libretro/RetroArch
1
0
Fork 0
mirror of https://github.com/libretro/RetroArch.git synced 2024-12-04 06:11:17 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
f2cae17021
RetroArch/gfx/common/metal_common.m
Stuart Carnie 367c15f488 feat(Metal): font rendering 
* FPS and other OSD messages render correctly
* add autoreleasepool around entire render loop to avoid memory leaks
2018-06-23 23:54:23 -07:00

1042 lines
32 KiB
Objective-C
Raw Blame History

//
// metal_common.m
// RetroArch_Metal
//
// Created by Stuart Carnie on 5/14/18.
//
#import <Foundation/Foundation.h>
#import "metal_common.h"
#import "../../ui/drivers/cocoa/cocoa_common.h"
#import <memory.h>
#import <gfx/video_frame.h>
#import <Metal/Metal.h>
#import <QuartzCore/QuartzCore.h>
#import <stddef.h>
#include <simd/simd.h>
#define STRUCT_ASSIGN(x, y) \
{ \
NSObject * __y = y; \
if (x != nil) { \
NSObject * __foo = (__bridge_transfer NSObject *)(__bridge void *)(x); \
__foo = nil; \
x = (__bridge __typeof__(x))nil; \
} \
if (__y != nil) \
x = (__bridge __typeof__(x))(__bridge_retained void *)((NSObject *)__y); \
}
@interface FrameView()
@property (readwrite) video_viewport_t *viewport;
- (instancetype)initWithDescriptor:(ViewDescriptor *)td renderer:(Renderer *)renderer;
- (void)drawWithContext:(Context *)ctx;
- (void)drawWithEncoder:(id<MTLRenderCommandEncoder>)rce;
@end
#pragma mark - private categories
@interface MetalMenu()
@property (readwrite) Renderer *renderer;
@end
@implementation MetalDriver
{
id<MTLDevice> _device;
Renderer *_renderer;
FrameView *_frameView;
video_info_t _video;
}
- (instancetype)init
{
if (self = [super init]) {
_frameCount = 0;
_viewport = (video_viewport_t *)calloc(1, sizeof(video_viewport_t));
_menu = [MetalMenu new];
}
return self;
}
- (void)dealloc
{
RARCH_LOG("[MetalDriver]: destroyed\n");
if (_viewport) {
free(_viewport);
_viewport = nil;
}
}
- (Context *)context {
return _renderer.context;
}
#pragma mark - video
- (void)setVideo:(const video_info_t *)video
{
_video = *video;
if (!_renderer) {
id<MTLDevice> device = MTLCreateSystemDefaultDevice();
_device = device;
MetalView *view = (MetalView *)apple_platform.renderView;
view.device = device;
CAMetalLayer *layer = (CAMetalLayer *)view.layer;
//layer.device = device;
_renderer = [[Renderer alloc] initWithDevice:device layer:layer];
_menu.renderer = _renderer;
}
if (!_frameView) {
ViewDescriptor *vd = [ViewDescriptor new];
vd.format = _video.rgb32 ? RPixelFormatBGRX8Unorm : RPixelFormatB5G6R5Unorm;
vd.size = CGSizeMake(video->width, video->height);
vd.filter = _video.smooth ? RTextureFilterLinear : RTextureFilterNearest;
_frameView = [[FrameView alloc] initWithDescriptor:vd renderer:_renderer];
_frameView.viewport = _viewport;
[_renderer addView:_frameView];
[_renderer sendViewToBack:_frameView];
[_frameView setFilteringIndex:0 smooth:video->smooth];
}
}
- (void)beginFrame
{
video_driver_update_viewport(_viewport, NO, _keepAspect);
[_renderer beginFrame];
}
- (void)drawViews {
[_renderer drawViews];
}
- (void)endFrame
{
[_renderer endFrame];
}
- (void)setNeedsResize
{
// TODO(sgc): resize all drawables
}
#pragma mark - MTKViewDelegate
- (void)mtkView:(MTKView *)view drawableSizeWillChange:(CGSize)size {
RARCH_LOG("[MetalDriver] drawableSizeWillChange: %s\n", NSStringFromSize(size).UTF8String);
_viewport->full_width = (unsigned int)size.width;
_viewport->full_height = (unsigned int)size.height;
video_driver_set_size(&_viewport->full_width, &_viewport->full_height);
[_renderer drawableSizeWillChange:size];
video_driver_update_viewport(_viewport, NO, _keepAspect);
