From 7f8487a90045b5ad4a95b8af6c0a2d293aec5d20 Mon Sep 17 00:00:00 2001 From: Omar El Sheikh Date: Tue, 9 Feb 2021 17:10:47 -0500 Subject: [PATCH] Split Vertex Position and Attribute Streams Implemented splitting of vertex positions and attributes in the vertex buffer Positions are sequential at the start of the buffer, followed by the additional attributes which are interleaved Made a project setting which enables/disabled the buffer formatting throughout the project Implemented in both GLES2 and GLES3 This improves performance particularly on tile-based GPUs as well as cache performance for something like shadow mapping which only needs position data Updated Docs and Project Setting --- doc/classes/ProjectSettings.xml | 3 + drivers/gles2/rasterizer_storage_gles2.cpp | 58 ++++++++------ drivers/gles3/rasterizer_storage_gles3.cpp | 57 +++++++++----- servers/visual_server.cpp | 88 ++++++++++++++++------ servers/visual_server.h | 2 +- 5 files changed, 141 insertions(+), 67 deletions(-) diff --git a/doc/classes/ProjectSettings.xml b/doc/classes/ProjectSettings.xml index aebe194033b..8ea56998807 100644 --- a/doc/classes/ProjectSettings.xml +++ b/doc/classes/ProjectSettings.xml @@ -1187,6 +1187,9 @@ The default compression level for lossless WebP. Higher levels result in smaller files at the cost of compression speed. Decompression speed is mostly unaffected by the compression level. Supported values are 0 to 9. Note that compression levels above 6 are very slow and offer very little savings. + + On import, mesh vertex data will be split into two streams within a single vertex buffer, one for position data and the other for interleaved attributes data. Recommended to be enabled if targeting mobile devices. Requires manual reimport of meshes after toggling. + If [code]true[/code], allocates the main framebuffer with high dynamic range. High dynamic range allows the use of [Color] values greater than 1. [b]Note:[/b] Only available on the GLES3 backend. diff --git a/drivers/gles2/rasterizer_storage_gles2.cpp b/drivers/gles2/rasterizer_storage_gles2.cpp index 0776902b1e5..17a59c0fdab 100644 --- a/drivers/gles2/rasterizer_storage_gles2.cpp +++ b/drivers/gles2/rasterizer_storage_gles2.cpp @@ -2240,10 +2240,13 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: } //bool has_morph = p_blend_shapes.size(); + bool use_split_stream = GLOBAL_GET("rendering/mesh_storage/split_stream"); Surface::Attrib attribs[VS::ARRAY_MAX]; - int stride = 0; + int attributes_base_offset = 0; + int attributes_stride = 0; + int positions_stride = 0; bool uses_half_float = false; for (int i = 0; i < VS::ARRAY_MAX; i++) { @@ -2256,7 +2259,7 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: } attribs[i].enabled = true; - attribs[i].offset = stride; + attribs[i].offset = attributes_base_offset + attributes_stride; attribs[i].integer = false; switch (i) { @@ -2269,26 +2272,31 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_VERTEX) { attribs[i].type = _GL_HALF_FLOAT_OES; - stride += attribs[i].size * 2; - uses_half_float = true; + positions_stride += attribs[i].size * 2; } else { attribs[i].type = GL_FLOAT; - stride += attribs[i].size * 4; + positions_stride += attribs[i].size * 4; } attribs[i].normalized = GL_FALSE; + if (use_split_stream) { + attributes_base_offset = positions_stride * p_vertex_count; + } else { + attributes_base_offset = positions_stride; + } + } break; case VS::ARRAY_NORMAL: { attribs[i].size = 3; if (p_format & VS::ARRAY_COMPRESS_NORMAL) { attribs[i].type = GL_BYTE; - stride += 4; //pad extra byte + attributes_stride += 4; //pad extra byte attribs[i].normalized = GL_TRUE; } else { attribs[i].type = GL_FLOAT; - stride += 12; + attributes_stride += 12; attribs[i].normalized = GL_FALSE; } @@ -2298,11 +2306,11 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_TANGENT) { attribs[i].type = GL_BYTE; - stride += 4; + attributes_stride += 4; attribs[i].normalized = GL_TRUE; } else { attribs[i].type = GL_FLOAT; - stride += 16; + attributes_stride += 16; attribs[i].normalized = GL_FALSE; } @@ -2312,11 +2320,11 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_COLOR) { attribs[i].type = GL_UNSIGNED_BYTE; - stride += 4; + attributes_stride += 4; attribs[i].normalized = GL_TRUE; } else { attribs[i].type = GL_FLOAT; - stride += 16; + attributes_stride += 16; attribs[i].normalized = GL_FALSE; } @@ -2326,11 +2334,11 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_TEX_UV) { attribs[i].type = _GL_HALF_FLOAT_OES; - stride += 4; + attributes_stride += 4; uses_half_float = true; } else { attribs[i].type = GL_FLOAT; - stride += 8; + attributes_stride += 8; } attribs[i].normalized = GL_FALSE; @@ -2341,11 +2349,11 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_TEX_UV2) { attribs[i].type = _GL_HALF_FLOAT_OES; - stride += 4; + attributes_stride += 4; uses_half_float = true; } else { attribs[i].type = GL_FLOAT; - stride += 8; + attributes_stride += 8; } attribs[i].normalized = GL_FALSE; @@ -2355,10 +2363,10 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_FLAG_USE_16_BIT_BONES) { attribs[i].type = GL_UNSIGNED_SHORT; - stride += 8; + attributes_stride += 8; } else { attribs[i].type = GL_UNSIGNED_BYTE; - stride += 4; + attributes_stride += 4; } attribs[i].normalized = GL_FALSE; @@ -2370,11 +2378,11 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_WEIGHTS) { attribs[i].type = GL_UNSIGNED_SHORT; - stride += 8; + attributes_stride += 8; attribs[i].normalized = GL_TRUE; } else { attribs[i].type = GL_FLOAT; - stride += 16; + attributes_stride += 16; attribs[i].normalized = GL_FALSE; } @@ -2396,13 +2404,21 @@ void RasterizerStorageGLES2::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: } } - for (int i = 0; i < VS::ARRAY_MAX - 1; i++) { - attribs[i].stride = stride; + if (use_split_stream) { + attribs[VS::ARRAY_VERTEX].stride = positions_stride; + for (int i = 1; i < VS::ARRAY_MAX - 1; i++) { + attribs[i].stride = attributes_stride; + } + } else { + for (int i = 0; i < VS::ARRAY_MAX - 1; i++) { + attribs[i].stride = positions_stride + attributes_stride; + } } //validate sizes PoolVector array = p_array; + int stride = positions_stride + attributes_stride; int array_size = stride * p_vertex_count; int index_array_size = 0; if (array.size() != array_size && array.size() + p_vertex_count * 2 == array_size) { diff --git a/drivers/gles3/rasterizer_storage_gles3.cpp b/drivers/gles3/rasterizer_storage_gles3.cpp index 9762b48b25c..d6233db3572 100644 --- a/drivers/gles3/rasterizer_storage_gles3.cpp +++ b/drivers/gles3/rasterizer_storage_gles3.cpp @@ -3349,10 +3349,13 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: } //bool has_morph = p_blend_shapes.size(); + bool use_split_stream = GLOBAL_GET("rendering/mesh_storage/split_stream"); Surface::Attrib attribs[VS::ARRAY_MAX]; - int stride = 0; + int attributes_base_offset = 0; + int attributes_stride = 0; + int positions_stride = 0; for (int i = 0; i < VS::ARRAY_MAX; i++) { attribs[i].index = i; @@ -3364,7 +3367,7 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: } attribs[i].enabled = true; - attribs[i].offset = stride; + attribs[i].offset = attributes_base_offset + attributes_stride; attribs[i].integer = false; switch (i) { @@ -3377,25 +3380,31 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_VERTEX) { attribs[i].type = GL_HALF_FLOAT; - stride += attribs[i].size * 2; + positions_stride += attribs[i].size * 2; } else { attribs[i].type = GL_FLOAT; - stride += attribs[i].size * 4; + positions_stride += attribs[i].size * 4; } attribs[i].normalized = GL_FALSE; + if (use_split_stream) { + attributes_base_offset = positions_stride * p_vertex_count; + } else { + attributes_base_offset = positions_stride; + } + } break; case VS::ARRAY_NORMAL: { attribs[i].size = 3; if (p_format & VS::ARRAY_COMPRESS_NORMAL) { attribs[i].type = GL_BYTE; - stride += 4; //pad extra byte + attributes_stride += 4; //pad extra byte attribs[i].normalized = GL_TRUE; } else { attribs[i].type = GL_FLOAT; - stride += 12; + attributes_stride += 12; attribs[i].normalized = GL_FALSE; } @@ -3405,11 +3414,11 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_TANGENT) { attribs[i].type = GL_BYTE; - stride += 4; + attributes_stride += 4; attribs[i].normalized = GL_TRUE; } else { attribs[i].type = GL_FLOAT; - stride += 16; + attributes_stride += 16; attribs[i].normalized = GL_FALSE; } @@ -3419,11 +3428,11 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_COLOR) { attribs[i].type = GL_UNSIGNED_BYTE; - stride += 4; + attributes_stride += 4; attribs[i].normalized = GL_TRUE; } else { attribs[i].type = GL_FLOAT; - stride += 16; + attributes_stride += 16; attribs[i].normalized = GL_FALSE; } @@ -3433,10 +3442,10 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_TEX_UV) { attribs[i].type = GL_HALF_FLOAT; - stride += 4; + attributes_stride += 4; } else { attribs[i].type = GL_FLOAT; - stride += 8; + attributes_stride += 8; } attribs[i].normalized = GL_FALSE; @@ -3447,10 +3456,10 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_TEX_UV2) { attribs[i].type = GL_HALF_FLOAT; - stride += 4; + attributes_stride += 4; } else { attribs[i].type = GL_FLOAT; - stride += 8; + attributes_stride += 8; } attribs[i].normalized = GL_FALSE; @@ -3460,10 +3469,10 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_FLAG_USE_16_BIT_BONES) { attribs[i].type = GL_UNSIGNED_SHORT; - stride += 8; + attributes_stride += 8; } else { attribs[i].type = GL_UNSIGNED_BYTE; - stride += 4; + attributes_stride += 4; } attribs[i].normalized = GL_FALSE; @@ -3475,11 +3484,11 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: if (p_format & VS::ARRAY_COMPRESS_WEIGHTS) { attribs[i].type = GL_UNSIGNED_SHORT; - stride += 8; + attributes_stride += 8; attribs[i].normalized = GL_TRUE; } else { attribs[i].type = GL_FLOAT; - stride += 16; + attributes_stride += 16; attribs[i].normalized = GL_FALSE; } @@ -3501,12 +3510,20 @@ void RasterizerStorageGLES3::mesh_add_surface(RID p_mesh, uint32_t p_format, VS: } } - for (int i = 0; i < VS::ARRAY_MAX - 1; i++) { - attribs[i].stride = stride; + if (use_split_stream) { + attribs[VS::ARRAY_VERTEX].stride = positions_stride; + for (int i = 1; i < VS::ARRAY_MAX - 1; i++) { + attribs[i].stride = attributes_stride; + } + } else { + for (int i = 0; i < VS::ARRAY_MAX - 1; i++) { + attribs[i].stride = positions_stride + attributes_stride; + } } //validate sizes + int stride = positions_stride + attributes_stride; int array_size = stride * p_vertex_count; int index_array_size = 0; if (array.size() != array_size && array.size() + p_vertex_count * 2 == array_size) { diff --git a/servers/visual_server.cpp b/servers/visual_server.cpp index 49410bb9039..fac03115916 100644 --- a/servers/visual_server.cpp +++ b/servers/visual_server.cpp @@ -329,7 +329,7 @@ RID VisualServer::get_white_texture() { #define SMALL_VEC2 Vector2(0.00001, 0.00001) #define SMALL_VEC3 Vector3(0.00001, 0.00001, 0.00001) -Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_t *p_offsets, uint32_t p_stride, PoolVector &r_vertex_array, int p_vertex_array_len, PoolVector &r_index_array, int p_index_array_len, AABB &r_aabb, Vector &r_bone_aabb) { +Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_t *p_offsets, uint32_t *p_stride, PoolVector &r_vertex_array, int p_vertex_array_len, PoolVector &r_index_array, int p_index_array_len, AABB &r_aabb, Vector &r_bone_aabb) { PoolVector::Write vw = r_vertex_array.write(); PoolVector::Write iw; @@ -360,7 +360,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ for (int i = 0; i < p_vertex_array_len; i++) { uint16_t vector[2] = { Math::make_half_float(src[i].x), Math::make_half_float(src[i].y) }; - memcpy(&vw[p_offsets[ai] + i * p_stride], vector, sizeof(uint16_t) * 2); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], vector, sizeof(uint16_t) * 2); if (i == 0) { aabb = Rect2(src[i], SMALL_VEC2); //must have a bit of size @@ -373,7 +373,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ for (int i = 0; i < p_vertex_array_len; i++) { float vector[2] = { src[i].x, src[i].y }; - memcpy(&vw[p_offsets[ai] + i * p_stride], vector, sizeof(float) * 2); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], vector, sizeof(float) * 2); if (i == 0) { aabb = Rect2(src[i], SMALL_VEC2); //must have a bit of size @@ -399,7 +399,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ for (int i = 0; i < p_vertex_array_len; i++) { uint16_t vector[4] = { Math::make_half_float(src[i].x), Math::make_half_float(src[i].y), Math::make_half_float(src[i].z), Math::make_half_float(1.0) }; - memcpy(&vw[p_offsets[ai] + i * p_stride], vector, sizeof(uint16_t) * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], vector, sizeof(uint16_t) * 4); if (i == 0) { aabb = AABB(src[i], SMALL_VEC3); @@ -412,7 +412,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ for (int i = 0; i < p_vertex_array_len; i++) { float vector[3] = { src[i].x, src[i].y, src[i].z }; - memcpy(&vw[p_offsets[ai] + i * p_stride], vector, sizeof(float) * 3); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], vector, sizeof(float) * 3); if (i == 0) { aabb = AABB(src[i], SMALL_VEC3); @@ -446,13 +446,13 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ 0, }; - memcpy(&vw[p_offsets[ai] + i * p_stride], vector, 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], vector, 4); } } else { for (int i = 0; i < p_vertex_array_len; i++) { float vector[3] = { src[i].x, src[i].y, src[i].z }; - memcpy(&vw[p_offsets[ai] + i * p_stride], vector, 3 * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], vector, 3 * 4); } } @@ -477,7 +477,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ (int8_t)CLAMP(src[i * 4 + 3] * 127, -128, 127) }; - memcpy(&vw[p_offsets[ai] + i * p_stride], xyzw, 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], xyzw, 4); } } else { @@ -489,7 +489,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ src[i * 4 + 3] }; - memcpy(&vw[p_offsets[ai] + i * p_stride], xyzw, 4 * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], xyzw, 4 * 4); } } @@ -512,11 +512,11 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ colors[j] = CLAMP(int((src[i][j]) * 255.0), 0, 255); } - memcpy(&vw[p_offsets[ai] + i * p_stride], colors, 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], colors, 4); } } else { for (int i = 0; i < p_vertex_array_len; i++) { - memcpy(&vw[p_offsets[ai] + i * p_stride], &src[i], 4 * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], &src[i], 4 * 4); } } @@ -535,14 +535,14 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ if (p_format & ARRAY_COMPRESS_TEX_UV) { for (int i = 0; i < p_vertex_array_len; i++) { uint16_t uv[2] = { Math::make_half_float(src[i].x), Math::make_half_float(src[i].y) }; - memcpy(&vw[p_offsets[ai] + i * p_stride], uv, 2 * 2); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], uv, 2 * 2); } } else { for (int i = 0; i < p_vertex_array_len; i++) { float uv[2] = { src[i].x, src[i].y }; - memcpy(&vw[p_offsets[ai] + i * p_stride], uv, 2 * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], uv, 2 * 4); } } @@ -562,14 +562,14 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ if (p_format & ARRAY_COMPRESS_TEX_UV2) { for (int i = 0; i < p_vertex_array_len; i++) { uint16_t uv[2] = { Math::make_half_float(src[i].x), Math::make_half_float(src[i].y) }; - memcpy(&vw[p_offsets[ai] + i * p_stride], uv, 2 * 2); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], uv, 2 * 2); } } else { for (int i = 0; i < p_vertex_array_len; i++) { float uv[2] = { src[i].x, src[i].y }; - memcpy(&vw[p_offsets[ai] + i * p_stride], uv, 2 * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], uv, 2 * 4); } } } break; @@ -591,7 +591,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ data[j] = CLAMP(src[i * VS::ARRAY_WEIGHTS_SIZE + j] * 65535, 0, 65535); } - memcpy(&vw[p_offsets[ai] + i * p_stride], data, 2 * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], data, 2 * 4); } } else { for (int i = 0; i < p_vertex_array_len; i++) { @@ -600,7 +600,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ data[j] = src[i * VS::ARRAY_WEIGHTS_SIZE + j]; } - memcpy(&vw[p_offsets[ai] + i * p_stride], data, 4 * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], data, 4 * 4); } } @@ -624,7 +624,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ max_bone = MAX(data[j], max_bone); } - memcpy(&vw[p_offsets[ai] + i * p_stride], data, 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], data, 4); } } else { @@ -635,7 +635,7 @@ Error VisualServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint32_ max_bone = MAX(data[j], max_bone); } - memcpy(&vw[p_offsets[ai] + i * p_stride], data, 2 * 4); + memcpy(&vw[p_offsets[ai] + i * p_stride[ai]], data, 2 * 4); } } @@ -856,6 +856,8 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim ERR_FAIL_INDEX(p_primitive, VS::PRIMITIVE_MAX); ERR_FAIL_COND(p_arrays.size() != VS::ARRAY_MAX); + bool use_split_stream = GLOBAL_GET("rendering/mesh_storage/split_stream"); + uint32_t format = 0; // validation @@ -908,8 +910,11 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim } uint32_t offsets[VS::ARRAY_MAX]; + uint32_t strides[VS::ARRAY_MAX]; - int total_elem_size = 0; + int attributes_base_offset = 0; + int attributes_stride = 0; + int positions_stride = 0; for (int i = 0; i < VS::ARRAY_MAX; i++) { offsets[i] = 0; //reset @@ -944,6 +949,14 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim elem_size = 8; } + offsets[i] = 0; + positions_stride = elem_size; + if (use_split_stream) { + attributes_base_offset = elem_size * array_len; + } else { + attributes_base_offset = elem_size; + } + } break; case VS::ARRAY_NORMAL: { if (p_compress_format & ARRAY_COMPRESS_NORMAL) { @@ -951,6 +964,8 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim } else { elem_size = sizeof(float) * 3; } + offsets[i] = attributes_base_offset + attributes_stride; + attributes_stride += elem_size; } break; @@ -960,6 +975,8 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim } else { elem_size = sizeof(float) * 4; } + offsets[i] = attributes_base_offset + attributes_stride; + attributes_stride += elem_size; } break; case VS::ARRAY_COLOR: { @@ -968,6 +985,9 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim } else { elem_size = sizeof(float) * 4; } + offsets[i] = attributes_base_offset + attributes_stride; + attributes_stride += elem_size; + } break; case VS::ARRAY_TEX_UV: { if (p_compress_format & ARRAY_COMPRESS_TEX_UV) { @@ -975,6 +995,8 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim } else { elem_size = sizeof(float) * 2; } + offsets[i] = attributes_base_offset + attributes_stride; + attributes_stride += elem_size; } break; @@ -984,6 +1006,8 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim } else { elem_size = sizeof(float) * 2; } + offsets[i] = attributes_base_offset + attributes_stride; + attributes_stride += elem_size; } break; case VS::ARRAY_WEIGHTS: { @@ -992,6 +1016,8 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim } else { elem_size = sizeof(float) * 4; } + offsets[i] = attributes_base_offset + attributes_stride; + attributes_stride += elem_size; } break; case VS::ARRAY_BONES: { @@ -1013,6 +1039,8 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim p_compress_format &= ~ARRAY_FLAG_USE_16_BIT_BONES; elem_size = sizeof(uint32_t); } + offsets[i] = attributes_base_offset + attributes_stride; + attributes_stride += elem_size; } break; case VS::ARRAY_INDEX: { @@ -1034,15 +1062,23 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim ERR_FAIL(); } } + } - offsets[i] = total_elem_size; - total_elem_size += elem_size; + if (use_split_stream) { + strides[VS::ARRAY_VERTEX] = positions_stride; + for (int i = 1; i < VS::ARRAY_MAX - 1; i++) { + strides[i] = attributes_stride; + } + } else { + for (int i = 0; i < VS::ARRAY_MAX - 1; i++) { + strides[i] = positions_stride + attributes_stride; + } } uint32_t mask = (1 << ARRAY_MAX) - 1; format |= (~mask) & p_compress_format; //make the full format - int array_size = total_elem_size * array_len; + int array_size = (positions_stride + attributes_stride) * array_len; PoolVector vertex_array; vertex_array.resize(array_size); @@ -1055,7 +1091,7 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim AABB aabb; Vector bone_aabb; - Error err = _surface_set_data(p_arrays, format, offsets, total_elem_size, vertex_array, array_len, index_array, index_array_len, aabb, bone_aabb); + Error err = _surface_set_data(p_arrays, format, offsets, strides, vertex_array, array_len, index_array, index_array_len, aabb, bone_aabb); ERR_FAIL_COND_MSG(err, "Invalid array format for surface."); Vector> blend_shape_data; @@ -1066,7 +1102,7 @@ void VisualServer::mesh_add_surface_from_arrays(RID p_mesh, PrimitiveType p_prim PoolVector noindex; AABB laabb; - Error err2 = _surface_set_data(p_blend_shapes[i], format & ~ARRAY_FORMAT_INDEX, offsets, total_elem_size, vertex_array_shape, array_len, noindex, 0, laabb, bone_aabb); + Error err2 = _surface_set_data(p_blend_shapes[i], format & ~ARRAY_FORMAT_INDEX, offsets, strides, vertex_array_shape, array_len, noindex, 0, laabb, bone_aabb); aabb.merge_with(laabb); ERR_FAIL_COND_MSG(err2 != OK, "Invalid blend shape array format for surface."); @@ -2299,6 +2335,8 @@ VisualServer::VisualServer() { GLOBAL_DEF("rendering/quality/shading/force_blinn_over_ggx", false); GLOBAL_DEF("rendering/quality/shading/force_blinn_over_ggx.mobile", true); + GLOBAL_DEF("rendering/mesh_storage/split_stream", false); + GLOBAL_DEF("rendering/quality/depth_prepass/enable", true); GLOBAL_DEF("rendering/quality/depth_prepass/disable_for_vendors", "PowerVR,Mali,Adreno,Apple"); diff --git a/servers/visual_server.h b/servers/visual_server.h index b46c5173937..a9ea43835be 100644 --- a/servers/visual_server.h +++ b/servers/visual_server.h @@ -60,7 +60,7 @@ protected: RID white_texture; RID test_material; - Error _surface_set_data(Array p_arrays, uint32_t p_format, uint32_t *p_offsets, uint32_t p_stride, PoolVector &r_vertex_array, int p_vertex_array_len, PoolVector &r_index_array, int p_index_array_len, AABB &r_aabb, Vector &r_bone_aabb); + Error _surface_set_data(Array p_arrays, uint32_t p_format, uint32_t *p_offsets, uint32_t *p_stride, PoolVector &r_vertex_array, int p_vertex_array_len, PoolVector &r_index_array, int p_index_array_len, AABB &r_aabb, Vector &r_bone_aabb); static VisualServer *(*create_func)(); static void _bind_methods();