obs-studio/libobs-d3d11/d3d11-shader.cpp

/******************************************************************************
    Copyright (C) 2023 by Lain Bailey <[email protected]>

    This program is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation, either version 2 of the License, or
    (at your option) any later version.

    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program.  If not, see <http://www.gnu.org/licenses/>.
******************************************************************************/

#include "d3d11-subsystem.hpp"
#include "d3d11-shaderprocessor.hpp"
#include <graphics/vec2.h>
#include <graphics/vec3.h>
#include <graphics/matrix3.h>
#include <graphics/matrix4.h>
#include <util/platform.h>
#include <util/util.hpp>

#include <filesystem>
#include <fstream>

void gs_vertex_shader::GetBuffersExpected(
	const vector<D3D11_INPUT_ELEMENT_DESC> &inputs)
{
	for (size_t i = 0; i < inputs.size(); i++) {
		const D3D11_INPUT_ELEMENT_DESC &input = inputs[i];
		if (strcmp(input.SemanticName, "NORMAL") == 0)
			hasNormals = true;
		else if (strcmp(input.SemanticName, "TANGENT") == 0)
			hasTangents = true;
		else if (strcmp(input.SemanticName, "COLOR") == 0)
			hasColors = true;
		else if (strcmp(input.SemanticName, "TEXCOORD") == 0)
			nTexUnits++;
	}
}

gs_vertex_shader::gs_vertex_shader(gs_device_t *device, const char *file,
				   const char *shaderString)
	: gs_shader(device, gs_type::gs_vertex_shader, GS_SHADER_VERTEX),
	  hasNormals(false),
	  hasColors(false),
	  hasTangents(false),
	  nTexUnits(0)
{
	ShaderProcessor processor(device);
	ComPtr<ID3D10Blob> shaderBlob;
	string outputString;
	HRESULT hr;

	processor.Process(shaderString, file);
	processor.BuildString(outputString);
	processor.BuildParams(params);
	processor.BuildInputLayout(layoutData);
	GetBuffersExpected(layoutData);
	BuildConstantBuffer();

	Compile(outputString.c_str(), file, "vs_4_0", shaderBlob.Assign());

	data.resize(shaderBlob->GetBufferSize());
	memcpy(&data[0], shaderBlob->GetBufferPointer(), data.size());

	hr = device->device->CreateVertexShader(data.data(), data.size(), NULL,
						shader.Assign());
	if (FAILED(hr))
		throw HRError("Failed to create vertex shader", hr);

	const UINT layoutSize = (UINT)layoutData.size();
	if (layoutSize > 0) {
		hr = device->device->CreateInputLayout(layoutData.data(),
						       (UINT)layoutSize,
						       data.data(), data.size(),
						       layout.Assign());
		if (FAILED(hr))
			throw HRError("Failed to create input layout", hr);
	}

	viewProj = gs_shader_get_param_by_name(this, "ViewProj");
	world = gs_shader_get_param_by_name(this, "World");
}

gs_pixel_shader::gs_pixel_shader(gs_device_t *device, const char *file,
				 const char *shaderString)
	: gs_shader(device, gs_type::gs_pixel_shader, GS_SHADER_PIXEL)
{
	ShaderProcessor processor(device);
	ComPtr<ID3D10Blob> shaderBlob;
	string outputString;
	HRESULT hr;

	processor.Process(shaderString, file);
	processor.BuildString(outputString);
	processor.BuildParams(params);
	processor.BuildSamplers(samplers);
	BuildConstantBuffer();

	Compile(outputString.c_str(), file, "ps_4_0", shaderBlob.Assign());

	data.resize(shaderBlob->GetBufferSize());
	memcpy(&data[0], shaderBlob->GetBufferPointer(), data.size());

	hr = device->device->CreatePixelShader(data.data(), data.size(), NULL,
					       shader.Assign());
	if (FAILED(hr))
		throw HRError("Failed to create pixel shader", hr);
}