}
- (void)drawInMTKView:(MTKView *)view {
}
@end
@implementation MetalMenu
{
Renderer *_renderer;
TexturedView *_view;
BOOL _enabled;
}
- (void)setEnabled:(BOOL)enabled
{
if (_enabled == enabled) return;
_enabled = enabled;
_view.visible = enabled;
}
- (BOOL)enabled
{
return _enabled;
}
- (void)updateWidth:(int)width
height:(int)height
format:(RPixelFormat)format
filter:(RTextureFilter)filter
{
CGSize size = CGSizeMake(width, height);
if (_view) {
if (!(CGSizeEqualToSize(_view.size, size) &&
_view.format == format &&
_view.filter == filter)) {
[_renderer removeView:_view];
_view = nil;
}
}
if (!_view) {
ViewDescriptor *vd = [ViewDescriptor new];
vd.format = format;
vd.filter = filter;
vd.size = size;
_view = [[TexturedView alloc] initWithDescriptor:vd renderer:_renderer];
[_renderer addView:_view];
_view.visible = _enabled;
}
}
- (void)updateFrame:(void const *)source
{
[_view updateFrame:source pitch:RPixelFormatToBPP(_view.format) * (NSUInteger)_view.size.width];
}
@end
#pragma mark - FrameView
#define ALIGN(x) __attribute__((aligned(x)))
typedef struct
{
float x;
float y;
float z;
float w;
} float4_t;
typedef struct texture
{
__unsafe_unretained id<MTLTexture> view;
float4_t size_data;
} texture_t;
typedef struct ALIGN(16)
{
matrix_float4x4 mvp;
struct
{
texture_t texture[GFX_MAX_FRAME_HISTORY + 1];
MTLViewport viewport;
float4_t output_size;
} frame;
struct
{
__unsafe_unretained id<MTLBuffer> buffers[SLANG_CBUFFER_MAX];
texture_t rt;
texture_t feedback;
uint32_t frame_count;
pass_semantics_t semantics;
MTLViewport viewport;
__unsafe_unretained id<MTLRenderPipelineState> _state;
} pass[GFX_MAX_SHADERS];
texture_t luts[GFX_MAX_TEXTURES];
} engine_t;
@implementation FrameView
{
__weak Renderer *_renderer;
Context *_context;
id<MTLTexture> _texture; // final render texture
Vertex _v[4];
CGSize _size; // size of view in pixels
CGRect _frame;
NSUInteger _bpp;
id<MTLBuffer> _pixels; // frame buffer in _srcFmt
bool _pixelsDirty;
id<MTLSamplerState> _samplers[RARCH_FILTER_MAX][RARCH_WRAP_MAX];
struct video_shader *_shader;
id<MTLFence> _fence;
engine_t _engine;
bool resize_render_targets;
bool init_history;
video_viewport_t *_viewport;
}
- (instancetype)initWithDescriptor:(ViewDescriptor *)d renderer:(Renderer *)r
{
self = [super init];
if (self) {
_renderer = r;
_context = r.context;
_format = d.format;
_bpp = RPixelFormatToBPP(_format);
_filter = d.filter;
if (_format == RPixelFormatBGRA8Unorm || _format == RPixelFormatBGRX8Unorm) {
_drawState = ViewDrawStateEncoder;
} else {
_drawState = ViewDrawStateAll;
}
_visible = YES;
_engine.mvp = matrix_proj_ortho(0, 1, 0, 1);
[self _initSamplers];
self.size = d.size;
self.frame = CGRectMake(0, 0, 1, 1);
resize_render_targets = YES;
}
return self;
}
- (void)_initSamplers
{
MTLSamplerDescriptor *sd = [MTLSamplerDescriptor new];
/* Initialize samplers */
for (unsigned i = 0; i < RARCH_WRAP_MAX; i++) {
switch (i) {
case RARCH_WRAP_BORDER:
sd.sAddressMode = MTLSamplerAddressModeClampToBorderColor;
break;
case RARCH_WRAP_EDGE:
sd.sAddressMode = MTLSamplerAddressModeClampToEdge;
break;
case RARCH_WRAP_REPEAT:
sd.sAddressMode = MTLSamplerAddressModeRepeat;
break;
case RARCH_WRAP_MIRRORED_REPEAT:
sd.sAddressMode = MTLSamplerAddressModeMirrorRepeat;
break;
default:
continue;
}
sd.tAddressMode = sd.sAddressMode;
sd.rAddressMode = sd.sAddressMode;
sd.minFilter = MTLSamplerMinMagFilterLinear;
sd.magFilter = MTLSamplerMinMagFilterLinear;
id<MTLSamplerState> ss = [_context.device newSamplerStateWithDescriptor:sd];
_samplers[RARCH_FILTER_LINEAR][i] = ss;
sd.minFilter = MTLSamplerMinMagFilterNearest;
