_FUNCTION_DEF(bitmap)
//- rjf: address range -> hash
U128 hash = ctrl_stored_hash_from_process_vaddr_range(process->ctrl_machine_id, process->ctrl_handle, vaddr_range, 0, 0, 0);
//- rjf: hash & topology -> texture
TEX_Topology topology = tex_topology_make(v2s32((S32)topology_info.width, (S32)topology_info.height), topology_info.fmt);
R_Handle texture = tex_texture_from_key_hash_topology(tex_scope, texture_key, hash, topology);
#endif
//- rjf: build preview
F32 rate = 1 - pow_f32(2, (-15.f * df_dt()));
@@ -941,6 +950,7 @@ DF_GFX_VIEW_RULE_BLOCK_UI_FUNCTION_DEF(bitmap)
{
if(dim.y > (F32)topology_info.height)
{
U128 hash = hs_hash_from_key(texture_key, 0);
String8 data = hs_data_from_hash(hs_scope, hash);
U64 bytes_per_pixel = r_tex2d_format_bytes_per_pixel_table[topology.fmt];
U64 mouse_pixel_off = mouse_bitmap_px_off.y*topology_info.width + mouse_bitmap_px_off.x;
+8 -6
View File
@@ -5024,7 +5024,8 @@ DF_VIEW_UI_FUNCTION_DEF(Code)
{
ProfBeginFunction();
Temp scratch = scratch_begin(0, 0);
DBGI_Scope *scope = dbgi_scope_open();
DBGI_Scope *dbgi_scope = dbgi_scope_open();
TXT_Scope *txt_scope = txt_scope_open();
DF_CodeViewState *tv = df_view_user_state(view, DF_CodeViewState);
//////////////////////////////
@@ -5050,10 +5051,10 @@ DF_VIEW_UI_FUNCTION_DEF(Code)
U64 unwind_count = ctrl_ctx.unwind_count;
U64 rip_vaddr = df_query_cached_rip_from_thread_unwind(thread, unwind_count);
DF_Entity *process = df_entity_ancestor_from_kind(thread, DF_EntityKind_Process);
EVAL_ParseCtx parse_ctx = df_eval_parse_ctx_from_process_vaddr(scope, process, rip_vaddr);
EVAL_ParseCtx parse_ctx = df_eval_parse_ctx_from_process_vaddr(dbgi_scope, process, rip_vaddr);
//////////////////////////////
//- rjf: unpack entity info
//- rjf: unpack file/text entity info
//
B32 entity_is_missing = !!(entity->flags & DF_EntityFlag_IsMissing && !(entity->flags & DF_EntityFlag_Output));
TXTI_Handle txti_handle = df_txti_handle_from_entity(entity);
@@ -5467,7 +5468,7 @@ DF_VIEW_UI_FUNCTION_DEF(Code)
String8 expr = txti_string_from_handle_txt_rng(scratch.arena, txti_handle, expr_rng);
if(expr.size != 0)
{
DF_Eval eval = df_eval_from_string(scratch.arena, scope, &ctrl_ctx, &parse_ctx, &eval_string2expr_map_nil, expr);
DF_Eval eval = df_eval_from_string(scratch.arena, dbgi_scope, &ctrl_ctx, &parse_ctx, &eval_string2expr_map_nil, expr);
if(eval.mode != EVAL_EvalMode_NULL)
{
df_set_hover_eval(ws, sig.mouse_expr_baseline_pos, ctrl_ctx, entity, sig.mouse_pt, 0, expr);
@@ -5788,7 +5789,7 @@ DF_VIEW_UI_FUNCTION_DEF(Code)
String8 full_path = df_full_path_from_entity(scratch.arena, entity);
TXTI_Handle handle = txti_handle_from_path(full_path);
TXTI_BufferInfo info = txti_buffer_info_from_handle(scratch.arena, handle);
DBGI_Parse *parse = df_dbgi_parse_from_binary_file(scope, binary);
