MIPS DSP composition functions optimizations.
List of optimized routines: - comp_func_DestinationOver - comp_func_SourceIn - comp_func_DestinationIn - comp_func_DestinationOut - comp_func_SourceAtop - comp_func_DestinationAtop - comp_func_XOR - comp_func_SourceOut - comp_func_solid_SourceOver - comp_func_solid_DestinationOver - comp_func_solid_SourceIn - comp_func_solid_DestinationIn - comp_func_solid_SourceAtop - comp_func_solid_DestinationAtop - copm_func_solid_XOR - comp_func_solid_SourceOut Previously optimized routines qt_blend_argb32_on_argb32_mips_dsp and comp_func_Source_mips_dsp are redesigned and rewritten. Overall improvement by running tst_bench_blendbench benchmark app from tests/benchmarks/gui/image/blendbench/ is 27%. Change-Id: I6ab09b17cac10f4aded59787074ab4c89e72ccac Reviewed-by: Thiago Macieira <thiago.macieira@intel.com> Reviewed-by: Samuel Rødal <samuel.rodal@nokia.com>
This commit is contained in:
Damir Tatalovic
Qt by Nokia
parent
60829b4a56
commit
614422c2e7
@ -5971,6 +5971,23 @@ void qInitDrawhelperAsm()
|
||||
#if defined(QT_COMPILER_SUPPORTS_MIPS_DSP)
|
||||
functionForMode_C[QPainter::CompositionMode_SourceOver] = comp_func_SourceOver_asm_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_Source] = comp_func_Source_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_DestinationOver] = comp_func_DestinationOver_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_SourceIn] = comp_func_SourceIn_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_DestinationIn] = comp_func_DestinationIn_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_DestinationOut] = comp_func_DestinationOut_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_SourceAtop] = comp_func_SourceAtop_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_DestinationAtop] = comp_func_DestinationAtop_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_Xor] = comp_func_XOR_mips_dsp;
|
||||
functionForMode_C[QPainter::CompositionMode_SourceOut] = comp_func_SourceOut_mips_dsp;
|
||||
|
||||
functionForModeSolid_C[QPainter::CompositionMode_SourceOver] = comp_func_solid_SourceOver_mips_dsp;
|
||||
functionForModeSolid_C[QPainter::CompositionMode_DestinationOver] = comp_func_solid_DestinationOver_mips_dsp;
|
||||
functionForModeSolid_C[QPainter::CompositionMode_SourceIn] = comp_func_solid_SourceIn_mips_dsp;
|
||||
functionForModeSolid_C[QPainter::CompositionMode_DestinationIn] = comp_func_solid_DestinationIn_mips_dsp;
|
||||
functionForModeSolid_C[QPainter::CompositionMode_SourceAtop] = comp_func_solid_SourceAtop_mips_dsp;
|
||||
functionForModeSolid_C[QPainter::CompositionMode_DestinationAtop] = comp_func_solid_DestinationAtop_mips_dsp;
|
||||
functionForModeSolid_C[QPainter::CompositionMode_Xor] = comp_func_solid_XOR_mips_dsp;
|
||||
functionForModeSolid_C[QPainter::CompositionMode_SourceOut] = comp_func_solid_SourceOut_mips_dsp;
|
||||
|
||||
qt_memfill32 = qt_memfill32_asm_mips_dsp;
|
||||
|
||||
|
||||
@ -45,24 +45,6 @@
|
||||
|
||||
QT_BEGIN_NAMESPACE
|
||||
|
||||
#if defined(QT_COMPILER_SUPPORTS_MIPS_DSP)
|
||||
|
||||
extern "C" uint INTERPOLATE_PIXEL_255_asm_mips_dsp(uint x, uint a, uint y, uint b);
|
||||
|
||||
extern "C" uint BYTE_MUL_asm_mips_dsp(uint x, uint a);
|
||||
|
||||
extern "C" uint * destfetchARGB32_asm_mips_dsp(uint *buffer, const uint *data, int length);
|
||||
|
||||
extern "C" uint * qt_destStoreARGB32_asm_mips_dsp(uint *buffer, const uint *data, int length);
|
||||
|
||||
#if defined(QT_COMPILER_SUPPORTS_MIPS_DSPR2)
|
||||
|
||||
extern "C" uint INTERPOLATE_PIXEL_255_asm_mips_dspr2(uint x, uint a, uint y, uint b);
