Skip to content
This repository has been archived by the owner on Aug 27, 2022. It is now read-only.

Commit

Permalink
Browse files Browse the repository at this point in the history
8238285: Lanai: java/awt/image/DrawImage tests fail
  • Loading branch information
@dekonoplyov
dekonoplyov authored and Alexey Ushakov committed Nov 27, 2020
1 parent dcff176 commit 77dca42
Show file tree
Hide file tree
Showing 2 changed files with 109 additions and 157 deletions.
33 changes: 33 additions & 0 deletions src/java.desktop/macosx/native/libawt_lwawt/awt/shaders.metal
View file
Expand Up @@ -667,6 +667,39 @@ kernel void stencil2tex(const device uchar *imageBuffer [[buffer(0)]],
outputBuffer[gid] = uchar4(p, p, p, p);
}

// work item deals with 4 byte pixel
// assuming that data is aligned
kernel void bgr_to_rgba(const device uchar *imageBuffer [[buffer(0)]],
device uchar *outputBuffer [[buffer(1)]],
uint gid [[thread_position_in_grid]])
{
outputBuffer[4 * gid] = imageBuffer[4 * gid + 2]; // r
outputBuffer[4 * gid + 1] = imageBuffer[4 * gid + 1]; // g
outputBuffer[4 * gid + 2] = imageBuffer[4 * gid]; // b
outputBuffer[4 * gid + 3] = 255; // a
}

kernel void xrgb_to_rgba(const device uchar *imageBuffer [[buffer(0)]],
device uchar *outputBuffer [[buffer(1)]],
uint gid [[thread_position_in_grid]])
{
outputBuffer[4 * gid] = imageBuffer[4 * gid + 1]; // r
outputBuffer[4 * gid + 1] = imageBuffer[4 * gid + 2]; // g
outputBuffer[4 * gid + 2] = imageBuffer[4 * gid + 3]; // b
outputBuffer[4 * gid + 3] = imageBuffer[4 * gid]; // a
}


kernel void xbgr_to_rgba(const device uchar *imageBuffer [[buffer(0)]],
device uchar *outputBuffer [[buffer(1)]],
uint gid [[thread_position_in_grid]])
{
outputBuffer[4 * gid] = imageBuffer[4 * gid + 3]; // r
outputBuffer[4 * gid + 1] = imageBuffer[4 * gid + 2]; // g
outputBuffer[4 * gid + 2] = imageBuffer[4 * gid + 1]; // b
outputBuffer[4 * gid + 3] = imageBuffer[4 * gid]; // a
}

// ----------------------------------------------------------------------------
// Shaders for rendering in XOR Mode
// ----------------------------------------------------------------------------
Expand Down
233 changes: 76 additions & 157 deletions src/java.desktop/macosx/native/libawt_lwawt/java2d/metal/MTLBlitLoops.m
View file
Expand Up @@ -49,26 +49,12 @@
//#define DEBUG_BLIT

typedef struct {
MTLPixelFormat format;
MTLPixelFormat format; // Consider deleting this field, since it's always MTLPixelFormatBGRA8Unorm
jboolean hasAlpha;
jboolean isPremult;
const uint8_t * permuteMap;
NSString* swizzleKernel;
} MTLRasterFormatInfo;

// 0 denotes the alpha channel, 1 the red channel, 2 the green channel, and 3 the blue channel.
const uint8_t permuteMap_rgbx[4] = { 1, 2, 3, 0 };
const uint8_t permuteMap_bgrx[4] = { 3, 2, 1, 0 };

static uint8_t revertPerm(const uint8_t * perm, uint8_t pos) {
for (int c = 0; c < 4; ++c) {
if (perm[c] == pos)
return c;
}
return -1;
}

#define uint2swizzle(channel) (channel == 0 ? MTLTextureSwizzleAlpha : (channel == 1 ? MTLTextureSwizzleRed : (channel == 2 ? MTLTextureSwizzleGreen : (channel == 3 ? MTLTextureSwizzleBlue : MTLTextureSwizzleZero))))

