remove dispatch_state

This commit is contained in:
jacob 2025-06-17 18:46:43 -05:00
parent 894044669c
commit 9aad25a699
4 changed files with 129 additions and 207 deletions

View File

@ -126,9 +126,7 @@ struct gpu_dispatch_params {
struct xform draw_target_view;
};
struct gpu_handle gpu_dispatch_state_alloc(void);
void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_params params);
void gpu_dispatch(struct gpu_dispatch_params params);
/* ========================== *
* Present

View File

@ -79,7 +79,6 @@ enum dx11_handle_kind {
DX11_HANDLE_KIND_NONE,
DX11_HANDLE_KIND_TEXTURE,
DX11_HANDLE_KIND_PLAN,
DX11_HANDLE_KIND_DISPATCH_STATE,
NUM_DX11_HANDLE_KINDS
};
@ -96,15 +95,6 @@ struct dx11_shader {
ID3D11PixelShader *ps;
};
struct dx11_dispatch_state {
struct dx11_handle_header header;
struct dx11_texture *albedo;
struct dx11_buffer *constant_buffer;
struct dx11_dispatch_state *next_free;
};
struct dx11_buffer {
D3D11_BUFFER_DESC desc;
@ -194,7 +184,9 @@ struct dx11_plan {
struct {
struct dx11_buffer *instance_buffer;
} test;
} buffers;
} cmd_buffers;
struct dx11_buffer *constant_buffer;
struct dx11_plan *next_free;
};
@ -243,9 +235,6 @@ GLOBAL struct {
struct dx11_texture backbuffer_texture;
i64 last_backbuffer_resize_ns;
struct gpu_handle present_blit_plan;
struct gpu_handle present_blit_dispatch_state;
ID3D11BlendState *blend_state;
ID3D11RasterizerState *rasterizer_state;
ID3D11DepthStencilState *depth_stencil_state;
@ -279,6 +268,9 @@ GLOBAL struct {
struct dx11_buffer *dummy_vertex_buffer;
struct dx11_buffer *quad_index_buffer;
/* Blit plan */
struct gpu_handle present_blit_plan;
} G = ZI, DEBUG_ALIAS(G, G_gpu_dx11);
/* ========================== *
@ -548,7 +540,6 @@ struct gpu_startup_receipt gpu_startup(struct work_startup_receipt *work_sr, str
#endif
G.present_blit_plan = gpu_plan_alloc();
G.present_blit_dispatch_state = gpu_dispatch_state_alloc();
return (struct gpu_startup_receipt) { 0 };
}
@ -587,12 +578,6 @@ void gpu_release(struct gpu_handle handle)
/* TODO */
ASSERT(false);
} break;
case DX11_HANDLE_KIND_DISPATCH_STATE:
{
/* TODO */
ASSERT(false);
} break;
}
}
@ -1354,29 +1339,29 @@ struct gpu_handle gpu_plan_alloc(void)
idesc.BindFlags = D3D11_BIND_INDEX_BUFFER;
idesc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
plan->buffers.mesh.vertex_buffer = dx11_buffer_alloc(vdesc, NULL);
plan->buffers.mesh.index_buffer = dx11_buffer_alloc(idesc, NULL);
plan->cmd_buffers.mesh.vertex_buffer = dx11_buffer_alloc(vdesc, NULL);
plan->cmd_buffers.mesh.index_buffer = dx11_buffer_alloc(idesc, NULL);
}
/* Texture buffers */
{
struct D3D11_BUFFER_DESC desc = structured_buffer_desc;
desc.StructureByteStride = sizeof(struct dx11_texture_instance);
plan->buffers.texture.instance_buffer = dx11_buffer_alloc(desc, NULL);
plan->cmd_buffers.texture.instance_buffer = dx11_buffer_alloc(desc, NULL);
}
/* Grid buffers */
{
struct D3D11_BUFFER_DESC desc = structured_buffer_desc;
desc.StructureByteStride = sizeof(struct dx11_grid_instance);
plan->buffers.grid.instance_buffer = dx11_buffer_alloc(desc, NULL);
plan->cmd_buffers.grid.instance_buffer = dx11_buffer_alloc(desc, NULL);
}
/* Test buffers */
{
struct D3D11_BUFFER_DESC desc = structured_buffer_desc;
desc.StructureByteStride = sizeof(struct dx11_test_instance);
plan->buffers.test.instance_buffer = dx11_buffer_alloc(desc, NULL);
plan->cmd_buffers.test.instance_buffer = dx11_buffer_alloc(desc, NULL);
}
}
@ -1410,8 +1395,8 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
/* TODO: Better count method */
cmd = arena_push(&plan->cpu_cmds_arena, struct dx11_cmd);
cmd->kind = params.kind;
cmd->mesh.vertex_offset = (plan->buffers.mesh.vertex_buffer->cpu_buffer_arena.pos / sizeof(struct dx11_mesh_vertex));
cmd->mesh.index_offset = (plan->buffers.mesh.index_buffer->cpu_buffer_arena.pos / sizeof(u32));
cmd->mesh.vertex_offset = (plan->cmd_buffers.mesh.vertex_buffer->cpu_buffer_arena.pos / sizeof(struct dx11_mesh_vertex));
cmd->mesh.index_offset = (plan->cmd_buffers.mesh.index_buffer->cpu_buffer_arena.pos / sizeof(u32));
