1
1
This repository has been archived on 2023-10-09. You can view files and clone it, but cannot push or open issues or pull requests.
Files
blender-archive/source/blender/python/gpu/gpu_py_batch.c
Jeroen Bakker f3515ffed9 Python: Add index and instance based parameter to GPUBatch.draw.
GPUBatch.draw only supported basic drawing methods. Although all
supported platforms support drawing a subset of the index buffer and/or
drawing of multiple instances of the same set of indexes.

This patch adds parameters to the existing `GPUBatch.draw` method to add
support for these drawing methods.

For drawing a subset of the index buffer the parameters `index_start`
and `index_count` can be used.

`my_batch.draw(my_shader, index_start=10, index_count=5)`

For drawing multiple indexes the parameters `instance_start` and
`instance_count` can be used.

`my_batch.draw(my_shader, instance_start=0, instance_count=10)`

Both set of parameters can be used together.

`my_batch.draw(my_shader, index_start=10, index_count=5, instance_start=0, instance_count=10)`

For compatibility reasons the `program` isn't required to be named as
demonstrated by `my_shader`.
2022-12-12 21:53:40 +01:00

391 lines
12 KiB
C

/* SPDX-License-Identifier: GPL-2.0-or-later
* Copyright 2015 Blender Foundation. */
/** \file
* \ingroup bpygpu
*
* This file defines the off-screen functionalities of the 'gpu' module
* used for off-screen OpenGL rendering.
*
* - Use `bpygpu_` for local API.
* - Use `BPyGPU` for public API.
*/
#include <Python.h>
#include "BLI_utildefines.h"
#include "GPU_batch.h"
#include "../generic/py_capi_utils.h"
#include "gpu_py.h"
#include "gpu_py_element.h"
#include "gpu_py_shader.h"
#include "gpu_py_vertex_buffer.h"
#include "gpu_py_batch.h" /* own include */
/* -------------------------------------------------------------------- */
/** \name Utility Functions
* \{ */
static bool pygpu_batch_is_program_or_error(BPyGPUBatch *self)
{
if (!self->batch->shader) {
PyErr_SetString(PyExc_RuntimeError, "batch does not have any program assigned to it");
return false;
}
return true;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name GPUBatch Type
* \{ */
static PyObject *pygpu_batch__tp_new(PyTypeObject *UNUSED(type), PyObject *args, PyObject *kwds)
{
const char *exc_str_missing_arg = "GPUBatch.__new__() missing required argument '%s' (pos %d)";
struct PyC_StringEnum prim_type = {bpygpu_primtype_items, GPU_PRIM_NONE};
BPyGPUVertBuf *py_vertbuf = NULL;
BPyGPUIndexBuf *py_indexbuf = NULL;
static const char *_keywords[] = {"type", "buf", "elem", NULL};
static _PyArg_Parser _parser = {
"|$" /* Optional keyword only arguments. */
"O&" /* `type` */
"O!" /* `buf` */
"O!" /* `elem` */
":GPUBatch.__new__",
_keywords,
0,
};
if (!_PyArg_ParseTupleAndKeywordsFast(args,
kwds,
&_parser,
PyC_ParseStringEnum,
&prim_type,
&BPyGPUVertBuf_Type,
&py_vertbuf,
&BPyGPUIndexBuf_Type,
&py_indexbuf)) {
return NULL;
}
BLI_assert(prim_type.value_found != GPU_PRIM_NONE);
if (prim_type.value_found == GPU_PRIM_LINE_LOOP) {
PyErr_WarnEx(PyExc_DeprecationWarning,
"'LINE_LOOP' is deprecated. Please use 'LINE_STRIP' and close the segment.",
1);
}
else if (prim_type.value_found == GPU_PRIM_TRI_FAN) {
PyErr_WarnEx(
PyExc_DeprecationWarning,
"'TRI_FAN' is deprecated. Please use 'TRI_STRIP' or 'TRIS' and try modifying your "
"vertices or indices to match the topology.",
1);
}
if (py_vertbuf == NULL) {
PyErr_Format(PyExc_TypeError, exc_str_missing_arg, _keywords[1], 2);
return NULL;
}
GPUBatch *batch = GPU_batch_create(
prim_type.value_found, py_vertbuf->buf, py_indexbuf ? py_indexbuf->elem : NULL);
BPyGPUBatch *ret = (BPyGPUBatch *)BPyGPUBatch_CreatePyObject(batch);
#ifdef USE_GPU_PY_REFERENCES
ret->references = PyList_New(py_indexbuf ? 2 : 1);
PyList_SET_ITEM(ret->references, 0, (PyObject *)py_vertbuf);
Py_INCREF(py_vertbuf);
if (py_indexbuf != NULL) {
PyList_SET_ITEM(ret->references, 1, (PyObject *)py_indexbuf);
Py_INCREF(py_indexbuf);
