refactoring, speedup by decoupling back to front transferes from drawing and usage of special memcpy routine, minor speedups in Painter

git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12437 a95241bf-73f2-0310-859d-f6bbb57e9c96
This commit is contained in:
Stephan Aßmus 2005-04-18 16:09:43 +00:00
parent 300af26345
commit 74b3612ac3
7 changed files with 210 additions and 159 deletions

View File

@ -523,68 +523,8 @@ AccelerantHWInterface::BackBuffer() const
void
AccelerantHWInterface::_DrawCursor(BRect area) const
{
BRect cf = _CursorFrame();
if (cf.IsValid() && area.Intersects(cf)) {
// clip to common area
area = area & cf;
int32 left = (int32)floorf(area.left);
int32 top = (int32)floorf(area.top);
int32 right = (int32)ceilf(area.right);
int32 bottom = (int32)ceilf(area.bottom);
int32 width = right - left + 1;
int32 height = bottom - top + 1;
// make a bitmap from the backbuffer
// that has the cursor blended on top of it
// blending buffer
uint8* buffer = new uint8[width * height * 4];
// offset into back buffer
uint8* src = (uint8*)fBackBuffer->Bits();
uint32 srcBPR = fBackBuffer->BytesPerRow();
src += top * srcBPR + left * 4;
// offset into cursor bitmap
uint8* crs = (uint8*)fCursor->Bits();
uint32 crsBPR = fCursor->BytesPerRow();
// since area is clipped to cf,
// the diff between top and cf.top is always positive,
// same for diff between left and cf.left
crs += (top - (int32)floorf(cf.top)) * crsBPR
+ (left - (int32)floorf(cf.left)) * 4;
uint8* dst = buffer;
// blending
for (int32 y = top; y <= bottom; y++) {
uint8* s = src;
uint8* c = crs;
uint8* d = dst;
for (int32 x = left; x <= right; x++) {
// assume backbuffer alpha = 255
// TODO: it appears alpha in cursor us upside down
uint8 a = 255 - c[3];
d[0] = (((s[0] - c[0]) * a) + (c[0] << 8)) >> 8;
d[1] = (((s[1] - c[1]) * a) + (c[1] << 8)) >> 8;
d[2] = (((s[2] - c[2]) * a) + (c[2] << 8)) >> 8;
d[3] = 255;
s += 4;
c += 4;
d += 4;
}
crs += crsBPR;
src += srcBPR;
dst += width * 4;
}
// copy result to front buffer
_CopyToFront(buffer, width * 4, left, top, right, bottom);
delete[] buffer;
}
// use the default implementation for now,
// until we have a hardware cursor
HWInterface::_DrawCursor(area);
}