sd.magFilter = MTLSamplerMinMagFilterNearest;
ss = [_context.device newSamplerStateWithDescriptor:sd];
_samplers[RARCH_FILTER_NEAREST][i] = ss;
}
}
- (void)setFilteringIndex:(int)index smooth:(bool)smooth
{
for (int i = 0; i < RARCH_WRAP_MAX; i++) {
if (smooth)
_samplers[RARCH_FILTER_UNSPEC][i] = _samplers[RARCH_FILTER_LINEAR][i];
else
_samplers[RARCH_FILTER_UNSPEC][i] = _samplers[RARCH_FILTER_NEAREST][i];
}
}
- (void)setSize:(CGSize)size
{
if (CGSizeEqualToSize(_size, size)) {
return;
}
_size = size;
resize_render_targets = YES;
if (_format != RPixelFormatBGRA8Unorm && _format != RPixelFormatBGRX8Unorm) {
_pixels = [_context.device newBufferWithLength:(NSUInteger)(size.width * size.height * 2)
options:MTLResourceStorageModeManaged];
}
}
- (CGSize)size
{
return _size;
}
- (void)setFrame:(CGRect)frame
{
if (CGRectEqualToRect(_frame, frame)) {
return;
}
_frame = frame;
// update vertices
CGPoint o = frame.origin;
CGSize s = frame.size;
CGFloat l = o.x;
CGFloat t = o.y;
CGFloat r = o.x + s.width;
CGFloat b = o.y + s.height;
Vertex v[4] = {
{{l, b, 0}, {0, 1}},
{{r, b, 0}, {1, 1}},
{{l, t, 0}, {0, 0}},
{{r, t, 0}, {1, 0}},
};
memcpy(_v, v, sizeof(_v));
}
- (CGRect)frame
{
return _frame;
}
- (void)_convertFormat
{
if (_format == RPixelFormatBGRA8Unorm || _format == RPixelFormatBGRX8Unorm)
return;
if (!_pixelsDirty)
return;
[_renderer.conv convertFormat:_format from:_pixels to:_texture];
_pixelsDirty = NO;
}
- (void)_updateHistory
{
if (_shader) {
if (_shader->history_size) {
if (init_history)
[self _initHistory];
else {
int k;
/* todo: what about frame-duping ?
* maybe clone d3d10_texture_t with AddRef */
texture_t tmp = _engine.frame.texture[_shader->history_size];
for (k = _shader->history_size; k > 0; k--)
_engine.frame.texture[k] = _engine.frame.texture[k - 1];
_engine.frame.texture[0] = tmp;
}
}
}
/* either no history, or we moved a texture of a different size in the front slot */
if (_engine.frame.texture[0].size_data.x != _size.width ||
_engine.frame.texture[0].size_data.y != _size.height) {
MTLTextureDescriptor *td = [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:MTLPixelFormatBGRA8Unorm
width:(NSUInteger)_size.width
height:(NSUInteger)_size.height
mipmapped:false];
td.usage = MTLTextureUsageShaderRead | MTLTextureUsageShaderWrite;
[self _initTexture:&_engine.frame.texture[0] withDescriptor:td];
}
}
- (void)updateFrame:(void const *)src pitch:(NSUInteger)pitch
{
if (_shader && (_engine.frame.output_size.x != _viewport->width ||
_engine.frame.output_size.y != _viewport->height)) {
resize_render_targets = YES;
}
_engine.frame.viewport.originX = _viewport->x;
_engine.frame.viewport.originY = _viewport->y;
_engine.frame.viewport.width = _viewport->width;
_engine.frame.viewport.height = _viewport->height;
_engine.frame.viewport.znear = 0.0f;
_engine.frame.viewport.zfar = 1.0f;
_engine.frame.output_size.x = _viewport->width;
_engine.frame.output_size.y = _viewport->height;
_engine.frame.output_size.z = 1.0f / _viewport->width;
_engine.frame.output_size.w = 1.0f / _viewport->height;
if (resize_render_targets) {
[self _updateRenderTargets];
}
[self _updateHistory];
if (_format == RPixelFormatBGRA8Unorm || _format == RPixelFormatBGRX8Unorm) {
id<MTLTexture> tex = _engine.frame.texture[0].view;
[tex replaceRegion:MTLRegionMake2D(0, 0, (NSUInteger)_size.width, (NSUInteger)_size.height)
mipmapLevel:0 withBytes:src
bytesPerRow:(NSUInteger)(4 * _size.width)];
}
else {
void *dst = _pixels.contents;
size_t len = (size_t)(_bpp * _size.width);
assert(len <= pitch); // the length can't be larger?