|
||||
|
||||
extern "C" uint BYTE_MUL_asm_mips_dspr2(uint x, uint a);
|
||||
|
||||
#endif // QT_COMPILER_SUPPORTS_MIPS_DSPR2
|
||||
|
||||
void qt_blend_argb32_on_argb32_mips_dsp(uchar *destPixels, int dbpl,
|
||||
const uchar *srcPixels, int sbpl,
|
||||
int w, int h,
|
||||
@ -80,32 +62,21 @@ void qt_blend_argb32_on_argb32_mips_dsp(uchar *destPixels, int dbpl,
|
||||
uint *dst = (uint *) destPixels;
|
||||
if (const_alpha == 256) {
|
||||
for (int y=0; y<h; ++y) {
|
||||
for (int x=0; x<w; ++x) {
|
||||
uint s = src[x];
|
||||
if (s >= 0xff000000)
|
||||
dst[x] = s;
|
||||
else if (s != 0)
|
||||
#if !defined(QT_COMPILER_SUPPORTS_MIPS_DSPR2)
|
||||
dst[x] = s + BYTE_MUL_asm_mips_dsp(dst[x], qAlpha(~s));
|
||||
#else
|
||||
dst[x] = s + BYTE_MUL_asm_mips_dspr2(dst[x], qAlpha(~s));
|
||||
#endif
|
||||
}
|
||||
qt_blend_argb32_on_argb32_const_alpha_256_mips_dsp_asm(dst, src, w);
|
||||
dst = (quint32 *)(((uchar *) dst) + dbpl);
|
||||
src = (const quint32 *)(((const uchar *) src) + sbpl);
|
||||
}
|
||||
} else if (const_alpha != 0) {
|
||||
const_alpha = (const_alpha * 255) >> 8;
|
||||
for (int y=0; y<h; ++y) {
|
||||
for (int x=0; x<w; ++x) {
|
||||
#if !defined(QT_COMPILER_SUPPORTS_MIPS_DSPR2)
|
||||
uint s = BYTE_MUL_asm_mips_dsp(src[x], const_alpha);
|
||||
dst[x] = s + BYTE_MUL_asm_mips_dsp(dst[x], qAlpha(~s));
|
||||
#else
|
||||
uint s = BYTE_MUL_asm_mips_dspr2(src[x], const_alpha);
|
||||
dst[x] = s + BYTE_MUL_asm_mips_dspr2(dst[x], qAlpha(~s));
|
||||
#endif
|
||||
if (h%2 > 0) {
|
||||
uint s = BYTE_MUL(src[0], const_alpha);
|
||||
dst[0] = s + BYTE_MUL(dst[0], qAlpha(~s));
|
||||
h--;
|
||||
dst++;
|
||||
src++;
|
||||
}
|
||||
qt_blend_argb32_on_argb32_mips_dsp_asm_x2(dst, src, h, const_alpha);
|
||||
dst = (quint32 *)(((uchar *) dst) + dbpl);
|
||||
src = (const quint32 *)(((const uchar *) src) + sbpl);
|
||||
}
|
||||
@ -145,13 +116,13 @@ void comp_func_Source_mips_dsp(uint *dest, const uint *src, int length, uint con
|
||||
::memcpy(dest, src, length * sizeof(uint));
|
||||
} else {
|
||||
int ialpha = 255 - const_alpha;
|
||||
for (int i = 0; i < length; ++i) {
|
||||
#if !defined(QT_COMPILER_SUPPORTS_MIPS_DSPR2)
|
||||
dest[i] = INTERPOLATE_PIXEL_255_asm_mips_dsp(src[i], const_alpha, dest[i], ialpha);
|
||||
#else
|
||||
dest[i] = INTERPOLATE_PIXEL_255_asm_mips_dspr2(src[i], const_alpha, dest[i], ialpha);
|
||||
#endif
|
||||
if (length%2 > 0) {
|
||||
dest[0] = INTERPOLATE_PIXEL_255(src[0], const_alpha, dest[0], ialpha);
|
||||
length--;
|
||||
dest++;
|
||||
src++;
|
||||
}
|
||||
comp_func_Source_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
}
|
||||
|
||||
@ -171,6 +142,285 @@ void QT_FASTCALL qt_destStoreARGB32_mips_dsp(QRasterBuffer *rasterBuffer, int x,
|
||||
qt_destStoreARGB32_asm_mips_dsp(data, buffer, length);
|
||||
}
|
||||
|
||||
#endif // QT_COMPILER_SUPPORTS_MIPS_DSP
|
||||
void QT_FASTCALL comp_func_solid_SourceOver_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
if (const_alpha != 255)
|
||||
color = BYTE_MUL(color, const_alpha);
|
||||
if (length%2 > 0) {
|
||||
dest[0] = color + BYTE_MUL(dest[0], qAlpha(~color));
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_Source_dsp_asm_x2(dest, length, color, qAlpha(~color));
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationOver_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
if (const_alpha != 255)
|
||||
color = BYTE_MUL(color, const_alpha);
|
||||
if (length%2 > 0) {
|
||||
uint d = dest[0];
|
||||
dest[0] = d + BYTE_MUL(color, qAlpha(~d));