/*
 * Shader compilers will pack constants in to single registers when possible.
 * For example:
 *
 *   uniform float3 test1;
 *   uniform float  test2;
 *
 * will inhabit a single constant register (c0.xyz for 'test1', and c0.w for
 * 'test2')
 *
 * However, if two constants cannot inhabit the same register, the second one
 * must begin at a new register, for example:
 *
 *   uniform float2 test1;
 *   uniform float3 test2;
 *
 * 'test1' will inhabit register constant c0.xy.  However, because there's no
 * room for 'test2, it must use a new register constant entirely (c1.xyz).
 *
 * So if we want to calculate the position of the constants in the constant
 * buffer, we must take this in to account.
 */

void gs_shader::BuildConstantBuffer()
{
	for (size_t i = 0; i < params.size(); i++) {
		gs_shader_param &param = params[i];
		size_t size = 0;

		switch (param.type) {
		case GS_SHADER_PARAM_BOOL:
		case GS_SHADER_PARAM_INT:
		case GS_SHADER_PARAM_FLOAT:
			size = sizeof(float);
			break;
		case GS_SHADER_PARAM_INT2:
		case GS_SHADER_PARAM_VEC2:
			size = sizeof(vec2);
			break;
		case GS_SHADER_PARAM_INT3:
		case GS_SHADER_PARAM_VEC3:
			size = sizeof(float) * 3;
			break;
		case GS_SHADER_PARAM_INT4:
		case GS_SHADER_PARAM_VEC4:
			size = sizeof(vec4);
			break;
		case GS_SHADER_PARAM_MATRIX4X4:
			size = sizeof(float) * 4 * 4;
			break;
		case GS_SHADER_PARAM_TEXTURE:
		case GS_SHADER_PARAM_STRING:
		case GS_SHADER_PARAM_UNKNOWN:
			continue;
		}

		if (param.arrayCount)
			size *= param.arrayCount;

		/* checks to see if this constant needs to start at a new
		 * register */
		if (size && (constantSize & 15) != 0) {
			size_t alignMax = (constantSize + 15) & ~15;

			if ((size + constantSize) > alignMax)
				constantSize = alignMax;
		}

		param.pos = constantSize;
		constantSize += size;
	}

	memset(&bd, 0, sizeof(bd));

	if (constantSize) {
		HRESULT hr;

		bd.ByteWidth = (constantSize + 15) & 0xFFFFFFF0; /* align */
		bd.Usage = D3D11_USAGE_DYNAMIC;
		bd.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
		bd.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;

		hr = device->device->CreateBuffer(&bd, NULL,
						  constants.Assign());
		if (FAILED(hr))
			throw HRError("Failed to create constant buffer", hr);
	}

	for (size_t i = 0; i < params.size(); i++)
		gs_shader_set_default(&params[i]);
}

static uint64_t fnv1a_hash(const char *str, size_t len)
{
	const uint64_t FNV_OFFSET = 14695981039346656037ULL;
	const uint64_t FNV_PRIME = 1099511628211ULL;
	uint64_t hash = FNV_OFFSET;
	for (size_t i = 0; i < len; i++) {
		hash ^= (uint64_t)str[i];
		hash *= FNV_PRIME;
	}
	return hash;
}

void gs_shader::Compile(const char *shaderString, const char *file,
			const char *target, ID3D10Blob **shader)
{
	ComPtr<ID3D10Blob> errorsBlob;
	HRESULT hr;

	bool is_cached = false;
	char hashstr[20];

	if (!shaderString)
		throw "No shader string specified";

	size_t shaderStrLen = strlen(shaderString);
	uint64_t hash = fnv1a_hash(shaderString, shaderStrLen);
	snprintf(hashstr, sizeof(hashstr), "%02llx", hash);

	BPtr program_data =
		os_get_program_data_path_ptr("obs-studio/shader-cache");
	auto cachePath = filesystem::u8path(program_data.Get()) / hashstr;
	// Increment if on-disk format changes
	cachePath += ".v2";

	std::fstream cacheFile;
	cacheFile.exceptions(fstream::badbit | fstream::eofbit);

	if (filesystem::exists(cachePath) && !filesystem::is_empty(cachePath))
		cacheFile.open(cachePath, ios::in | ios::binary | ios::ate);

	if (cacheFile.is_open()) {
		uint64_t checksum;

		try {
			streampos len = cacheFile.tellg();
			// Not enough data for checksum + shader
			if (len <= sizeof(checksum))
				throw length_error("File truncated");

			cacheFile.seekg(0, ios::beg);