if (plan->cpu_last_cmd) {
plan->cpu_last_cmd->next = cmd;
} else {
@ -1424,7 +1409,7 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
u64 vertex_count = params.mesh.vertices.count;
u64 start_index = cmd->mesh.vertex_count;
cmd->mesh.vertex_count += vertex_count;
struct dx11_mesh_vertex *verts = dx11_buffer_push(plan->buffers.mesh.vertex_buffer, sizeof(struct dx11_mesh_vertex) * vertex_count);
struct dx11_mesh_vertex *verts = dx11_buffer_push(plan->cmd_buffers.mesh.vertex_buffer, sizeof(struct dx11_mesh_vertex) * vertex_count);
for (u64 i = 0; i < vertex_count; ++i) {
struct dx11_mesh_vertex *vert = &verts[i];
vert->pos = params.mesh.vertices.points[i];
@ -1433,7 +1418,7 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
/* Push indices */
u64 index_count = params.mesh.indices.count;
u32 *indices = dx11_buffer_push(plan->buffers.mesh.index_buffer, sizeof(u32) * index_count);
u32 *indices = dx11_buffer_push(plan->cmd_buffers.mesh.index_buffer, sizeof(u32) * index_count);
cmd->mesh.index_count += index_count;
for (u64 i = 0; i < index_count; ++i) {
indices[i] = start_index + params.mesh.indices.indices[i];
@ -1458,7 +1443,7 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
cmd->kind = params.kind;
cmd->texture.sprite = params.texture.sprite;
cmd->texture.texture = params.texture.texture;
cmd->texture.instance_offset = (plan->buffers.texture.instance_buffer->cpu_buffer_arena.pos / sizeof(struct dx11_texture_instance));
cmd->texture.instance_offset = (plan->cmd_buffers.texture.instance_buffer->cpu_buffer_arena.pos / sizeof(struct dx11_texture_instance));
if (plan->cpu_last_cmd) {
plan->cpu_last_cmd->next = cmd;
} else {
@ -1469,7 +1454,7 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
/* Push instance data */
++cmd->texture.instance_count;
struct dx11_texture_instance *instance = dx11_buffer_push(plan->buffers.texture.instance_buffer, sizeof(struct dx11_texture_instance));
struct dx11_texture_instance *instance = dx11_buffer_push(plan->cmd_buffers.texture.instance_buffer, sizeof(struct dx11_texture_instance));
instance->xf = params.texture.xf;
instance->uv0 = params.texture.clip.p0;
instance->uv1 = params.texture.clip.p1;
@ -1490,7 +1475,7 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
/* TODO: Better count method */
cmd = arena_push(&plan->cpu_cmds_arena, struct dx11_cmd);
cmd->kind = params.kind;
cmd->grid.instance_offset = (plan->buffers.grid.instance_buffer->cpu_buffer_arena.pos / sizeof(struct dx11_grid_instance));
cmd->grid.instance_offset = (plan->cmd_buffers.grid.instance_buffer->cpu_buffer_arena.pos / sizeof(struct dx11_grid_instance));
if (plan->cpu_last_cmd) {
plan->cpu_last_cmd->next = cmd;
} else {
@ -1501,7 +1486,7 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
/* Push instance data */
++cmd->grid.instance_count;
struct dx11_grid_instance *instance = dx11_buffer_push(plan->buffers.grid.instance_buffer, sizeof(struct dx11_grid_instance));
struct dx11_grid_instance *instance = dx11_buffer_push(plan->cmd_buffers.grid.instance_buffer, sizeof(struct dx11_grid_instance));
instance->xf = params.grid.xf;
instance->line_thickness = params.grid.line_thickness;
instance->line_spacing = params.grid.line_spacing;
@ -1517,7 +1502,7 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
{
struct dx11_cmd *cmd = arena_push(&plan->cpu_cmds_arena, struct dx11_cmd);
cmd->kind = params.kind;
cmd->test.instance_offset = (plan->buffers.test.instance_buffer->cpu_buffer_arena.pos / sizeof(struct dx11_test_instance));
cmd->test.instance_offset = (plan->cmd_buffers.test.instance_buffer->cpu_buffer_arena.pos / sizeof(struct dx11_test_instance));
if (plan->cpu_last_cmd) {
plan->cpu_last_cmd->next = cmd;
} else {
@ -1527,7 +1512,7 @@ void gpu_push_cmd(struct gpu_handle gpu_plan, struct gpu_cmd_params params)
/* Push instance data */
++cmd->test.instance_count;
struct dx11_test_instance *instance = dx11_buffer_push(plan->buffers.test.instance_buffer, sizeof(struct dx11_test_instance));
struct dx11_test_instance *instance = dx11_buffer_push(plan->cmd_buffers.test.instance_buffer, sizeof(struct dx11_test_instance));