if (len < pitch) {
for (int i = 0; i < _size.height; i++) {
memcpy(dst, src, len);
dst += len;
src += pitch;
}
}
else {
memcpy(dst, src, _pixels.length);
}
[_pixels didModifyRange:NSMakeRange(0, _pixels.length)];
_pixelsDirty = YES;
}
}
- (void)_initTexture:(texture_t *)t withDescriptor:(MTLTextureDescriptor *)td
{
STRUCT_ASSIGN(t->view, [_context.device newTextureWithDescriptor:td]);
t->size_data.x = td.width;
t->size_data.y = td.height;
t->size_data.z = 1.0f / td.width;
t->size_data.w = 1.0f / td.height;
}
- (void)_initHistory
{
MTLTextureDescriptor *td = [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:MTLPixelFormatBGRA8Unorm
width:(NSUInteger)_size.width
height:(NSUInteger)_size.height
mipmapped:false];
td.usage = MTLTextureUsageShaderRead | MTLTextureUsageShaderWrite | MTLTextureUsageRenderTarget;
for (int i = 0; i < _shader->history_size + 1; i++) {
[self _initTexture:&_engine.frame.texture[i] withDescriptor:td];
}
init_history = NO;
}
typedef struct vertex
{
simd_float4 pos;
simd_float2 tex;
} vertex_t;
static vertex_t vertex_bytes[] = {
{{0, 1, 0, 1}, {0, 1}},
{{1, 1, 0, 1}, {1, 1}},
{{0, 0, 0, 1}, {0, 0}},
{{1, 0, 0, 1}, {1, 0}},
};
- (void)drawWithEncoder:(id<MTLRenderCommandEncoder>)rce
{
if (_texture) {
[rce setViewport:_engine.frame.viewport];
[rce setVertexBytes:&_v length:sizeof(_v) atIndex:BufferIndexPositions];
[rce setFragmentTexture:_texture atIndex:TextureIndexColor];
[rce drawPrimitives:MTLPrimitiveTypeTriangleStrip vertexStart:0 vertexCount:4];
}
}
- (void)drawWithContext:(Context *)ctx
{
_texture = _engine.frame.texture[0].view;
[self _convertFormat];
if (!_shader || _shader->passes == 0) {
return;
}
for (unsigned i = 0; i < _shader->passes; i++) {
if (_shader->pass[i].feedback) {
texture_t tmp = _engine.pass[i].feedback;
_engine.pass[i].feedback = _engine.pass[i].rt;
_engine.pass[i].rt = tmp;
}
}
id<MTLCommandBuffer> cb = ctx.commandBuffer;
MTLRenderPassDescriptor *rpd = [MTLRenderPassDescriptor new];
rpd.colorAttachments[0].clearColor = MTLClearColorMake(0, 0, 0, 1.0);
rpd.colorAttachments[0].loadAction = MTLLoadActionClear;
rpd.colorAttachments[0].storeAction = MTLStoreActionStore;
BOOL firstPass = YES;
for (unsigned i = 0; i < _shader->passes; i++) {
BOOL backBuffer = (_engine.pass[i].rt.view == nil);
if (backBuffer) {
rpd.colorAttachments[0].texture = _context.nextDrawable.texture;
}
else {
rpd.colorAttachments[0].texture = _engine.pass[i].rt.view;
}
id<MTLRenderCommandEncoder> rce = [cb renderCommandEncoderWithDescriptor:rpd];
if (firstPass) {
firstPass = NO;
} else {
[rce waitForFence:_fence beforeStages:MTLRenderStageVertex];
}
[rce setRenderPipelineState:_engine.pass[i]._state];
_engine.pass[i].frame_count = (uint32_t)_frameCount;
if (_shader->pass[i].frame_count_mod)
_engine.pass[i].frame_count %= _shader->pass[i].frame_count_mod;
for (unsigned j = 0; j < SLANG_CBUFFER_MAX; j++) {
id<MTLBuffer> buffer = _engine.pass[i].buffers[j];
cbuffer_sem_t *buffer_sem = &_engine.pass[i].semantics.cbuffers[j];
if (buffer_sem->stage_mask && buffer_sem->uniforms) {
void *data = buffer.contents;