}
BLI_assert(!PyObject_GC_IsTracked((PyObject *)ret));
PyObject_GC_Track(ret);
#endif
return (PyObject *)ret;
}
PyDoc_STRVAR(pygpu_batch_vertbuf_add_doc,
".. method:: vertbuf_add(buf)\n"
"\n"
" Add another vertex buffer to the Batch.\n"
" It is not possible to add more vertices to the batch using this method.\n"
" Instead it can be used to add more attributes to the existing vertices.\n"
" A good use case would be when you have a separate\n"
" vertex buffer for vertex positions and vertex normals.\n"
" Current a batch can have at most " STRINGIFY(GPU_BATCH_VBO_MAX_LEN) " vertex buffers.\n"
"\n"
" :arg buf: The vertex buffer that will be added to the batch.\n"
" :type buf: :class:`gpu.types.GPUVertBuf`\n"
);
static PyObject *pygpu_batch_vertbuf_add(BPyGPUBatch *self, BPyGPUVertBuf *py_buf)
{
if (!BPyGPUVertBuf_Check(py_buf)) {
PyErr_Format(PyExc_TypeError, "Expected a GPUVertBuf, got %s", Py_TYPE(py_buf)->tp_name);
return NULL;
}
if (GPU_vertbuf_get_vertex_len(self->batch->verts[0]) !=
GPU_vertbuf_get_vertex_len(py_buf->buf)) {
PyErr_Format(PyExc_TypeError,
"Expected %d length, got %d",
GPU_vertbuf_get_vertex_len(self->batch->verts[0]),
GPU_vertbuf_get_vertex_len(py_buf->buf));
return NULL;
}
if (self->batch->verts[GPU_BATCH_VBO_MAX_LEN - 1] != NULL) {
PyErr_SetString(
PyExc_RuntimeError,
"Maximum number of vertex buffers exceeded: " STRINGIFY(GPU_BATCH_VBO_MAX_LEN));
return NULL;
}
#ifdef USE_GPU_PY_REFERENCES
/* Hold user */
PyList_Append(self->references, (PyObject *)py_buf);
#endif
GPU_batch_vertbuf_add(self->batch, py_buf->buf);
Py_RETURN_NONE;
}
PyDoc_STRVAR(
pygpu_batch_program_set_doc,
".. method:: program_set(program)\n"
"\n"
" Assign a shader to this batch that will be used for drawing when not overwritten later.\n"
" Note: This method has to be called in the draw context that the batch will be drawn in.\n"
" This function does not need to be called when you always\n"
" set the shader when calling :meth:`gpu.types.GPUBatch.draw`.\n"
"\n"
" :arg program: The program/shader the batch will use in future draw calls.\n"
" :type program: :class:`gpu.types.GPUShader`\n");
static PyObject *pygpu_batch_program_set(BPyGPUBatch *self, BPyGPUShader *py_shader)
{
if (!BPyGPUShader_Check(py_shader)) {
PyErr_Format(PyExc_TypeError, "Expected a GPUShader, got %s", Py_TYPE(py_shader)->tp_name);
return NULL;
}
GPUShader *shader = py_shader->shader;
GPU_batch_set_shader(self->batch, shader);
#ifdef USE_GPU_PY_REFERENCES
/* Remove existing user (if any), hold new user. */
int i = PyList_GET_SIZE(self->references);
while (--i != -1) {
PyObject *py_shader_test = PyList_GET_ITEM(self->references, i);
if (BPyGPUShader_Check(py_shader_test)) {
PyList_SET_ITEM(self->references, i, (PyObject *)py_shader);
Py_INCREF(py_shader);
Py_DECREF(py_shader_test);
/* Only ever reference one shader. */
break;
}
}
if (i != -1) {
PyList_Append(self->references, (PyObject *)py_shader);
}
#endif
Py_RETURN_NONE;
}
PyDoc_STRVAR(pygpu_batch_draw_doc,
".. method:: draw("
"program=None, "
"index_start=0, "
"index_count=0, "
"instance_start=0, "
"instance_count=0)\n"
"\n"
" Run the drawing program with the parameters assigned to the batch.\n"
"\n"
" :arg program: Program that performs the drawing operations.\n"
" If ``None`` is passed, the last program set to this batch will run.\n"
" :type program: :class:`gpu.types.GPUShader`\n"
" :arg index_start: First index to draw.\n"
" :type index_start: int\n"
" :arg index_count: Number of indexes to draw.\n"
" :type index_count: int\n"
" :arg instance_start: First instance to draw.\n"
" :type instance_start: int\n"
" :arg instance_count: Number of instances to draw.\n"
" :type instance_count: int\n");
static PyObject *pygpu_batch_draw(BPyGPUBatch *self, PyObject *args, PyObject *kw)
{
BPyGPUShader *py_program = NULL;
int index_start = 0;
int index_count = 0;
int instance_start = 0;
int instance_count = 0;
static const char *_keywords[] = {