instance->xf = params.test.xf;
} break;
}
@ -1551,43 +1536,17 @@ void gpu_submit_plan(struct gpu_handle gpu_plan)
arena_reset(&plan->cpu_cmds_arena);
/* Submit mesh buffers */
dx11_buffer_submit(plan->buffers.mesh.vertex_buffer);
dx11_buffer_submit(plan->buffers.mesh.index_buffer);
dx11_buffer_submit(plan->cmd_buffers.mesh.vertex_buffer);
dx11_buffer_submit(plan->cmd_buffers.mesh.index_buffer);
/* Submit texture buffers */
dx11_buffer_submit(plan->buffers.texture.instance_buffer);
dx11_buffer_submit(plan->cmd_buffers.texture.instance_buffer);
/* Submit grid buffers */
dx11_buffer_submit(plan->buffers.grid.instance_buffer);
dx11_buffer_submit(plan->cmd_buffers.grid.instance_buffer);
/* Submit test buffers */
dx11_buffer_submit(plan->buffers.test.instance_buffer);
}
/* ========================== *
* Dispatch state
* ========================== */
struct gpu_handle gpu_dispatch_state_alloc(void)
{
__prof;
struct dx11_dispatch_state *state = NULL;
{
struct sys_lock lock = sys_mutex_lock_e(&G.dispatch_states_mutex);
if (G.first_free_dispatch_state) {
state = G.first_free_dispatch_state;
G.first_free_dispatch_state = state->next_free;
} else {
state = arena_push_no_zero(&G.dispatch_states_arena, struct dx11_dispatch_state);
}
sys_mutex_unlock(&lock);
}
MEMZERO_STRUCT(state);
state->header.kind = DX11_HANDLE_KIND_DISPATCH_STATE;
struct gpu_handle res = ZI;
res.v = (u64)state;
return res;
dx11_buffer_submit(plan->cmd_buffers.test.instance_buffer);
}
/* ========================== *
@ -1661,12 +1620,11 @@ INTERNAL void dx11_unbind(u32 flags)
}
/* TODO: Lock resources during dispatch */
void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_params params)
void gpu_dispatch(struct gpu_dispatch_params params)
{
__prof;
__profscope_dx11(G.profiling_ctx, Dispatch, RGB32_F(0.5, 0.2, 0.2));
struct sprite_scope *sprite_scope = sprite_scope_begin();
struct dx11_dispatch_state *state = (struct dx11_dispatch_state *)gpu_dispatch_state.v;
struct dx11_plan *plan = (struct dx11_plan *)params.plan.v;
struct rect viewport = params.draw_target_viewport;
@ -1682,10 +1640,9 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
ID3D11DeviceContext_RSSetViewports(G.devcon, 1, &d3d11_viewport);
struct dx11_texture *final_tex = (struct dx11_texture *)params.draw_target.v;
struct v2i32 final_tex_size = final_tex->size;
/* Allocate constant buffer */
struct dx11_buffer *constant_buffer = state->constant_buffer;
struct dx11_buffer *constant_buffer = plan->constant_buffer;
if (!constant_buffer) {
const D3D11_BUFFER_DESC desc = {
.Usage = D3D11_USAGE_DYNAMIC,
@ -1693,17 +1650,7 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE
};
constant_buffer = dx11_buffer_alloc(desc, NULL);
state->constant_buffer = constant_buffer;
}
/* Allocate / resize albedo */
struct dx11_texture *albedo_tex = state->albedo;
if (!albedo_tex || !v2i32_eq(albedo_tex->size, final_tex->size)) {
if (albedo_tex) {
dx11_texture_release(albedo_tex);
}
albedo_tex = dx11_texture_alloc(DXGI_FORMAT_R8G8B8A8_UNORM, D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_RENDER_TARGET, final_tex_size, NULL);
state->albedo = albedo_tex;
plan->constant_buffer = constant_buffer;
}
/* Regular pass */
@ -1727,8 +1674,8 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
__profscope_dx11(G.profiling_ctx, Draw mesh, RGB32_F(0.5, 0.2, 0.2));
struct dx11_shader *shader = &G.shaders[DX11_SHADER_KIND_MESH];
if (shader->valid) {
struct dx11_buffer *vertex_buffer = plan->buffers.mesh.vertex_buffer;
struct dx11_buffer *index_buffer = plan->buffers.mesh.index_buffer;
struct dx11_buffer *vertex_buffer = plan->cmd_buffers.mesh.vertex_buffer;
struct dx11_buffer *index_buffer = plan->cmd_buffers.mesh.index_buffer;
u32 vertex_offset = cmd->mesh.vertex_offset;
u32 index_offset = cmd->mesh.index_offset;
@ -1786,7 +1733,7 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
}
if (texture && texture->srv) {
struct dx11_buffer *instance_buffer = plan->buffers.texture.instance_buffer;