uniform_sem_t *uniform = buffer_sem->uniforms;
while (uniform->size) {
if (uniform->data)
memcpy((uint8_t *)data + uniform->offset, uniform->data, uniform->size);
uniform++;
}
if (buffer_sem->stage_mask & SLANG_STAGE_VERTEX_MASK)
[rce setVertexBuffer:buffer offset:0 atIndex:buffer_sem->binding];
if (buffer_sem->stage_mask & SLANG_STAGE_FRAGMENT_MASK)
[rce setFragmentBuffer:buffer offset:0 atIndex:buffer_sem->binding];
[buffer didModifyRange:NSMakeRange(0, buffer.length)];
}
}
__unsafe_unretained id<MTLTexture> textures[SLANG_NUM_BINDINGS] = {NULL};
id<MTLSamplerState> samplers[SLANG_NUM_BINDINGS] = {NULL};
texture_sem_t *texture_sem = _engine.pass[i].semantics.textures;
while (texture_sem->stage_mask) {
int binding = texture_sem->binding;
id<MTLTexture> tex = (__bridge id<MTLTexture>)*(void **)texture_sem->texture_data;
textures[binding] = tex;
samplers[binding] = _samplers[texture_sem->filter][texture_sem->wrap];
texture_sem++;
}
if (backBuffer) {
[rce setViewport:_engine.frame.viewport];
}
else {
[rce setViewport:_engine.pass[i].viewport];
}
[rce setFragmentTextures:textures withRange:NSMakeRange(0, SLANG_NUM_BINDINGS)];
[rce setFragmentSamplerStates:samplers withRange:NSMakeRange(0, SLANG_NUM_BINDINGS)];
[rce setVertexBytes:vertex_bytes length:sizeof(vertex_bytes) atIndex:4];
[rce drawPrimitives:MTLPrimitiveTypeTriangleStrip vertexStart:0 vertexCount:4];
[rce updateFence:_fence afterStages:MTLRenderStageFragment];
[rce endEncoding];
_texture = _engine.pass[i].rt.view;
}
if (_texture == nil) {
_drawState = ViewDrawStateContext;
} else {
_drawState = ViewDrawStateAll;
}
}
- (void)_updateRenderTargets
{
if (!_shader || !resize_render_targets) return;
// release existing targets
for (int i = 0; i < _shader->passes; i++) {
STRUCT_ASSIGN(_engine.pass[i].rt.view, nil);
STRUCT_ASSIGN(_engine.pass[i].feedback.view, nil);
memset(&_engine.pass[i].rt, 0, sizeof(_engine.pass[i].rt));
memset(&_engine.pass[i].feedback, 0, sizeof(_engine.pass[i].feedback));
}
NSUInteger width = (NSUInteger)_size.width, height = (NSUInteger)_size.height;
for (unsigned i = 0; i < _shader->passes; i++) {
struct video_shader_pass *shader_pass = &_shader->pass[i];
if (shader_pass->fbo.valid) {
switch (shader_pass->fbo.type_x) {
case RARCH_SCALE_INPUT:
width *= shader_pass->fbo.scale_x;
break;
case RARCH_SCALE_VIEWPORT:
width = (NSUInteger)(_viewport->width * shader_pass->fbo.scale_x);
break;
case RARCH_SCALE_ABSOLUTE:
width = shader_pass->fbo.abs_x;
break;
default:
break;
}
if (!width)
width = _viewport->width;
switch (shader_pass->fbo.type_y) {
case RARCH_SCALE_INPUT:
height *= shader_pass->fbo.scale_y;
break;
case RARCH_SCALE_VIEWPORT:
height = (NSUInteger)(_viewport->height * shader_pass->fbo.scale_y);
break;
case RARCH_SCALE_ABSOLUTE:
height = shader_pass->fbo.abs_y;
break;
default:
break;
}
if (!height)
height = _viewport->height;
}
else if (i == (_shader->passes - 1)) {
width = _viewport->width;
height = _viewport->height;
}
RARCH_LOG("[Metal]: Updating framebuffer size %u x %u.\n", width, height);
MTLPixelFormat fmt = SelectOptimalPixelFormat(glslang_format_to_metal(_engine.pass[i].semantics.format));