DBGI_Parse *parse = df_dbgi_parse_from_binary_file(dbgi_scope, binary);
if(parse->exe_props.modified < info.timestamp)
{
file_is_out_of_date = 1;
@@ -5851,7 +5852,8 @@ DF_VIEW_UI_FUNCTION_DEF(Code)
}
}
dbgi_scope_close(scope);
txt_scope_close(txt_scope);
dbgi_scope_close(dbgi_scope);
scratch_end(scratch);
ProfEnd();
}
+21 -82
View File
@@ -34,16 +34,6 @@ tex_init(void)
tex_shared->stripes[idx].rw_mutex = os_rw_mutex_alloc();
tex_shared->stripes[idx].cv = os_condition_variable_alloc();
}
tex_shared->fallback_slots_count = 1024;
tex_shared->fallback_stripes_count = Min(tex_shared->fallback_slots_count, os_logical_core_count());
tex_shared->fallback_slots = push_array(arena, TEX_KeyFallbackSlot, tex_shared->fallback_slots_count);
tex_shared->fallback_stripes = push_array(arena, TEX_Stripe, tex_shared->fallback_stripes_count);
for(U64 idx = 0; idx < tex_shared->fallback_stripes_count; idx += 1)
{
tex_shared->fallback_stripes[idx].arena = arena_alloc();
tex_shared->fallback_stripes[idx].rw_mutex = os_rw_mutex_alloc();
tex_shared->fallback_stripes[idx].cv = os_condition_variable_alloc();
}
tex_shared->u2x_ring_size = KB(64);
tex_shared->u2x_ring_base = push_array_no_zero(arena, U8, tex_shared->u2x_ring_size);
tex_shared->u2x_ring_cv = os_condition_variable_alloc();
@@ -158,10 +148,9 @@ tex_scope_touch_node__stripe_r_guarded(TEX_Scope *scope, TEX_Node *node)
//~ rjf: Cache Lookups
internal R_Handle
tex_texture_from_key_hash_topology(TEX_Scope *scope, U128 key, U128 hash, TEX_Topology topology)
tex_texture_from_hash_topology(TEX_Scope *scope, U128 hash, TEX_Topology topology)
{
R_Handle handle = {0};
if(!u128_match(u128_zero(), hash))
{
U64 slot_idx = hash.u64[1]%tex_shared->slots_count;
U64 stripe_idx = slot_idx%tex_shared->stripes_count;
@@ -217,42 +206,23 @@ tex_texture_from_key_hash_topology(TEX_Scope *scope, U128 key, U128 hash, TEX_To
}
if(node_is_new)
{
tex_u2x_enqueue_req(key, hash, topology, max_U64);
tex_u2x_enqueue_req(hash, topology, max_U64);
}
if(r_handle_match(handle, r_handle_zero()))
}
return handle;
}
internal R_Handle
tex_texture_from_key_topology(TEX_Scope *scope, U128 key, TEX_Topology topology)
{
R_Handle handle = {0};
for(U64 rewind_idx = 0; rewind_idx < 2; rewind_idx += 1)
{
U128 hash = hs_hash_from_key(key, rewind_idx);
handle = tex_texture_from_hash_topology(scope, hash, topology);
if(!r_handle_match(handle, r_handle_zero()))
{
U128 fallback_hash = {0};
U64 fallback_slot_idx = key.u64[1]%tex_shared->fallback_slots_count;
U64 fallback_stripe_idx = fallback_slot_idx%tex_shared->fallback_stripes_count;
TEX_KeyFallbackSlot *fallback_slot = &tex_shared->fallback_slots[fallback_slot_idx];
TEX_Stripe *fallback_stripe = &tex_shared->fallback_stripes[fallback_stripe_idx];
OS_MutexScopeR(fallback_stripe->rw_mutex) for(TEX_KeyFallbackNode *n = fallback_slot->first; n != 0; n = n->next)