struct dx11_buffer *instance_buffer = plan->cmd_buffers.texture.instance_buffer;
u32 instance_offset = cmd->texture.instance_offset;
u32 instance_count = cmd->texture.instance_count;
@ -1835,7 +1782,7 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
__profscope_dx11(G.profiling_ctx, Draw grid, RGB32_F(0.2, 0.2, 0.5));
struct dx11_shader *shader = &G.shaders[DX11_SHADER_KIND_GRID];
if (shader->valid) {
struct dx11_buffer *instance_buffer = plan->buffers.grid.instance_buffer;
struct dx11_buffer *instance_buffer = plan->cmd_buffers.grid.instance_buffer;
u32 instance_offset = cmd->grid.instance_offset;
u32 instance_count = cmd->grid.instance_count;
@ -1883,7 +1830,7 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
__profscope_dx11(G.profiling_ctx, Test, RGB32_F(1, 0.2, 1));
struct dx11_shader *shader = &G.shaders[DX11_SHADER_KIND_TEST];
if (shader->valid) {
struct dx11_buffer *instance_buffer = plan->buffers.test.instance_buffer;
struct dx11_buffer *instance_buffer = plan->cmd_buffers.test.instance_buffer;
u32 instance_offset = cmd->test.instance_offset;
u32 instance_count = cmd->test.instance_count;
@ -2035,7 +1982,7 @@ INTERNAL void present_blit(struct dx11_texture *dst, struct dx11_texture *src, s
params.draw_target = dst_texture_handle;
params.draw_target_viewport = RECT_FROM_V2(V2(0, 0), V2(dst->size.x, dst->size.y));
params.draw_target_view = XFORM_IDENT;
gpu_dispatch(G.present_blit_dispatch_state, params);
gpu_dispatch(params);
}
}

View File

@ -80,11 +80,11 @@ struct pipeline_error {
struct string msg;
};
struct dx12_descriptor {
struct dx12_cpu_descriptor_heap *heap;
struct descriptor {
struct cpu_descriptor_heap *heap;
D3D12_CPU_DESCRIPTOR_HANDLE handle;
struct dx12_descriptor *next_free;
struct descriptor *next_free;
};
struct dx12_resource {
@ -96,10 +96,10 @@ struct dx12_resource {
D3D12_CPU_DESCRIPTOR_HANDLE uav_handle;
D3D12_CPU_DESCRIPTOR_HANDLE rtv_handle;
#else
struct dx12_descriptor *cbv_descriptor;
struct dx12_descriptor *srv_descriptor;
struct dx12_descriptor *uav_descriptor;
struct dx12_descriptor *rtv_descriptor;
struct descriptor *cbv_descriptor;
struct descriptor *srv_descriptor;
struct descriptor *uav_descriptor;
struct descriptor *rtv_descriptor;
#endif
D3D12_GPU_VIRTUAL_ADDRESS gpu_address; /* NOTE: 0 for textures */
@ -107,7 +107,7 @@ struct dx12_resource {
struct dx12_resource *next_free;
};
struct dx12_cpu_descriptor_heap {
struct cpu_descriptor_heap {
enum D3D12_DESCRIPTOR_HEAP_TYPE type;
struct arena arena;
struct sys_mutex mutex;
@ -116,13 +116,13 @@ struct dx12_cpu_descriptor_heap {
u32 num_descriptors_reserved;
u32 num_descriptors_capacity;
struct dx12_descriptor *first_free_descriptor;
struct descriptor *first_free_descriptor;
ID3D12DescriptorHeap *heap;
struct D3D12_CPU_DESCRIPTOR_HANDLE handle;
};
struct dx12_gpu_descriptor_heap {
struct gpu_descriptor_heap {
D3D12_DESCRIPTOR_HEAP_TYPE type;
ID3D12DescriptorHeap *heap;
D3D12_CPU_DESCRIPTOR_HANDLE cpu_handle;
@ -132,26 +132,25 @@ struct dx12_gpu_descriptor_heap {
ID3D12Fence *free_fence;
u64 free_fence_value;
struct dx12_gpu_descriptor_heap *prev_free;
struct dx12_gpu_descriptor_heap *next_free;
struct gpu_descriptor_heap *prev_free;
struct gpu_descriptor_heap *next_free;
};
enum dx12_handle_kind {
enum handle_kind {
DX12_HANDLE_KIND_NONE,
DX12_HANDLE_KIND_RESOURCE,
DX12_HANDLE_KIND_PLAN,
DX12_HANDLE_KIND_DISPATCH_STATE,
NUM_DX12_HANDLE_KINDS
};
struct dx12_handle_entry {
enum dx12_handle_kind kind;
struct handle_entry {
enum handle_kind kind;
u64 gen;
u64 idx;
void *data;
struct dx12_handle_entry *next_free;
struct handle_entry *next_free;
};
/* ========================== *
@ -162,14 +161,14 @@ GLOBAL struct {
/* Handles pool */
struct sys_mutex handle_entries_mutex;
struct arena handle_entries_arena;
struct dx12_handle_entry *first_free_handle_entry;
struct handle_entry *first_free_handle_entry;
u64 num_handle_entries_reserved;
/* Descriptor heaps pool */
struct sys_mutex gpu_descriptor_heaps_mutex;
struct arena gpu_descriptor_heaps_arena;
struct dx12_gpu_descriptor_heap *first_free_gpu_descriptor_heap;