if ((i != (_shader->passes - 1)) ||
(width != _viewport->width) || (height != _viewport->height) ||
fmt != MTLPixelFormatBGRA8Unorm)
{
_engine.pass[i].viewport.width = width;
_engine.pass[i].viewport.height = height;
_engine.pass[i].viewport.znear = 0.0;
_engine.pass[i].viewport.zfar = 1.0;
MTLTextureDescriptor *td = [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:fmt
width:width
height:height
mipmapped:false];
td.usage = MTLTextureUsageShaderRead | MTLTextureUsageRenderTarget;
[self _initTexture:&_engine.pass[i].rt withDescriptor:td];
if (shader_pass->feedback) {
[self _initTexture:&_engine.pass[i].feedback withDescriptor:td];
}
}
else {
_engine.pass[i].rt.size_data.x = width;
_engine.pass[i].rt.size_data.y = height;
_engine.pass[i].rt.size_data.z = 1.0f / width;
_engine.pass[i].rt.size_data.w = 1.0f / height;
}
}
resize_render_targets = NO;
}
- (void)_freeVideoShader:(struct video_shader *)shader
{
if (!shader)
return;
for (int i = 0; i < GFX_MAX_SHADERS; i++) {
STRUCT_ASSIGN(_engine.pass[i].rt.view, nil);
STRUCT_ASSIGN(_engine.pass[i].feedback.view, nil);
memset(&_engine.pass[i].rt, 0, sizeof(_engine.pass[i].rt));
memset(&_engine.pass[i].feedback, 0, sizeof(_engine.pass[i].feedback));
STRUCT_ASSIGN(_engine.pass[i]._state, nil);
for (unsigned j = 0; j < SLANG_CBUFFER_MAX; j++) {
STRUCT_ASSIGN(_engine.pass[i].buffers[j], nil);
}
}
for (int i = 0; i < GFX_MAX_TEXTURES; i++) {
STRUCT_ASSIGN(_engine.luts[i].view, nil);
}
free(shader);
_fence = nil;
}
- (BOOL)setShaderFromPath:(NSString *)path
{
[self _freeVideoShader:_shader];
_shader = nil;
config_file_t *conf = config_file_new(path.UTF8String);
struct video_shader *shader = (struct video_shader *)calloc(1, sizeof(*shader));
@try {
if (!video_shader_read_conf_cgp(conf, shader))
return NO;
video_shader_resolve_relative(shader, path.UTF8String);
texture_t *source = &_engine.frame.texture[0];
for (unsigned i = 0; i < shader->passes; source = &_engine.pass[i++].rt) {
/* clang-format off */
semantics_map_t semantics_map = {
{
/* Original */
{&_engine.frame.texture[0].view, 0,
&_engine.frame.texture[0].size_data, 0},
/* Source */
{&source->view, 0,
&source->size_data, 0},
/* OriginalHistory */
{&_engine.frame.texture[0].view, sizeof(*_engine.frame.texture),
&_engine.frame.texture[0].size_data, sizeof(*_engine.frame.texture)},
/* PassOutput */
{&_engine.pass[0].rt.view, sizeof(*_engine.pass),
&_engine.pass[0].rt.size_data, sizeof(*_engine.pass)},
/* PassFeedback */
{&_engine.pass[0].feedback.view, sizeof(*_engine.pass),
&_engine.pass[0].feedback.size_data, sizeof(*_engine.pass)},
/* User */
{&_engine.luts[0].view, sizeof(*_engine.luts),
&_engine.luts[0].size_data, sizeof(*_engine.luts)},
},
{
&_engine.mvp, /* MVP */
&_engine.pass[i].rt.size_data, /* OutputSize */
&_engine.frame.output_size, /* FinalViewportSize */
&_engine.pass[i].frame_count, /* FrameCount */
}
};
/* clang-format on */
if (!slang_process(shader, i, RARCH_SHADER_METAL, 20000, &semantics_map, &_engine.pass[i].semantics))
return NO;
#ifdef DEBUG
bool save_msl = true;