			len -= sizeof(checksum);
			device->d3dCreateBlob(len, shader);
			cacheFile.read((char *)(*shader)->GetBufferPointer(),
				       len);
			uint64_t calculated_checksum = fnv1a_hash(
				(char *)(*shader)->GetBufferPointer(), len);

			cacheFile.read((char *)&checksum, sizeof(checksum));
			if (calculated_checksum != checksum)
				throw exception("Checksum mismatch");

			is_cached = true;
		} catch (const exception &e) {
			// Something went wrong reading the cache file, delete it
			blog(LOG_WARNING,
			     "Loading shader cache file failed with \"%s\": %s",
			     e.what(), file);
			cacheFile.close();
			filesystem::remove(cachePath);
		}
	}

	if (!is_cached) {
		hr = device->d3dCompile(shaderString, shaderStrLen, file, NULL,
					NULL, "main", target,
					D3D10_SHADER_OPTIMIZATION_LEVEL3, 0,
					shader, errorsBlob.Assign());
		if (FAILED(hr)) {
			if (errorsBlob != NULL && errorsBlob->GetBufferSize())
				throw ShaderError(errorsBlob, hr);
			else
				throw HRError("Failed to compile shader", hr);
		}

		cacheFile.open(cachePath, ios::out | ios::binary);
		if (cacheFile.is_open()) {
			try {
				uint64_t calculated_checksum = fnv1a_hash(
					(char *)(*shader)->GetBufferPointer(),
					(*shader)->GetBufferSize());

				cacheFile.write(
					(char *)(*shader)->GetBufferPointer(),
					(*shader)->GetBufferSize());
				cacheFile.write((char *)&calculated_checksum,
						sizeof(calculated_checksum));
			} catch (const exception &e) {
				blog(LOG_WARNING,
				     "Writing shader cache file failed with \"%s\": %s",
				     e.what(), file);
				cacheFile.close();
				filesystem::remove(cachePath);
			}
		}
	}

#ifdef DISASSEMBLE_SHADERS
	ComPtr<ID3D10Blob> asmBlob;

	if (!device->d3dDisassemble)
		return;

	hr = device->d3dDisassemble((*shader)->GetBufferPointer(),
				    (*shader)->GetBufferSize(), 0, nullptr,
				    &asmBlob);

	if (SUCCEEDED(hr) && !!asmBlob && asmBlob->GetBufferSize()) {
		blog(LOG_INFO, "=============================================");
		blog(LOG_INFO, "Disassembly output for shader '%s':\n%s", file,
		     asmBlob->GetBufferPointer());
	}
#endif
}

inline void gs_shader::UpdateParam(vector<uint8_t> &constData,
				   gs_shader_param &param, bool &upload)
{
	if (param.type != GS_SHADER_PARAM_TEXTURE) {
		if (!param.curValue.size())
			throw "Not all shader parameters were set";

		/* padding in case the constant needs to start at a new
		 * register */
		if (param.pos > constData.size()) {
			uint8_t zero = 0;

			constData.insert(constData.end(),
					 param.pos - constData.size(), zero);
		}

		constData.insert(constData.end(), param.curValue.begin(),
				 param.curValue.end());

		if (param.changed) {
			upload = true;
			param.changed = false;
		}

	} else if (param.curValue.size() == sizeof(struct gs_shader_texture)) {
		struct gs_shader_texture shader_tex;
		memcpy(&shader_tex, param.curValue.data(), sizeof(shader_tex));
		if (shader_tex.srgb)
			device_load_texture_srgb(device, shader_tex.tex,
						 param.textureID);
		else
			device_load_texture(device, shader_tex.tex,
					    param.textureID);

		if (param.nextSampler) {
			ID3D11SamplerState *state = param.nextSampler->state;
			device->context->PSSetSamplers(param.textureID, 1,
						       &state);
			param.nextSampler = nullptr;
		}
	}
}

void gs_shader::UploadParams()
{
	vector<uint8_t> constData;
	bool upload = false;

	constData.reserve(constantSize);

	for (size_t i = 0; i < params.size(); i++)
		UpdateParam(constData, params[i], upload);

	if (constData.size() != constantSize)
		throw "Invalid constant data size given to shader";

	if (upload) {
		D3D11_MAPPED_SUBRESOURCE map;
		HRESULT hr;