struct dx12_gpu_descriptor_heap *last_free_gpu_descriptor_heap;
struct gpu_descriptor_heap *first_free_gpu_descriptor_heap;
struct gpu_descriptor_heap *last_free_gpu_descriptor_heap;
/* Resources pool */
struct sys_mutex resources_mutex;
@ -204,8 +203,8 @@ GLOBAL struct {
u32 desc_counts[D3D12_DESCRIPTOR_HEAP_TYPE_NUM_TYPES];
/* Global descriptor heaps */
struct dx12_cpu_descriptor_heap *cbv_srv_uav_heap;
struct dx12_cpu_descriptor_heap *rtv_heap;
struct cpu_descriptor_heap *cbv_srv_uav_heap;
struct cpu_descriptor_heap *rtv_heap;
/* Command queues */
/* TODO: Add optional mode to route everything to direct queue */
@ -233,7 +232,7 @@ GLOBAL struct {
* ========================== */
INTERNAL APP_EXIT_CALLBACK_FUNC_DEF(gpu_shutdown);
INTERNAL struct dx12_cpu_descriptor_heap *dx12_cpu_descriptor_heap_alloc(enum D3D12_DESCRIPTOR_HEAP_TYPE type);
INTERNAL struct cpu_descriptor_heap *cpu_descriptor_heap_alloc(enum D3D12_DESCRIPTOR_HEAP_TYPE type);
INTERNAL void dx12_init_device(void);
INTERNAL void dx12_init_objects(void);
INTERNAL void dx12_init_swapchain(struct sys_window *window);
@ -356,11 +355,11 @@ INTERNAL APP_EXIT_CALLBACK_FUNC_DEF(gpu_shutdown)
INTERNAL void dx12_resource_release(struct dx12_resource *t);
INTERNAL struct gpu_handle handle_alloc(enum dx12_handle_kind kind, void *data)
INTERNAL struct gpu_handle handle_alloc(enum handle_kind kind, void *data)
{
u64 old_gen = 0;
u64 idx = 0;
struct dx12_handle_entry *entry = NULL;
struct handle_entry *entry = NULL;
{
struct sys_lock lock = sys_mutex_lock_e(&G.handle_entries_mutex);
if (G.first_free_handle_entry) {
@ -369,7 +368,7 @@ INTERNAL struct gpu_handle handle_alloc(enum dx12_handle_kind kind, void *data)
old_gen = entry->gen;
idx = entry->idx;
} else {
entry = arena_push_no_zero(&G.handle_entries_arena, struct dx12_handle_entry);
entry = arena_push_no_zero(&G.handle_entries_arena, struct handle_entry);
idx = G.num_handle_entries_reserved++;
}
sys_mutex_unlock(&lock);
@ -386,12 +385,12 @@ INTERNAL struct gpu_handle handle_alloc(enum dx12_handle_kind kind, void *data)
return res;
}
INTERNAL struct dx12_handle_entry *handle_get_entry(struct gpu_handle handle, struct sys_lock *lock)
INTERNAL struct handle_entry *handle_get_entry(struct gpu_handle handle, struct sys_lock *lock)
{
sys_assert_locked_e_or_s(lock, &G.handle_entries_mutex);
struct dx12_handle_entry *res = NULL;
struct handle_entry *res = NULL;
if (handle.idx > 0 && handle.idx < G.num_handle_entries_reserved) {
struct dx12_handle_entry *tmp = &((struct dx12_handle_entry *)G.handle_entries_arena.base)[handle.idx];
struct handle_entry *tmp = &((struct handle_entry *)G.handle_entries_arena.base)[handle.idx];
if (tmp->gen == handle.gen) {
res = tmp;
}
@ -399,12 +398,12 @@ INTERNAL struct dx12_handle_entry *handle_get_entry(struct gpu_handle handle, st
return res;
}
INTERNAL void *handle_get_data(struct gpu_handle handle, enum dx12_handle_kind kind)
INTERNAL void *handle_get_data(struct gpu_handle handle, enum handle_kind kind)
{
void *data = NULL;
struct sys_lock lock = sys_mutex_lock_s(&G.handle_entries_mutex);
{
struct dx12_handle_entry *entry = handle_get_entry(handle, &lock);
struct handle_entry *entry = handle_get_entry(handle, &lock);
data = entry->data;
#if RTC
/* Handle should match expected kind */
@ -420,13 +419,13 @@ INTERNAL void *handle_get_data(struct gpu_handle handle, enum dx12_handle_kind k
* to ensure freed textures aren't being used in pending command lists. */
void gpu_release(struct gpu_handle handle)
{
enum dx12_handle_kind kind = 0;
enum handle_kind kind = 0;
void *data = NULL;
/* Release handle entry */
struct sys_lock lock = sys_mutex_lock_e(&G.handle_entries_mutex);
{
struct dx12_handle_entry *entry = handle_get_entry(handle, &lock);
struct handle_entry *entry = handle_get_entry(handle, &lock);
if (entry) {
kind = entry->kind;
data = entry->data;
@ -589,8 +588,8 @@ INTERNAL void dx12_init_objects(void)
G.desc_counts[D3D12_DESCRIPTOR_HEAP_TYPE_RTV] = DX12_NUM_RTV_DESCRIPTORS;
/* Create global descriptor heaps */