#else
bool save_msl = false;
#endif
NSString *vs_src = [NSString stringWithUTF8String:shader->pass[i].source.string.vertex];
NSString *fs_src = [NSString stringWithUTF8String:shader->pass[i].source.string.fragment];
// vertex descriptor
@try {
MTLVertexDescriptor *vd = [MTLVertexDescriptor new];
vd.attributes[0].offset = offsetof(vertex_t, pos);
vd.attributes[0].format = MTLVertexFormatFloat4;
vd.attributes[0].bufferIndex = 4;
vd.attributes[1].offset = offsetof(vertex_t, tex);
vd.attributes[1].format = MTLVertexFormatFloat2;
vd.attributes[1].bufferIndex = 4;
vd.layouts[4].stride = sizeof(vertex_t);
vd.layouts[4].stepFunction = MTLVertexStepFunctionPerVertex;
MTLRenderPipelineDescriptor *psd = [MTLRenderPipelineDescriptor new];
psd.label = [NSString stringWithFormat:@"pass %d", i];
MTLRenderPipelineColorAttachmentDescriptor *ca = psd.colorAttachments[0];
ca.pixelFormat = SelectOptimalPixelFormat(glslang_format_to_metal(_engine.pass[i].semantics.format));
// TODO(sgc): confirm we never need blending for render passes
ca.blendingEnabled = NO;
ca.sourceAlphaBlendFactor = MTLBlendFactorSourceAlpha;
ca.sourceRGBBlendFactor = MTLBlendFactorSourceAlpha;
ca.destinationAlphaBlendFactor = MTLBlendFactorOneMinusSourceAlpha;
ca.destinationRGBBlendFactor = MTLBlendFactorOneMinusSourceAlpha;
psd.sampleCount = 1;
psd.vertexDescriptor = vd;
NSError *err;
id<MTLLibrary> lib = [_context.device newLibraryWithSource:vs_src options:nil error:&err];
if (err != nil) {
if (lib == nil) {
save_msl = true;
RARCH_ERR("Metal]: unable to compile vertex shader: %s\n", err.localizedDescription.UTF8String);
return NO;
}
#if DEBUG
RARCH_WARN("[Metal]: warnings compiling vertex shader: %s\n", err.localizedDescription.UTF8String);
#endif
}
psd.vertexFunction = [lib newFunctionWithName:@"main0"];
lib = [_context.device newLibraryWithSource:fs_src options:nil error:&err];
if (err != nil) {
if (lib == nil) {
save_msl = true;
RARCH_ERR("Metal]: unable to compile fragment shader: %s\n", err.localizedDescription.UTF8String);
return NO;
}
#if DEBUG
RARCH_WARN("[Metal]: warnings compiling fragment shader: %s\n", err.localizedDescription.UTF8String);
#endif
}
psd.fragmentFunction = [lib newFunctionWithName:@"main0"];
STRUCT_ASSIGN(_engine.pass[i]._state,
[_context.device newRenderPipelineStateWithDescriptor:psd error:&err]);
if (err != nil) {
save_msl = true;
RARCH_ERR("error creating pipeline state: %s", err.localizedDescription.UTF8String);
return NO;
}
for (unsigned j = 0; j < SLANG_CBUFFER_MAX; j++) {
unsigned int size = _engine.pass[i].semantics.cbuffers[j].size;
if (size == 0) {
continue;
}
id<MTLBuffer> buf = [_context.device newBufferWithLength:size options:MTLResourceStorageModeManaged];
STRUCT_ASSIGN(_engine.pass[i].buffers[j], buf);
}
} @finally {
if (save_msl) {
RARCH_LOG("[Metal]: saving metal shader files\n");
NSError *err = nil;
NSString *basePath = [[NSString stringWithUTF8String:shader->pass[i].source.path] stringByDeletingPathExtension];
[vs_src writeToFile:[basePath stringByAppendingPathExtension:@"vs.metal"]
atomically:NO