		hr = device->context->Map(constants, 0, D3D11_MAP_WRITE_DISCARD,
					  0, &map);
		if (FAILED(hr))
			throw HRError("Could not lock constant buffer", hr);

		memcpy(map.pData, constData.data(), constData.size());
		device->context->Unmap(constants, 0);
	}
}

void gs_shader_destroy(gs_shader_t *shader)
{
	if (shader && shader->device->lastVertexShader == shader)
		shader->device->lastVertexShader = nullptr;
	delete shader;
}

int gs_shader_get_num_params(const gs_shader_t *shader)
{
	return (int)shader->params.size();
}

gs_sparam_t *gs_shader_get_param_by_idx(gs_shader_t *shader, uint32_t param)
{
	return &shader->params[param];
}

gs_sparam_t *gs_shader_get_param_by_name(gs_shader_t *shader, const char *name)
{
	for (size_t i = 0; i < shader->params.size(); i++) {
		gs_shader_param &param = shader->params[i];
		if (strcmp(param.name.c_str(), name) == 0)
			return &param;
	}

	return NULL;
}

gs_sparam_t *gs_shader_get_viewproj_matrix(const gs_shader_t *shader)
{
	if (shader->type != GS_SHADER_VERTEX)
		return NULL;

	return static_cast<const gs_vertex_shader *>(shader)->viewProj;
}

gs_sparam_t *gs_shader_get_world_matrix(const gs_shader_t *shader)
{
	if (shader->type != GS_SHADER_VERTEX)
		return NULL;

	return static_cast<const gs_vertex_shader *>(shader)->world;
}

void gs_shader_get_param_info(const gs_sparam_t *param,
			      struct gs_shader_param_info *info)
{
	if (!param)
		return;

	info->name = param->name.c_str();
	info->type = param->type;
}

static inline void shader_setval_inline(gs_shader_param *param,
					const void *data, size_t size)
{
	assert(param);
	if (!param)
		return;

	bool size_changed = param->curValue.size() != size;
	if (size_changed)
		param->curValue.resize(size);

	if (size_changed || memcmp(param->curValue.data(), data, size) != 0) {
		memcpy(param->curValue.data(), data, size);
		param->changed = true;
	}
}

void gs_shader_set_bool(gs_sparam_t *param, bool val)
{
	int b_val = (int)val;
	shader_setval_inline(param, &b_val, sizeof(int));
}

void gs_shader_set_float(gs_sparam_t *param, float val)
{
	shader_setval_inline(param, &val, sizeof(float));
}

void gs_shader_set_int(gs_sparam_t *param, int val)
{
	shader_setval_inline(param, &val, sizeof(int));
}

void gs_shader_set_matrix3(gs_sparam_t *param, const struct matrix3 *val)
{
	struct matrix4 mat;
	matrix4_from_matrix3(&mat, val);
	shader_setval_inline(param, &mat, sizeof(matrix4));
}

void gs_shader_set_matrix4(gs_sparam_t *param, const struct matrix4 *val)
{
	shader_setval_inline(param, val, sizeof(matrix4));
}

void gs_shader_set_vec2(gs_sparam_t *param, const struct vec2 *val)
{
	shader_setval_inline(param, val, sizeof(vec2));
}

void gs_shader_set_vec3(gs_sparam_t *param, const struct vec3 *val)
{
	shader_setval_inline(param, val, sizeof(float) * 3);
}

void gs_shader_set_vec4(gs_sparam_t *param, const struct vec4 *val)
{
	shader_setval_inline(param, val, sizeof(vec4));
}

void gs_shader_set_texture(gs_sparam_t *param, gs_texture_t *val)
{
	shader_setval_inline(param, &val, sizeof(gs_texture_t *));
}

void gs_shader_set_val(gs_sparam_t *param, const void *val, size_t size)
{
	shader_setval_inline(param, val, size);
}

void gs_shader_set_default(gs_sparam_t *param)
{
	if (param->defaultValue.size())
		shader_setval_inline(param, param->defaultValue.data(),
				     param->defaultValue.size());
}

void gs_shader_set_next_sampler(gs_sparam_t *param, gs_samplerstate_t *sampler)
{
	param->nextSampler = sampler;
}