G.cbv_srv_uav_heap = dx12_cpu_descriptor_heap_alloc(D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
G.rtv_heap = dx12_cpu_descriptor_heap_alloc(D3D12_DESCRIPTOR_HEAP_TYPE_RTV);
G.cbv_srv_uav_heap = cpu_descriptor_heap_alloc(D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
G.rtv_heap = cpu_descriptor_heap_alloc(D3D12_DESCRIPTOR_HEAP_TYPE_RTV);
/* Create direct command queue */
{
@ -1221,12 +1220,12 @@ INTERNAL void pipeline_release(struct pipeline *pipeline)
* CPU descriptor heap
* ========================== */
INTERNAL struct dx12_cpu_descriptor_heap *dx12_cpu_descriptor_heap_alloc(enum D3D12_DESCRIPTOR_HEAP_TYPE type)
INTERNAL struct cpu_descriptor_heap *cpu_descriptor_heap_alloc(enum D3D12_DESCRIPTOR_HEAP_TYPE type)
{
struct dx12_cpu_descriptor_heap *dh = NULL;
struct cpu_descriptor_heap *dh = NULL;
{
struct arena arena = arena_alloc(MEGABYTE(64));
dh = arena_push(&arena, struct dx12_cpu_descriptor_heap);
dh = arena_push(&arena, struct cpu_descriptor_heap);
dh->arena = arena;
}
dh->mutex = sys_mutex_alloc();
@ -1257,7 +1256,7 @@ INTERNAL struct dx12_cpu_descriptor_heap *dx12_cpu_descriptor_heap_alloc(enum D3
}
#if 0
INTERNAL void dx12_cpu_descriptor_heap_release(struct dx12_cpu_descriptor_heap *dh)
INTERNAL void cpu_descriptor_heap_release(struct cpu_descriptor_heap *dh)
{
/* TODO */
(UNUSED)dh;
@ -1268,9 +1267,9 @@ INTERNAL void dx12_cpu_descriptor_heap_release(struct dx12_cpu_descriptor_heap *
* Descriptor
* ========================== */
INTERNAL struct dx12_descriptor *dx12_descriptor_alloc(struct dx12_cpu_descriptor_heap *dh)
INTERNAL struct descriptor *descriptor_alloc(struct cpu_descriptor_heap *dh)
{
struct dx12_descriptor *d = NULL;
struct descriptor *d = NULL;
D3D12_CPU_DESCRIPTOR_HANDLE handle = ZI;
{
struct sys_lock lock = sys_mutex_lock_e(&dh->mutex);
@ -1282,7 +1281,7 @@ INTERNAL struct dx12_descriptor *dx12_descriptor_alloc(struct dx12_cpu_descripto
sys_panic(LIT("Max descriptors reached in heap"));
}
d = arena_push_no_zero(&dh->arena, struct dx12_descriptor);
d = arena_push_no_zero(&dh->arena, struct descriptor);
handle.ptr = dh->handle.ptr + (dh->num_descriptors_reserved * dh->descriptor_size);
++dh->num_descriptors_reserved;
}
@ -1298,12 +1297,12 @@ INTERNAL struct dx12_descriptor *dx12_descriptor_alloc(struct dx12_cpu_descripto
* GPU (shader visible) descriptor heap
* ========================== */
INTERNAL struct dx12_gpu_descriptor_heap *dx12_gpu_descriptor_heap_alloc(struct dx12_cpu_descriptor_heap *dh_cpu)
INTERNAL struct gpu_descriptor_heap *gpu_descriptor_heap_alloc(struct cpu_descriptor_heap *dh_cpu)
{
ASSERT(dh_cpu->type == D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV); /* Src heap must have expected type */
/* Allocate GPU heap */
struct dx12_gpu_descriptor_heap *dh_gpu = NULL;
struct gpu_descriptor_heap *dh_gpu = NULL;
ID3D12DescriptorHeap *heap = NULL;
D3D12_CPU_DESCRIPTOR_HANDLE cpu_handle = ZI;
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle = ZI;
@ -1313,7 +1312,7 @@ INTERNAL struct dx12_gpu_descriptor_heap *dx12_gpu_descriptor_heap_alloc(struct
struct sys_lock lock = sys_mutex_lock_e(&G.gpu_descriptor_heaps_mutex);
/* Find first free & ready heap for reuse */
/* FIXME: Rather than storing fence per heap, store & increment fence per queue and check against it */
for (struct dx12_gpu_descriptor_heap *tmp = G.first_free_gpu_descriptor_heap; tmp; tmp = tmp->next_free) {
for (struct gpu_descriptor_heap *tmp = G.first_free_gpu_descriptor_heap; tmp; tmp = tmp->next_free) {
if (ID3D12Fence_GetCompletedValue(tmp->free_fence) >= tmp->free_fence_value) {
dh_gpu = tmp;
break;
@ -1328,8 +1327,8 @@ INTERNAL struct dx12_gpu_descriptor_heap *dx12_gpu_descriptor_heap_alloc(struct
free_fence = dh_gpu->free_fence;
free_fence_value = dh_gpu->free_fence_value;
/* Remove from free list */
struct dx12_gpu_descriptor_heap *prev = dh_gpu->prev_free;
struct dx12_gpu_descriptor_heap *next = dh_gpu->next_free;
struct gpu_descriptor_heap *prev = dh_gpu->prev_free;
struct gpu_descriptor_heap *next = dh_gpu->next_free;
if (prev) {
prev->next_free = next;