encoding:NSStringEncodingConversionAllowLossy
error:&err];
if (err != nil) {
RARCH_ERR("[Metal]: unable to save vertex shader source: %s\n", err.localizedDescription.UTF8String);
}
err = nil;
[fs_src writeToFile:[basePath stringByAppendingPathExtension:@"fs.metal"]
atomically:NO
encoding:NSStringEncodingConversionAllowLossy
error:&err];
if (err != nil) {
RARCH_ERR("[Metal]: unable to save fragment shader source: %s\n", err.localizedDescription.UTF8String);
}
}
free(shader->pass[i].source.string.vertex);
free(shader->pass[i].source.string.fragment);
shader->pass[i].source.string.vertex = NULL;
shader->pass[i].source.string.fragment = NULL;
}
}
for (unsigned i = 0; i < shader->luts; i++) {
struct texture_image image = {0};
image.supports_rgba = true;
if (!image_texture_load(&image, shader->lut[i].path))
return NO;
MTLTextureDescriptor *td = [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:MTLPixelFormatRGBA8Unorm
width:image.width
height:image.height
mipmapped:shader->lut[i].mipmap];
td.usage = MTLTextureUsageShaderRead;
[self _initTexture:&_engine.luts[i] withDescriptor:td];
[_engine.luts[i].view replaceRegion:MTLRegionMake2D(0, 0, image.width, image.height)
mipmapLevel:0 withBytes:image.pixels
bytesPerRow:4 * image.width];
// TODO(sgc): generate mip maps
image_texture_free(&image);
}
video_shader_resolve_current_parameters(conf, shader);
_shader = shader;
shader = nil;
_fence = [_context.device newFence];
}
@finally {
if (shader) {
[self _freeVideoShader:shader];
}
if (conf) {
config_file_free(conf);
conf = nil;
}
}
resize_render_targets = YES;
init_history = YES;
return YES;
}
@end
MTLPixelFormat glslang_format_to_metal(glslang_format fmt)
{
#undef FMT2
#define FMT2(x,y) case SLANG_FORMAT_##x: return MTLPixelFormat##y
switch (fmt)
{
FMT2(R8_UNORM, R8Unorm);
FMT2(R8_SINT, R8Sint);
FMT2(R8_UINT, R8Uint);
FMT2(R8G8_UNORM, RG8Unorm);
FMT2(R8G8_SINT, RG8Sint);
FMT2(R8G8_UINT, RG8Uint);
FMT2(R8G8B8A8_UNORM, RGBA8Unorm);
FMT2(R8G8B8A8_SINT, RGBA8Sint);
FMT2(R8G8B8A8_UINT, RGBA8Uint);
FMT2(R8G8B8A8_SRGB, RGBA8Unorm_sRGB);
FMT2(A2B10G10R10_UNORM_PACK32, RGB10A2Unorm);
FMT2(A2B10G10R10_UINT_PACK32, RGB10A2Uint);
FMT2(R16_UINT, R16Uint);
FMT2(R16_SINT, R16Sint);
FMT2(R16_SFLOAT, R16Float);
FMT2(R16G16_UINT, RG16Uint);
FMT2(R16G16_SINT, RG16Sint);
FMT2(R16G16_SFLOAT, RG16Float);
FMT2(R16G16B16A16_UINT, RGBA16Uint);
FMT2(R16G16B16A16_SINT, RGBA16Sint);
FMT2(R16G16B16A16_SFLOAT, RGBA16Float);
FMT2(R32_UINT, R32Uint);
FMT2(R32_SINT, R32Sint);
FMT2(R32_SFLOAT, R32Float);
FMT2(R32G32_UINT, RG32Uint);
FMT2(R32G32_SINT, RG32Sint);
FMT2(R32G32_SFLOAT, RG32Float);
FMT2(R32G32B32A32_UINT, RGBA32Uint);
FMT2(R32G32B32A32_SINT, RGBA32Sint);
FMT2(R32G32B32A32_SFLOAT, RGBA32Float);
case SLANG_FORMAT_UNKNOWN:
default:
break;
}
#undef FMT2
return MTLPixelFormatInvalid;
}
MTLPixelFormat SelectOptimalPixelFormat(MTLPixelFormat fmt)
{
switch (fmt)
{
case MTLPixelFormatRGBA8Unorm:
return MTLPixelFormatBGRA8Unorm;
case MTLPixelFormatRGBA8Unorm_sRGB:
return MTLPixelFormatBGRA8Unorm_sRGB;
default:
return fmt;
}
}
Reference in New Issue View Git Blame Copy Permalink