{
if(u128_match(key, n->key))
{
fallback_hash = n->hash;
break;
}
}
if(!u128_match(fallback_hash, u128_zero()))
{
U64 retry_slot_idx = fallback_hash.u64[1]%tex_shared->slots_count;
U64 retry_stripe_idx = retry_slot_idx%tex_shared->stripes_count;
TEX_Slot *retry_slot = &tex_shared->slots[retry_slot_idx];
TEX_Stripe *retry_stripe = &tex_shared->stripes[retry_stripe_idx];
OS_MutexScopeR(retry_stripe->rw_mutex)
{
for(TEX_Node *n = retry_slot->first; n != 0; n = n->next)
{
if(u128_match(fallback_hash, n->hash) && MemoryMatchStruct(&topology, &n->topology))
{
handle = n->texture;
tex_scope_touch_node__stripe_r_guarded(scope, n);
break;
}
}
}
}
break;
}
}
return handle;
@@ -262,17 +232,16 @@ tex_texture_from_key_hash_topology(TEX_Scope *scope, U128 key, U128 hash, TEX_To
//~ rjf: Transfer Threads
internal B32
tex_u2x_enqueue_req(U128 key, U128 hash, TEX_Topology top, U64 endt_us)
tex_u2x_enqueue_req(U128 hash, TEX_Topology top, U64 endt_us)
{
B32 good = 0;
OS_MutexScope(tex_shared->u2x_ring_mutex) for(;;)
{
U64 unconsumed_size = tex_shared->u2x_ring_write_pos-tex_shared->u2x_ring_read_pos;
U64 available_size = tex_shared->u2x_ring_size-unconsumed_size;
if(available_size >= sizeof(key)+sizeof(hash)+sizeof(top))
if(available_size >= sizeof(hash)+sizeof(top))
{
good = 1;
tex_shared->u2x_ring_write_pos += ring_write_struct(tex_shared->u2x_ring_base, tex_shared->u2x_ring_size, tex_shared->u2x_ring_write_pos, &key);
tex_shared->u2x_ring_write_pos += ring_write_struct(tex_shared->u2x_ring_base, tex_shared->u2x_ring_size, tex_shared->u2x_ring_write_pos, &hash);
tex_shared->u2x_ring_write_pos += ring_write_struct(tex_shared->u2x_ring_base, tex_shared->u2x_ring_size, tex_shared->u2x_ring_write_pos, &top);
break;
@@ -291,14 +260,13 @@ tex_u2x_enqueue_req(U128 key, U128 hash, TEX_Topology top, U64 endt_us)
}
internal void
tex_u2x_dequeue_req(U128 *key_out, U128 *hash_out, TEX_Topology *top_out)
tex_u2x_dequeue_req(U128 *hash_out, TEX_Topology *top_out)
{
OS_MutexScope(tex_shared->u2x_ring_mutex) for(;;)
{
U64 unconsumed_size = tex_shared->u2x_ring_write_pos-tex_shared->u2x_ring_read_pos;
if(unconsumed_size >= sizeof(*key_out)+sizeof(*hash_out)+sizeof(*top_out))
if(unconsumed_size >= sizeof(*hash_out)+sizeof(*top_out))
{
tex_shared->u2x_ring_read_pos += ring_read_struct(tex_shared->u2x_ring_base, tex_shared->u2x_ring_size, tex_shared->u2x_ring_read_pos, key_out);
tex_shared->u2x_ring_read_pos += ring_read_struct(tex_shared->u2x_ring_base, tex_shared->u2x_ring_size, tex_shared->u2x_ring_read_pos, hash_out);
tex_shared->u2x_ring_read_pos += ring_read_struct(tex_shared->u2x_ring_base, tex_shared->u2x_ring_size, tex_shared->u2x_ring_read_pos, top_out);
break;
@@ -316,10 +284,9 @@ tex_xfer_thread__entry_point(void *p)
HS_Scope *scope = hs_scope_open();
//- rjf: decode
U128 key = {0};