} else {
@ -1342,7 +1341,7 @@ INTERNAL struct dx12_gpu_descriptor_heap *dx12_gpu_descriptor_heap_alloc(struct
}
} else {
/* No available heap available for reuse, allocate new */
dh_gpu = arena_push_no_zero(&G.gpu_descriptor_heaps_arena, struct dx12_gpu_descriptor_heap);
dh_gpu = arena_push_no_zero(&G.gpu_descriptor_heaps_arena, struct gpu_descriptor_heap);
}
sys_mutex_unlock(&lock);
}
@ -1379,7 +1378,7 @@ INTERNAL struct dx12_gpu_descriptor_heap *dx12_gpu_descriptor_heap_alloc(struct
return dh_gpu;
}
INTERNAL void dx12_gpu_descriptor_heap_release(struct dx12_gpu_descriptor_heap *dh, ID3D12CommandQueue *cq)
INTERNAL void gpu_descriptor_heap_release(struct gpu_descriptor_heap *dh, ID3D12CommandQueue *cq)
{
/* Queue fence signal */
++dh->free_fence_value;
@ -1401,18 +1400,46 @@ INTERNAL void dx12_gpu_descriptor_heap_release(struct dx12_gpu_descriptor_heap *
* Plan
* ========================== */
struct dx12_plan {
struct dx12_plan *next_free;
/* TODO: Move command list out of plan struct */
struct plan {
struct arena arena;
ID3D12CommandAllocator *ca_direct;
ID3D12GraphicsCommandList *cl_direct;
struct plan *next_free;
};
INTERNAL struct dx12_plan *dx12_plan_alloc(void)
INTERNAL struct plan *plan_alloc(void)
{
return NULL;
HRESULT hr = 0;
struct plan *plan = NULL;
{
struct arena arena = arena_alloc(MEGABYTE(64));
plan = arena_push(&arena, struct plan);
plan->arena = arena;
}
hr = ID3D12Device_CreateCommandAllocator(G.device, D3D12_COMMAND_LIST_TYPE_DIRECT, &IID_ID3D12CommandAllocator, (void **)&plan->ca_direct);
if (FAILED(hr)) {
sys_panic(LIT("Failed to create command allocator"));
}
hr = ID3D12Device_CreateCommandList(G.device, 0, D3D12_COMMAND_LIST_TYPE_DIRECT, plan->ca_direct, NULL, &IID_ID3D12GraphicsCommandList, (void **)&plan->cl_direct);
if (FAILED(hr)) {
sys_panic(LIT("Failed to create command list"));
}
hr = ID3D12GraphicsCommandList_Close(plan->cl_direct);
if (FAILED(hr)) {
sys_panic(LIT("Failed to close command list during initialization"));
}
return plan;
}
struct gpu_handle gpu_plan_alloc(void)
{
struct dx12_plan *plan = dx12_plan_alloc();
struct plan *plan = plan_alloc();
return handle_alloc(DX12_HANDLE_KIND_PLAN, plan);
}
@ -1467,7 +1494,7 @@ INTERNAL struct dx12_resource *dx12_resource_alloc(D3D12_HEAP_PROPERTIES heap_pr
}
if (view_flags & DX12_RESOURCE_VIEW_FLAG_CBV) {
r->cbv_descriptor = dx12_descriptor_alloc(G.cbv_srv_uav_heap);
r->cbv_descriptor = descriptor_alloc(G.cbv_srv_uav_heap);
D3D12_CONSTANT_BUFFER_VIEW_DESC cbv_desc = ZI;
cbv_desc.BufferLocation = r->gpu_address;
//cbv_desc.SizeInBytes = desc.ByteWidth;
@ -1476,15 +1503,15 @@ INTERNAL struct dx12_resource *dx12_resource_alloc(D3D12_HEAP_PROPERTIES heap_pr
ID3D12Device_CreateConstantBufferView(G.device, &cbv_desc, r->cbv_descriptor->handle);
}
if (view_flags & DX12_RESOURCE_VIEW_FLAG_SRV) {
r->srv_descriptor = dx12_descriptor_alloc(G.cbv_srv_uav_heap);
r->srv_descriptor = descriptor_alloc(G.cbv_srv_uav_heap);
ID3D12Device_CreateShaderResourceView(G.device, r->resource, NULL, r->srv_descriptor->handle);
}
if (view_flags & DX12_RESOURCE_VIEW_FLAG_UAV) {
r->uav_descriptor = dx12_descriptor_alloc(G.cbv_srv_uav_heap);
r->uav_descriptor = descriptor_alloc(G.cbv_srv_uav_heap);
ID3D12Device_CreateUnorderedAccessView(G.device, r->resource, NULL, NULL, r->uav_descriptor->handle);
}
if (view_flags & DX12_RESOURCE_VIEW_FLAG_RTV) {
r->rtv_descriptor = dx12_descriptor_alloc(G.rtv_heap);
r->rtv_descriptor = descriptor_alloc(G.rtv_heap);
ID3D12Device_CreateRenderTargetView(G.device, r->resource, NULL, r->rtv_descriptor->handle);
}
@ -1583,48 +1610,7 @@ struct v2i32 gpu_texture_get_size(struct gpu_handle resource)
* Dispatch
* ========================== */
/* TODO: Move command list off of dispatch state */
struct dx12_dispatch_state {
struct arena arena;
ID3D12CommandAllocator *ca_direct;
ID3D12GraphicsCommandList *cl_direct;
};
INTERNAL struct dx12_dispatch_state *dx12_dispatch_state_alloc(void)
{
HRESULT hr = 0;
struct dx12_dispatch_state *ds = NULL;
{
struct arena arena = arena_alloc(MEGABYTE(64));