/**
* This table contains the "pixel formats" for all system memory surfaces
* that Metal is capable of handling, indexed by the "PF_" constants defined
Expand All @@ -77,12 +63,12 @@ static uint8_t revertPerm(const uint8_t * perm, uint8_t pos) {
* an Metal surface
*/
MTLRasterFormatInfo RasterFormatInfos[] = {
{ MTLPixelFormatBGRA8Unorm, 1, 0, NULL }, /* 0 - IntArgb */ // Argb (in java notation)
{ MTLPixelFormatBGRA8Unorm, 1, 1, NULL }, /* 1 - IntArgbPre */
{ MTLPixelFormatBGRA8Unorm, 0, 1, NULL }, /* 2 - IntRgb */ // xrgb
{ MTLPixelFormatBGRA8Unorm, 0, 1, permuteMap_rgbx }, /* 3 - IntRgbx */
{ MTLPixelFormatRGBA8Unorm, 0, 1, NULL }, /* 4 - IntBgr */ // xbgr
{ MTLPixelFormatBGRA8Unorm, 0, 1, permuteMap_bgrx }, /* 5 - IntBgrx */
{ MTLPixelFormatBGRA8Unorm, 1, 0, nil }, /* 0 - IntArgb */ // Argb (in java notation)
{ MTLPixelFormatBGRA8Unorm, 1, 1, nil }, /* 1 - IntArgbPre */
{ MTLPixelFormatBGRA8Unorm, 0, 1, nil }, /* 2 - IntRgb */
{ MTLPixelFormatBGRA8Unorm, 0, 1, @"xrgb_to_rgba" }, /* 3 - IntRgbx */
{ MTLPixelFormatBGRA8Unorm, 0, 1, @"bgr_to_rgba" }, /* 4 - IntBgr */
{ MTLPixelFormatBGRA8Unorm, 0, 1, @"xbgr_to_rgba" }, /* 5 - IntBgrx */

// TODO: support 2-byte formats
// { GL_BGRA, GL_UNSIGNED_SHORT_1_5_5_5_REV,
Expand Down Expand Up @@ -169,94 +155,55 @@ void drawTex2Tex(MTLContext *mtlc,
[encoder drawPrimitives:MTLPrimitiveTypeTriangle vertexStart:0 vertexCount:6];
}

static
id<MTLTexture> replaceTextureRegion(MTLContext *mtlc, id<MTLTexture> dest, const SurfaceDataRasInfo * srcInfo, const MTLRasterFormatInfo * rfi, int dx1, int dy1, int dx2, int dy2) {
static void
replaceTextureRegion(MTLContext *mtlc, id<MTLTexture> dest, const SurfaceDataRasInfo *srcInfo,
const MTLRasterFormatInfo *rfi,
int dx1, int dy1, int dx2, int dy2) {
const int dw = dx2 - dx1;
const int dh = dy2 - dy1;

const void * raster = srcInfo->rasBase;
const void *raster = srcInfo->rasBase;
raster += srcInfo->bounds.y1*srcInfo->scanStride + srcInfo->bounds.x1*srcInfo->pixelStride;

id<MTLTexture> result = nil;
if (rfi->permuteMap != NULL) {
#if defined(__MAC_10_15) && __MAC_OS_X_VERSION_MAX_ALLOWED >= __MAC_10_15
if (@available(macOS 10.15, *)) {
@autoreleasepool {
const uint8_t swzRed = revertPerm(rfi->permuteMap, 1);
const uint8_t swzGreen = revertPerm(rfi->permuteMap, 2);
const uint8_t swzBlue = revertPerm(rfi->permuteMap, 3);
const uint8_t swzAlpha = revertPerm(rfi->permuteMap, 0);
MTLTextureSwizzleChannels swizzle = MTLTextureSwizzleChannelsMake(
uint2swizzle(swzRed),
uint2swizzle(swzGreen),
uint2swizzle(swzBlue),
rfi->hasAlpha ? uint2swizzle(swzAlpha) : MTLTextureSwizzleOne
);
result = [dest
newTextureViewWithPixelFormat:MTLPixelFormatBGRA8Unorm
textureType:MTLTextureType2D
levels:NSMakeRange(0, 1) slices:NSMakeRange(0, 1)
swizzle:swizzle];
J2dTraceLn5(J2D_TRACE_VERBOSE, "replaceTextureRegion [use swizzle for pooled]: %d, %d, %d, %d, hasA=%d",
swizzle.red, swizzle.green, swizzle.blue, swizzle.alpha, rfi->hasAlpha);
}
} else
#endif // __MAC_10_15 && __MAC_OS_X_VERSION_MAX_ALLOWED >= __MAC_10_15
{
// perform raster conversion
// invoked only from rq-thread, so use static buffers
// but it's better to use thread-local buffers (or special buffer manager)
const int destRasterSize = dw*dh*4;