U128 hash = {0};
TEX_Topology top = {0};
tex_u2x_dequeue_req(&key, &hash, &top);
tex_u2x_dequeue_req(&hash, &top);
//- rjf: unpack hash
U64 slot_idx = hash.u64[1]%tex_shared->slots_count;
@@ -370,34 +337,6 @@ tex_xfer_thread__entry_point(void *p)
}
}
//- rjf: commit this key/hash pair to fallback cache
if(got_task && !u128_match(key, u128_zero()) && !u128_match(hash, u128_zero()))
{
U64 fallback_slot_idx = key.u64[1]%tex_shared->fallback_slots_count;
U64 fallback_stripe_idx = fallback_slot_idx%tex_shared->fallback_stripes_count;
TEX_KeyFallbackSlot *fallback_slot = &tex_shared->fallback_slots[fallback_slot_idx];
TEX_Stripe *fallback_stripe = &tex_shared->fallback_stripes[fallback_stripe_idx];
OS_MutexScopeW(fallback_stripe->rw_mutex)
{
TEX_KeyFallbackNode *node = 0;
for(TEX_KeyFallbackNode *n = fallback_slot->first; n != 0; n = n->next)
{
if(u128_match(n->key, key))
{
node = n;
break;
}
}
if(node == 0)
{
node = push_array(fallback_stripe->arena, TEX_KeyFallbackNode, 1);
SLLQueuePush(fallback_slot->first, fallback_slot->last, node);
}
node->key = key;
node->hash = hash;
}
}
hs_scope_close(scope);
}
}
+4 -24
View File
@@ -17,21 +17,6 @@ struct TEX_Topology
////////////////////////////////
//~ rjf: Cache Types
typedef struct TEX_KeyFallbackNode TEX_KeyFallbackNode;
struct TEX_KeyFallbackNode
{
TEX_KeyFallbackNode *next;
U128 key;
U128 hash;
};
typedef struct TEX_KeyFallbackSlot TEX_KeyFallbackSlot;
struct TEX_KeyFallbackSlot
{
TEX_KeyFallbackNode *first;
TEX_KeyFallbackNode *last;
};
typedef struct TEX_Node TEX_Node;
struct TEX_Node
{
@@ -109,12 +94,6 @@ struct TEX_Shared
TEX_Stripe *stripes;
TEX_Node **stripes_free_nodes;
// rjf: fallback cache
U64 fallback_slots_count;
U64 fallback_stripes_count;
TEX_KeyFallbackSlot *fallback_slots;
TEX_Stripe *fallback_stripes;
// rjf: user -> xfer thread
U64 u2x_ring_size;
U8 *u2x_ring_base;
@@ -168,13 +147,14 @@ internal void tex_scope_touch_node__stripe_r_guarded(TEX_Scope *scope, TEX_Node
////////////////////////////////
//~ rjf: Cache Lookups
internal R_Handle tex_texture_from_key_hash_topology(TEX_Scope *scope, U128 key, U128 hash, TEX_Topology topology);
internal R_Handle tex_texture_from_hash_topology(TEX_Scope *scope, U128 hash, TEX_Topology topology);
internal R_Handle tex_texture_from_key_topology(TEX_Scope *scope, U128 key, TEX_Topology topology);
////////////////////////////////
//~ rjf: Transfer Threads
internal B32 tex_u2x_enqueue_req(U128 key, U128 hash, TEX_Topology top, U64 endt_us);
internal void tex_u2x_dequeue_req(U128 *key_out, U128 *hash_out, TEX_Topology *top_out);
internal B32 tex_u2x_enqueue_req(U128 hash, TEX_Topology top, U64 endt_us);
internal void tex_u2x_dequeue_req(U128 *hash_out, TEX_Topology *top_out);
internal void tex_xfer_thread__entry_point(void *p);
////////////////////////////////