ds = arena_push(&arena, struct dx12_dispatch_state);
ds->arena = arena;
}
hr = ID3D12Device_CreateCommandAllocator(G.device, D3D12_COMMAND_LIST_TYPE_DIRECT, &IID_ID3D12CommandAllocator, (void **)&ds->ca_direct);
if (FAILED(hr)) {
sys_panic(LIT("Failed to create command allocator"));
}
hr = ID3D12Device_CreateCommandList(G.device, 0, D3D12_COMMAND_LIST_TYPE_DIRECT, ds->ca_direct, NULL, &IID_ID3D12GraphicsCommandList, (void **)&ds->cl_direct);
if (FAILED(hr)) {
sys_panic(LIT("Failed to create command list"));
}
hr = ID3D12GraphicsCommandList_Close(ds->cl_direct);
if (FAILED(hr)) {
sys_panic(LIT("Failed to close command list during initialization"));
}
return ds;
}
struct gpu_handle gpu_dispatch_state_alloc(void)
{
struct dx12_dispatch_state *ds = dx12_dispatch_state_alloc();
return handle_alloc(DX12_HANDLE_KIND_DISPATCH_STATE, ds);
}
void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_params params)
void gpu_dispatch(struct gpu_dispatch_params params)
{
HRESULT hr = 0;
@ -1646,13 +1632,12 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
d3d12_scissor.right = viewport.x + viewport.width;
d3d12_scissor.bottom = viewport.y + viewport.height;
struct dx12_dispatch_state *dispatch_state = handle_get_data(gpu_dispatch_state, DX12_HANDLE_KIND_DISPATCH_STATE);
struct dx12_plan *plan = handle_get_data(params.plan, DX12_HANDLE_KIND_PLAN);
struct plan *plan = handle_get_data(params.plan, DX12_HANDLE_KIND_PLAN);
struct dx12_resource *target = handle_get_data(params.draw_target, DX12_HANDLE_KIND_RESOURCE);
ID3D12CommandQueue *cq = G.cq_direct;
ID3D12CommandAllocator *ca = dispatch_state->ca_direct;
ID3D12GraphicsCommandList *cl = dispatch_state->cl_direct;
ID3D12CommandAllocator *ca = plan->ca_direct;
ID3D12GraphicsCommandList *cl = plan->cl_direct;
/* FIXME: Use fence to ensure command allocator has finished execution on GPU before resetting */
hr = ID3D12CommandAllocator_Reset(ca);
@ -1666,7 +1651,7 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
}
/* Create temporary srv heap */
struct dx12_gpu_descriptor_heap *temp_descriptor_heap = dx12_gpu_descriptor_heap_alloc(G.cbv_srv_uav_heap);
struct gpu_descriptor_heap *temp_descriptor_heap = gpu_descriptor_heap_alloc(G.cbv_srv_uav_heap);
/* Material pass */
{
@ -1714,7 +1699,7 @@ void gpu_dispatch(struct gpu_handle gpu_dispatch_state, struct gpu_dispatch_para
sys_panic(LIT("Failed to close command list before execution"));
}
dx12_gpu_descriptor_heap_release(temp_descriptor_heap, cq);
gpu_descriptor_heap_release(temp_descriptor_heap, cq);
#if 0
__prof;

View File

@ -75,10 +75,6 @@ GLOBAL struct {
struct gpu_handle world_gpu_plan;
struct gpu_handle ui_gpu_plan;
struct gpu_handle backbuffer_gpu_plan;
struct gpu_handle user_dispatch_state;
struct gpu_handle backbuffer_dispatch_state;
struct xform world_to_user_xf;
@ -259,9 +255,6 @@ struct user_startup_receipt user_startup(struct work_startup_receipt *work_sr,
G.world_to_user_xf = XFORM_IDENT;
G.world_gpu_plan = gpu_plan_alloc();
G.ui_gpu_plan = gpu_plan_alloc();
G.backbuffer_gpu_plan = gpu_plan_alloc();
G.user_dispatch_state = gpu_dispatch_state_alloc();
G.backbuffer_dispatch_state = gpu_dispatch_state_alloc();
G.console_logs_mutex = sys_mutex_alloc();
G.console_logs_arena = arena_alloc(GIGABYTE(64));
@ -2082,7 +2075,6 @@ INTERNAL void user_update(void)
/* Send plans to GPU */
gpu_submit_plan(G.world_gpu_plan);
gpu_submit_plan(G.ui_gpu_plan);
gpu_submit_plan(G.backbuffer_gpu_plan);
/* Clear textures */
gpu_texture_clear(G.user_texture, 0);
@ -2094,7 +2086,7 @@ INTERNAL void user_update(void)
params.draw_target = G.user_texture;
params.draw_target_viewport = user_viewport;
params.draw_target_view = G.world_to_user_xf;
gpu_dispatch(G.user_dispatch_state, params);
gpu_dispatch(params);
}
/* Render ui to user texture */
@ -2104,7 +2096,7 @@ INTERNAL void user_update(void)
params.draw_target = G.user_texture;
params.draw_target_viewport = user_viewport;
params.draw_target_view = XFORM_IDENT;
gpu_dispatch(G.user_dispatch_state, params);
gpu_dispatch(params);
}
/* Present user texture */