static int bufferSize = 0;
static void * buffer = NULL;
if (buffer == NULL || bufferSize < destRasterSize) {
bufferSize = destRasterSize;
buffer = realloc(buffer, bufferSize);
}
if (buffer == NULL) {
J2dTraceLn1(J2D_TRACE_ERROR, "replaceTextureRegion: can't alloc buffer for raster conversion, size=%d", bufferSize);
bufferSize = 0;
return nil;
}
vImage_Buffer srcBuf;
srcBuf.height = dh;
srcBuf.width = dw;
srcBuf.rowBytes = srcInfo->scanStride;
srcBuf.data = raster;

vImage_Buffer destBuf;
destBuf.height = dh;
destBuf.width = dw;
destBuf.rowBytes = dw*4;
destBuf.data = buffer;

vImagePermuteChannels_ARGB8888(&srcBuf, &destBuf, rfi->permuteMap, kvImageNoFlags);
raster = buffer;

J2dTraceLn5(J2D_TRACE_VERBOSE, "replaceTextureRegion [use conversion]: %d, %d, %d, %d, hasA=%d",
rfi->permuteMap[0], rfi->permuteMap[1], rfi->permuteMap[2], rfi->permuteMap[3], rfi->hasAlpha);
}
}

MTLRegion region = MTLRegionMake2D(dx1, dy1, dw, dh);
if (result != nil)
dest = result;

@autoreleasepool {
id <MTLBlitCommandEncoder> blitEncoder = [mtlc.encoderManager createBlitEncoder];

J2dTraceLn4(J2D_TRACE_VERBOSE, "replaceTextureRegion src (dw, dh) : [%d, %d] dest (dx1, dy1) =[%d, %d]",
dw, dh, dx1, dy1);
// NOTE: we might want to fill alpha channel when !rfi->hasAlpha
id<MTLBuffer> buff = [mtlc.device newBufferWithBytes:raster length:srcInfo->scanStride * dh options:MTLResourceStorageModeManaged];
if (rfi->swizzleKernel != nil) {
id <MTLBuffer> swizzled = [mtlc.device newBufferWithLength:srcInfo->scanStride * dh options:MTLResourceStorageModeManaged];

id <MTLBuffer> buff = [[mtlc.device newBufferWithBytes:raster length:srcInfo->scanStride * dh options:MTLResourceStorageModeManaged] autorelease];
// this should be cheap, since data is already on GPU
id<MTLCommandBuffer> cb = [mtlc createCommandBuffer];
id<MTLComputeCommandEncoder> computeEncoder = [cb computeCommandEncoder];
id<MTLComputePipelineState> computePipelineState = [mtlc.pipelineStateStorage
getComputePipelineState:rfi->swizzleKernel];
[computeEncoder setComputePipelineState:computePipelineState];

[computeEncoder setBuffer:buff offset:0 atIndex:0];
[computeEncoder setBuffer:swizzled offset:0 atIndex:1];