View File

@ -19,8 +19,8 @@ HWInterface::HWInterface()
fCursor(NULL),
fCursorVisible(true),
fCursorLocation(0, 0),
// fUpdateExecutor(new UpdateQueue(this))
fUpdateExecutor(NULL)
fUpdateExecutor(new UpdateQueue(this))
// fUpdateExecutor(NULL)
{
}
@ -105,14 +105,18 @@ HWInterface::GetCursorPosition()
status_t
HWInterface::Invalidate(const BRect& frame)
{
return CopyBackToFront(frame);;
// return CopyBackToFront(frame);
// TODO: get this working, the locking in the DisplayDriverPainter needs
// to be based on locking this object, which essentially means the access
// to the back buffer is locked, or more precise the access to the invalid
// region scheduled to be copied to the front buffer
// fUpdateExecutor->AddRect(frame);
// return B_OK;
// TODO: the remaining problem is the immediate wake up of the
// thread carrying out the updates, when I enable it, there
// seems to be a deadlock, but I didn't figure it out yet.
// Maybe the same bug is there without the wakeup, only, triggered
// less often.... scarry, huh?
if (frame.IsValid()) {
fUpdateExecutor->AddRect(frame);
return B_OK;
}
return B_BAD_VALUE;
}
// CopyBackToFront
@ -155,6 +159,155 @@ HWInterface::CopyBackToFront(const BRect& frame)
return B_BAD_VALUE;
}
// _DrawCursor
// * default implementation, can be used as fallback or for
// software cursor
void
HWInterface::_DrawCursor(BRect area) const
{
BRect cf = _CursorFrame();
RenderingBuffer* backBuffer = BackBuffer();
if (backBuffer && cf.IsValid() && area.Intersects(cf)) {
// clip to common area
area = area & cf;
int32 left = (int32)area.left;
int32 top = (int32)area.top;
int32 right = (int32)area.right;
int32 bottom = (int32)area.bottom;
int32 width = right - left + 1;
int32 height = bottom - top + 1;
// make a bitmap from the backbuffer
// that has the cursor blended on top of it
// blending buffer
uint8* buffer = new uint8[width * height * 4];
// offset into back buffer
uint8* src = (uint8*)backBuffer->Bits();
uint32 srcBPR = backBuffer->BytesPerRow();
src += top * srcBPR + left * 4;
// offset into cursor bitmap
uint8* crs = (uint8*)fCursor->Bits();
uint32 crsBPR = fCursor->BytesPerRow();
// since area is clipped to cf,
// the diff between top and cf.top is always positive,
// same for diff between left and cf.left
crs += (top - (int32)floorf(cf.top)) * crsBPR
+ (left - (int32)floorf(cf.left)) * 4;
uint8* dst = buffer;
// blending
for (int32 y = top; y <= bottom; y++) {
uint8* s = src;
uint8* c = crs;
uint8* d = dst;
for (int32 x = left; x <= right; x++) {
// assume backbuffer alpha = 255
// TODO: it appears alpha in cursor is upside down
uint8 a = 255 - c[3];
d[0] = (((s[0] - c[0]) * a) + (c[0] << 8)) >> 8;
d[1] = (((s[1] - c[1]) * a) + (c[1] << 8)) >> 8;
d[2] = (((s[2] - c[2]) * a) + (c[2] << 8)) >> 8;
d[3] = 255;
s += 4;
c += 4;
d += 4;
}
crs += crsBPR;
src += srcBPR;
dst += width * 4;
}
// copy result to front buffer
_CopyToFront(buffer, width * 4, left, top, right, bottom);
delete[] buffer;
}
}
/*
// gfxcpy
inline
void
gfxcpy(uint8* dst, uint8* src, int32 numBytes)
{
uint64* d64 = (uint64*)dst;
uint64* s64 = (uint64*)src;
int32 numBytesBegin = numBytes;
while (numBytes >= 32) {
*d64++ = *s64++;
*d64++ = *s64++;
*d64++ = *s64++;
*d64++ = *s64++;
numBytes -= 32;
}
while (numBytes >= 16) {
*d64++ = *s64++;
*d64++ = *s64++;
numBytes -= 16;
}
while (numBytes >= 8) {
*d64++ = *s64++;
numBytes -= 8;
}
if (numBytes > 0) {
// update original pointers
dst += numBytesBegin - numBytes;
src += numBytesBegin - numBytes;
numBytesBegin = numBytes;
uint32* d32 = (uint32*)dst;
uint32* s32 = (uint32*)src;
while (numBytes >= 4) {
*d32++ = *s32++;
numBytes -= 4;
}
// update original pointers
dst += numBytesBegin - numBytes;
src += numBytesBegin - numBytes;
while (numBytes > 0) {
*dst++ = *src++;
numBytes--;
}
}
}*/
// gfxcpy32
// * numBytes is expected to be a multiple of 4
inline
void
gfxcpy32(uint8* dst, uint8* src, int32 numBytes)
{
uint64* d64 = (uint64*)dst;
uint64* s64 = (uint64*)src;
int32 numBytesStart = numBytes;
while (numBytes >= 32) {
*d64++ = *s64++;
*d64++ = *s64++;
*d64++ = *s64++;
*d64++ = *s64++;
numBytes -= 32;
}
if (numBytes >= 16) {
*d64++ = *s64++;
*d64++ = *s64++;
numBytes -= 16;
}
if (numBytes >= 8) {
*d64++ = *s64++;
numBytes -= 8;
}
if (numBytes == 4) {
uint32* d32 = (uint32*)(dst + numBytesStart - numBytes);
uint32* s32 = (uint32*)(src + numBytesStart - numBytes);
*d32 = *s32;
}
}
// _CopyToFront
//
@ -183,7 +336,12 @@ HWInterface::_CopyToFront(uint8* src, uint32 srcBPR,
dst += y * dstBPR + x * 4;
// copy
for (; y <= bottom; y++) {
#ifndef __HAIKU__
memcpy(dst, src, bytes);
#else
// bytes is guaranteed to be multiple of 4
gfxcpy32(dst, src, bytes);
#endif // __HAIKU__
dst += dstBPR;
src += srcBPR;
}