"program", "index_start", "index_count", "instance_start", "instance_count", NULL};
static _PyArg_Parser _parser = {
"O!" /* `program` Optional argument for compatibility reasons. */
"|$" /* Optional, keyword only arguments. */
"i" /* `index_start' */
"i" /* `index_count' */
"i" /* `instance_start' */
"i" /* `instance_count' */
":GPUBatch.draw",
_keywords,
0,
};
if (!_PyArg_ParseTupleAndKeywordsFast(args,
kw,
&_parser,
&BPyGPUShader_Type,
&py_program,
&index_start,
&index_count,
&instance_start,
&instance_count)) {
return NULL;
}
if (py_program == NULL) {
if (!pygpu_batch_is_program_or_error(self)) {
return NULL;
}
}
else if (self->batch->shader != py_program->shader) {
GPU_batch_set_shader(self->batch, py_program->shader);
}
GPU_batch_draw_advanced(self->batch, index_start, index_count, instance_start, instance_count);
Py_RETURN_NONE;
}
static PyObject *pygpu_batch_program_use_begin(BPyGPUBatch *self)
{
if (!pygpu_batch_is_program_or_error(self)) {
return NULL;
}
GPU_shader_bind(self->batch->shader);
Py_RETURN_NONE;
}
static PyObject *pygpu_batch_program_use_end(BPyGPUBatch *self)
{
if (!pygpu_batch_is_program_or_error(self)) {
return NULL;
}
GPU_shader_unbind();
Py_RETURN_NONE;
}
static struct PyMethodDef pygpu_batch__tp_methods[] = {
{"vertbuf_add", (PyCFunction)pygpu_batch_vertbuf_add, METH_O, pygpu_batch_vertbuf_add_doc},
{"program_set", (PyCFunction)pygpu_batch_program_set, METH_O, pygpu_batch_program_set_doc},
{"draw", (PyCFunction)pygpu_batch_draw, METH_VARARGS | METH_KEYWORDS, pygpu_batch_draw_doc},
{"_program_use_begin", (PyCFunction)pygpu_batch_program_use_begin, METH_NOARGS, ""},
{"_program_use_end", (PyCFunction)pygpu_batch_program_use_end, METH_NOARGS, ""},
{NULL, NULL, 0, NULL},
};
#ifdef USE_GPU_PY_REFERENCES
static int pygpu_batch__tp_traverse(BPyGPUBatch *self, visitproc visit, void *arg)
{
Py_VISIT(self->references);
return 0;
}
static int pygpu_batch__tp_clear(BPyGPUBatch *self)
{
Py_CLEAR(self->references);
return 0;
}
static int pygpu_batch__tp_is_gc(BPyGPUBatch *self)
{
return self->references != NULL;
}
#endif
static void pygpu_batch__tp_dealloc(BPyGPUBatch *self)
{
GPU_batch_discard(self->batch);
#ifdef USE_GPU_PY_REFERENCES
PyObject_GC_UnTrack(self);
if (self->references) {
pygpu_batch__tp_clear(self);
Py_XDECREF(self->references);
}
#endif
Py_TYPE(self)->tp_free(self);
}
PyDoc_STRVAR(
pygpu_batch__tp_doc,
".. class:: GPUBatch(type, buf, elem=None)\n"
"\n"
" Reusable container for drawable geometry.\n"
"\n"
" :arg type: The primitive type of geometry to be drawn.\n"
" Possible values are `POINTS`, `LINES`, `TRIS`, `LINE_STRIP`, `LINE_LOOP`, `TRI_STRIP`, "
"`TRI_FAN`, `LINES_ADJ`, `TRIS_ADJ` and `LINE_STRIP_ADJ`.\n"
" :type type: str\n"
" :arg buf: Vertex buffer containing all or some of the attributes required for drawing.\n"
" :type buf: :class:`gpu.types.GPUVertBuf`\n"
" :arg elem: An optional index buffer.\n"
" :type elem: :class:`gpu.types.GPUIndexBuf`\n");
PyTypeObject BPyGPUBatch_Type = {
PyVarObject_HEAD_INIT(NULL, 0).tp_name = "GPUBatch",
.tp_basicsize = sizeof(BPyGPUBatch),
.tp_dealloc = (destructor)pygpu_batch__tp_dealloc,
#ifdef USE_GPU_PY_REFERENCES
.tp_flags = Py_TPFLAGS_DEFAULT | Py_TPFLAGS_HAVE_GC,
.tp_doc = pygpu_batch__tp_doc,
.tp_traverse = (traverseproc)pygpu_batch__tp_traverse,
.tp_clear = (inquiry)pygpu_batch__tp_clear,
.tp_is_gc = (inquiry)pygpu_batch__tp_is_gc,
#else
.tp_flags = Py_TPFLAGS_DEFAULT,
#endif
.tp_methods = pygpu_batch__tp_methods,
.tp_new = pygpu_batch__tp_new,
};
/** \} */
/* -------------------------------------------------------------------- */
/** \name Public API
* \{ */
PyObject *BPyGPUBatch_CreatePyObject(GPUBatch *batch)
{
BPyGPUBatch *self;
#ifdef USE_GPU_PY_REFERENCES
self = (BPyGPUBatch *)_PyObject_GC_New(&BPyGPUBatch_Type);
self->references = NULL;
#else
self = PyObject_New(BPyGPUBatch, &BPyGPUBatch_Type);
#endif
self->batch = batch;
return (PyObject *)self;
}
/** \} */
#undef BPY_GPU_BATCH_CHECK_OBJ