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_DestinationOver_dsp_asm_x2(dest, length, color);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_DestinationOver_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
uint d = dest[0];
|
||||
dest[0] = d + BYTE_MUL(src[0], qAlpha(~d));
|
||||
} else {
|
||||
uint d = dest[0];
|
||||
uint s = BYTE_MUL(src[0], const_alpha);
|
||||
dest[0] = d + BYTE_MUL(s, qAlpha(~d));
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
src++;
|
||||
}
|
||||
comp_func_DestinationOver_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_solid_SourceIn_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
dest[0] = BYTE_MUL(color, qAlpha(dest[0]));
|
||||
} else {
|
||||
uint tmp_color = BYTE_MUL(color, const_alpha);
|
||||
uint cia = 255 - const_alpha;
|
||||
uint d = dest[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(tmp_color, qAlpha(d), d, cia);
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_SourceIn_dsp_asm_x2(dest, length, color, const_alpha);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_SourceIn_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
dest[0] = BYTE_MUL(src[0], qAlpha(dest[0]));
|
||||
} else {
|
||||
uint cia = 255 - const_alpha;
|
||||
uint d = dest[0];
|
||||
uint s = BYTE_MUL(src[0], const_alpha);
|
||||
dest[0] = INTERPOLATE_PIXEL_255(s, qAlpha(d), d, cia);
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
src++;
|
||||
}
|
||||
comp_func_SourceIn_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationIn_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
uint a = qAlpha(color);
|
||||
if (const_alpha != 255) {
|
||||
a = BYTE_MUL(a, const_alpha) + 255 - const_alpha;
|
||||
}
|
||||
if (length%2 > 0) {
|
||||
dest[0] = BYTE_MUL(dest[0], a);
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_DestinationIn_dsp_asm_x2(dest, length, a);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_DestinationIn_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
dest[0] = BYTE_MUL(dest[0], qAlpha(src[0]));
|
||||
} else {
|
||||
int cia = 255 - const_alpha;
|
||||
uint a = BYTE_MUL(qAlpha(src[0]), const_alpha) + cia;
|
||||
dest[0] = BYTE_MUL(dest[0], a);
|
||||
}
|
||||
length--;
|
||||
src++;
|
||||
dest++;
|
||||
}
|
||||
comp_func_DestinationIn_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationOut_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
uint a = qAlpha(~color);
|
||||
if (const_alpha != 255) {
|
||||
a = BYTE_MUL(a, const_alpha) + 255 - const_alpha;
|
||||
}
|
||||
if (length%2 > 0) {
|
||||
dest[0] = BYTE_MUL(dest[0], a);
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_DestinationIn_dsp_asm_x2(dest, length, a);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_DestinationOut_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
dest[0] = BYTE_MUL(dest[0], qAlpha(~src[0]));
|
||||
} else {
|
||||
int cia = 255 - const_alpha;
|
||||
uint sia = BYTE_MUL(qAlpha(~src[0]), const_alpha) + cia;
|
||||
dest[0] = BYTE_MUL(dest[0], sia);
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
src++;
|
||||
}
|
||||
comp_func_DestinationOut_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_solid_SourceAtop_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
if (const_alpha != 255) {
|
||||
color = BYTE_MUL(color, const_alpha);
|
||||
}
|
||||
uint sia = qAlpha(~color);
|
||||
if (length%2 > 0) {
|
||||
dest[0] = INTERPOLATE_PIXEL_255(color, qAlpha(dest[0]), dest[0], sia);
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_SourceAtop_dsp_asm_x2(dest, length, color, sia);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_SourceAtop_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