View File

@ -67,7 +67,7 @@ class HWInterface : public BLocker {
protected:
// implement this in derived classes
virtual void _DrawCursor(BRect area) const = 0;
virtual void _DrawCursor(BRect area) const;
// does the actual transfer and handles color space conversion
void _CopyToFront(uint8* src, uint32 srcBPR,

View File

@ -363,6 +363,11 @@ Painter::StrokeLine(BPoint b, DrawData* context)
return StrokeLine(context->penlocation, b, context);
}
typedef union {
uint32 data32;
int8 data[4];
} color32;
// StraightLine
bool
Painter::StraightLine(BPoint a, BPoint b, const rgb_color& c) const
@ -376,6 +381,11 @@ Painter::StraightLine(BPoint a, BPoint b, const rgb_color& c) const
dst += x * 4;
int32 y1 = (int32)min_c(a.y, b.y);
int32 y2 = (int32)max_c(a.y, b.y);
color32 color;
color.data[0] = c.blue;
color.data[1] = c.green;
color.data[2] = c.red;
color.data[3] = 255;
// draw a line, iterate over clipping boxes
fBaseRenderer->first_clip_box();
do {
@ -385,9 +395,7 @@ Painter::StraightLine(BPoint a, BPoint b, const rgb_color& c) const
int32 end = min_c(fBaseRenderer->ymax(), y2);
uint8* handle = dst + i * bpr;
for (; i <= end; i++) {
handle[0] = c.blue;
handle[1] = c.green;
handle[2] = c.red;
*(uint32*)handle = color.data32;
handle += bpr;
}
}
@ -403,6 +411,11 @@ Painter::StraightLine(BPoint a, BPoint b, const rgb_color& c) const
dst += y * bpr;
int32 x1 = (int32)min_c(a.x, b.x);
int32 x2 = (int32)max_c(a.x, b.x);
color32 color;
color.data[0] = c.blue;
color.data[1] = c.green;
color.data[2] = c.red;
color.data[3] = 255;
// draw a line, iterate over clipping boxes
fBaseRenderer->first_clip_box();
do {
@ -412,9 +425,7 @@ Painter::StraightLine(BPoint a, BPoint b, const rgb_color& c) const
int32 end = min_c(fBaseRenderer->xmax(), x2);
uint8* handle = dst + i * 4;
for (; i <= end; i++) {
handle[0] = c.blue;
handle[1] = c.green;
handle[2] = c.red;
*(uint32*)handle = color.data32;
handle += 4;
}
}
@ -626,6 +637,12 @@ Painter::FillRect(const BRect& r, const rgb_color& c) const
int32 top = (int32)r.top;
int32 right = (int32)r.right;
int32 bottom = (int32)r.bottom;
// get a 32 bit pixel ready with the color
color32 color;
color.data[0] = c.blue;
color.data[1] = c.green;
color.data[2] = c.red;
color.data[3] = 255;
// fill rects, iterate over clipping boxes
fBaseRenderer->first_clip_box();
do {
@ -636,12 +653,9 @@ Painter::FillRect(const BRect& r, const rgb_color& c) const
int32 y2 = min_c(fBaseRenderer->ymax(), bottom);
uint8* offset = dst + x1 * 4;
for (; y1 <= y2; y1++) {
uint8* handle = offset + y1 * bpr;
uint32* handle = (uint32*)(offset + y1 * bpr);
for (int32 x = x1; x <= x2; x++) {
handle[0] = c.blue;
handle[1] = c.green;
handle[2] = c.red;
handle += 4;
*handle++ = color.data32;
}
}
}