NSUInteger threadGroupSize = computePipelineState.maxTotalThreadsPerThreadgroup;
NSUInteger pixelCount = buff.length / srcInfo->pixelStride;
MTLSize threadsPerGroup = MTLSizeMake(threadGroupSize, 1, 1);
MTLSize threadGroups = MTLSizeMake((pixelCount + threadGroupSize - 1) / threadGroupSize,
1, 1);
[computeEncoder dispatchThreadgroups:threadGroups
threadsPerThreadgroup:threadsPerGroup];
[computeEncoder endEncoding];
[cb commit];

buff = swizzled;
}

id<MTLBlitCommandEncoder> blitEncoder = [mtlc.encoderManager createBlitEncoder];
[blitEncoder copyFromBuffer:buff
sourceOffset:0 sourceBytesPerRow:srcInfo->scanStride sourceBytesPerImage:srcInfo->scanStride * dh sourceSize:MTLSizeMake(dw, dh, 1)
toTexture:dest destinationSlice:0 destinationLevel:0 destinationOrigin:MTLOriginMake(dx1, dy1, 0)];
sourceOffset:0 sourceBytesPerRow:srcInfo->scanStride
sourceBytesPerImage:srcInfo->scanStride * dh sourceSize:MTLSizeMake(dw, dh, 1)
toTexture:dest
destinationSlice:0 destinationLevel:0 destinationOrigin:MTLOriginMake(dx1, dy1, 0)];
[blitEncoder endEncoding];
}

return result;
}

/**
Expand All @@ -283,10 +230,11 @@ void drawTex2Tex(MTLContext *mtlc,
[[mtlc getCommandBufferWrapper] registerPooledTexture:texHandle];

id<MTLTexture> texBuff = texHandle.texture;
id<MTLTexture> swizzledTexture = replaceTextureRegion(mtlc, texBuff, srcInfo, rfi, 0, 0, sw, sh);
replaceTextureRegion(mtlc, texBuff, srcInfo, rfi, 0, 0, sw, sh);
// TODO: useBlitEncoder is always false, remove dead code
if (useBlitEncoder) {
id <MTLBlitCommandEncoder> blitEncoder = [mtlc.encoderManager createBlitEncoder];
[blitEncoder copyFromTexture:swizzledTexture != nil ? swizzledTexture : texBuff
[blitEncoder copyFromTexture:texBuff
sourceSlice:0
sourceLevel:0
sourceOrigin:MTLOriginMake(0, 0, 0)
Expand All @@ -297,13 +245,9 @@ void drawTex2Tex(MTLContext *mtlc,
destinationOrigin:MTLOriginMake(dx1, dy1, 0)];
[blitEncoder endEncoding];
} else {
drawTex2Tex(mtlc, swizzledTexture != nil ? swizzledTexture : texBuff, dest, !rfi->hasAlpha, bmtlsdOps->isOpaque, hint,
drawTex2Tex(mtlc, texBuff, dest, !rfi->hasAlpha, bmtlsdOps->isOpaque, hint,
0, 0, sw, sh, dx1, dy1, dx2, dy2);
}

if (swizzledTexture != nil) {
[swizzledTexture release];
}
}

static
Expand Down Expand Up @@ -737,25 +681,20 @@ void copyFromMTLBuffer(void *pDst, id<MTLBuffer> srcBuf, jint offset, jint len,
const int byteLength = width * height * 4; // NOTE: assume that src format is MTLPixelFormatBGRA8Unorm

// Create MTLBuffer (or use static)
MTLRasterFormatInfo rfi = RasterFormatInfos[dsttype];
const jboolean directCopy = rfi.permuteMap == NULL;

id<MTLBuffer> mtlbuf;
#ifdef USE_STATIC_BUFFER
if (directCopy) {
// NOTE: theoretically we can use newBufferWithBytesNoCopy, but pDst must be allocated with special API
// mtlbuf = [mtlc.device
// newBufferWithBytesNoCopy:pDst
// length:(NSUInteger) srcLength
// options:MTLResourceCPUCacheModeDefaultCache
// deallocator:nil];
//
// see https://developer.apple.com/documentation/metal/mtldevice/1433382-newbufferwithbytesnocopy?language=objc
//
// The storage allocation of the returned new MTLBuffer object is the same as the pointer input value.
// The existing memory allocation must be covered by a single VM region, typically allocated with vm_allocate or mmap.
// Memory allocated by malloc is specifically disallowed.
}
// NOTE: theoretically we can use newBufferWithBytesNoCopy, but pDst must be allocated with special API
// mtlbuf = [mtlc.device
// newBufferWithBytesNoCopy:pDst
// length:(NSUInteger) srcLength
// options:MTLResourceCPUCacheModeDefaultCache
// deallocator:nil];
//
// see https://developer.apple.com/documentation/metal/mtldevice/1433382-newbufferwithbytesnocopy?language=objc
//
// The storage allocation of the returned new MTLBuffer object is the same as the pointer input value.
// The existing memory allocation must be covered by a single VM region, typically allocated with vm_allocate or mmap.
// Memory allocated by malloc is specifically disallowed.