View File

@ -53,8 +53,15 @@ UpdateQueue::AddRect(const BRect& rect)
{
// Lock();
//printf("UpdateQueue::AddRect()\n");
// NOTE: The access to fUpdateRegion
// is protected by the HWInterface lock.
// When trying to access the fUpdateRegion,
// our thread will first try to lock the
// HWInterface, while on the other hand
// HWInterface will always be locked when
// it calls AddRect().
fUpdateRegion.Include(rect);
_Reschedule();
// _Reschedule();
// Unlock();
}
@ -80,9 +87,7 @@ UpdateQueue::_ExecuteUpdates()
// execute updates
if (fInterface->LockWithTimeout(5000) >= B_OK) {
int32 count = fUpdateRegion.CountRects();
//printf("%ld copy dirty region\n", find_thread(NULL));
if (count > 0) {
printf("%ld rects\n", count);
for (int32 i = 0; i < count; i++) {
fInterface->CopyBackToFront(fUpdateRegion.RectAt(i));
}
@ -96,7 +101,7 @@ printf("%ld rects\n", count);
break;
default:
printf("other error: %s\n", strerror(err));
running = false;
//running = false;
snooze(20000);
break;
}
@ -110,6 +115,11 @@ running = false;
void
UpdateQueue::_Reschedule()
{
// TODO: _Reschedule() is supposed to cause the
// immediate wake up of the update thread, but
// the HWInterface is still locked when we get here.
// Somehow this causes a deadlock, but I don't
// see why yet...
if (fStatus == B_OK) {
release_sem(fThreadControl);
}

View File

@ -611,74 +611,6 @@ ViewHWInterface::CopyBackToFront(const BRect& frame)
return ret;
}
// _DrawCursor
void
ViewHWInterface::_DrawCursor(BRect area) const
{
BRect cf = _CursorFrame();
if (cf.IsValid() && area.Intersects(cf)) {
// clip to common area
area = area & cf;
int32 left = (int32)floorf(area.left);
int32 top = (int32)floorf(area.top);
int32 right = (int32)ceilf(area.right);
int32 bottom = (int32)ceilf(area.bottom);
int32 width = right - left + 1;
int32 height = bottom - top + 1;
// make a bitmap from the backbuffer
// that has the cursor blended on top of it
// blending buffer
uint8* buffer = new uint8[width * height * 4];
// offset into back buffer
uint8* src = (uint8*)fBackBuffer->Bits();
uint32 srcBPR = fBackBuffer->BytesPerRow();
src += top * srcBPR + left * 4;
// offset into cursor bitmap
uint8* crs = (uint8*)fCursor->Bits();
uint32 crsBPR = fCursor->BytesPerRow();
// since area is clipped to cf,
// the diff between top and cf.top is always positive,
// same for diff between left and cf.left
crs += (top - (int32)floorf(cf.top)) * crsBPR
+ (left - (int32)floorf(cf.left)) * 4;
uint8* dst = buffer;
// blending
for (int32 y = top; y <= bottom; y++) {
uint8* s = src;
uint8* c = crs;
uint8* d = dst;
for (int32 x = left; x <= right; x++) {
// assume backbuffer alpha = 255
// TODO: it appears alpha in cursor us upside down
uint8 a = 255 - c[3];
d[0] = (((s[0] - c[0]) * a) + (c[0] << 8)) >> 8;
d[1] = (((s[1] - c[1]) * a) + (c[1] << 8)) >> 8;
d[2] = (((s[2] - c[2]) * a) + (c[2] << 8)) >> 8;
d[3] = 255;
s += 4;
c += 4;
d += 4;
}
crs += crsBPR;
src += srcBPR;
dst += width * 4;
}
// copy result to front buffer
_CopyToFront(buffer, width * 4, left, top, right, bottom);
delete[] buffer;
}
}
/*void ViewHWInterface::CopyBitmap(ServerBitmap *bitmap, const BRect &source, const BRect &dest, const DrawData *d)
{

View File

@ -55,9 +55,6 @@ class ViewHWInterface : public HWInterface {
virtual status_t CopyBackToFront(const BRect& frame);
protected:
virtual void _DrawCursor(BRect area) const;
private:
BitmapBuffer* fBackBuffer;
BitmapBuffer* fFrontBuffer;