static id<MTLBuffer> mtlIntermediateBuffer = nil; // need to reimplement with MTLBufferManager
if (mtlIntermediateBuffer == nil || mtlIntermediateBuffer.length < srcLength) {
Expand Down Expand Up @@ -795,43 +734,23 @@ void copyFromMTLBuffer(void *pDst, id<MTLBuffer> srcBuf, jint offset, jint len,

// Perform conversion if necessary
BOOL convertFromPre = !RasterFormatInfos[dsttype].isPremult && !srcOps->isOpaque;
if (directCopy) {
if ((dstInfo.scanStride == width * dstInfo.pixelStride) &&
(height == (dstInfo.bounds.y2 - dstInfo.bounds.y1))) {
// mtlbuf.contents have same dimensions as of pDst
copyFromMTLBuffer(pDst, mtlbuf, 0, byteLength, convertFromPre);
} else {
// mtlbuf.contents have smaller dimensions than pDst
// copy each row from mtlbuf.contents at appropriate position in pDst
// Note : pDst is already addjusted for offsets using PtrAddBytes above

int rowSize = width * dstInfo.pixelStride;
for (int y = 0; y < height; y++) {
copyFromMTLBuffer(pDst, mtlbuf, y * rowSize, rowSize, convertFromPre);
pDst = PtrAddBytes(pDst, dstInfo.scanStride);
}
}

if ((dstInfo.scanStride == width * dstInfo.pixelStride) &&
(height == (dstInfo.bounds.y2 - dstInfo.bounds.y1))) {
// mtlbuf.contents have same dimensions as of pDst
copyFromMTLBuffer(pDst, mtlbuf, 0, byteLength, convertFromPre);
} else {
J2dTraceLn6(J2D_TRACE_VERBOSE,"MTLBlitLoops_SurfaceToSwBlit: dsttype=%d, raster conversion will be performed, dest rfi: %d, %d, %d, %d, hasA=%d",
dsttype, rfi.permuteMap[0], rfi.permuteMap[1], rfi.permuteMap[2], rfi.permuteMap[3], rfi.hasAlpha);

// perform raster conversion: mtlIntermediateBuffer(8888) -> pDst(rfi)
// invoked only from rq-thread, so use static buffers
// but it's better to use thread-local buffers (or special buffer manager)
vImage_Buffer srcBuf;
srcBuf.height = height;
srcBuf.width = width;
srcBuf.rowBytes = 4*width;
srcBuf.data = mtlbuf.contents;

vImage_Buffer destBuf;
destBuf.height = height;
destBuf.width = width;
destBuf.rowBytes = dstInfo.scanStride;
destBuf.data = pDst;

vImagePermuteChannels_ARGB8888(&srcBuf, &destBuf, rfi.permuteMap, kvImageNoFlags);
// mtlbuf.contents have smaller dimensions than pDst
// copy each row from mtlbuf.contents at appropriate position in pDst
// Note : pDst is already addjusted for offsets using PtrAddBytes above

int rowSize = width * dstInfo.pixelStride;
for (int y = 0; y < height; y++) {
copyFromMTLBuffer(pDst, mtlbuf, y * rowSize, rowSize, convertFromPre);
pDst = PtrAddBytes(pDst, dstInfo.scanStride);
}
}

#ifndef USE_STATIC_BUFFER
[mtlbuf release];
#endif // USE_STATIC_BUFFER
Expand Down

0 comments on commit 77dca42

Please sign in to comment.