uint s = src[0];
|
||||
uint d = dest[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(s, qAlpha(d), d, qAlpha(~s));
|
||||
} else {
|
||||
uint s = BYTE_MUL(src[0], const_alpha);
|
||||
uint d = dest[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(s, qAlpha(d), d, qAlpha(~s));
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
src++;
|
||||
}
|
||||
comp_func_SourceAtop_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationAtop_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
uint a = qAlpha(color);
|
||||
if (const_alpha != 255) {
|
||||
color = BYTE_MUL(color, const_alpha);
|
||||
a = qAlpha(color) + 255 - const_alpha;
|
||||
}
|
||||
if (length%2 > 0) {
|
||||
uint d = dest[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(d, a, color, qAlpha(~d));
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_DestinationAtop_dsp_asm_x2(dest, length, color, a);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_DestinationAtop_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
uint s = src[0];
|
||||
uint d = dest[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(d, qAlpha(s), s, qAlpha(~d));
|
||||
} else {
|
||||
int cia = 255 - const_alpha;
|
||||
uint s = BYTE_MUL(src[0], const_alpha);
|
||||
uint d = dest[0];
|
||||
uint a = qAlpha(s) + cia;
|
||||
dest[0] = INTERPOLATE_PIXEL_255(d, a, s, qAlpha(~d));
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
src++;
|
||||
}
|
||||
comp_func_DestinationAtop_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_solid_XOR_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
if (const_alpha != 255)
|
||||
color = BYTE_MUL(color, const_alpha);
|
||||
uint sia = qAlpha(~color);
|
||||
|
||||
if (length%2 > 0) {
|
||||
uint d = dest[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(color, qAlpha(~d), d, sia);
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_XOR_dsp_asm_x2(dest, length, color, sia);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_XOR_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
uint d = dest[0];
|
||||
uint s = src[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(s, qAlpha(~d), d, qAlpha(~s));
|
||||
} else {
|
||||
uint d = dest[0];
|
||||
uint s = BYTE_MUL(src[0], const_alpha);
|
||||
dest[0] = INTERPOLATE_PIXEL_255(s, qAlpha(~d), d, qAlpha(~s));
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
src++;
|
||||
}
|
||||
comp_func_XOR_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_solid_SourceOut_mips_dsp(uint *dest, int length, uint color, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
dest[0] = BYTE_MUL(color, qAlpha(~dest[0]));
|
||||
} else {
|
||||
uint tmp_color = BYTE_MUL(color, const_alpha);
|
||||
int cia = 255 - const_alpha;
|
||||
uint d = dest[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(tmp_color, qAlpha(~d), d, cia);
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
}
|
||||
comp_func_solid_SourceOut_dsp_asm_x2(dest, length, color, const_alpha);
|
||||
}
|
||||
|
||||
void QT_FASTCALL comp_func_SourceOut_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha)
|
||||
{
|
||||
if (length%2 > 0) {
|
||||
if (const_alpha == 255) {
|
||||
dest[0] = BYTE_MUL(src[0], qAlpha(~dest[0]));
|
||||
} else {
|
||||
int cia = 255 - const_alpha;
|
||||
uint s = BYTE_MUL(src[0], const_alpha);
|
||||
uint d = dest[0];
|
||||
dest[0] = INTERPOLATE_PIXEL_255(s, qAlpha(~d), d, cia);
|
||||
}
|
||||
length--;
|
||||
dest++;
|
||||
src++;
|
||||
}
|
||||
comp_func_SourceOut_dsp_asm_x2(dest, src, length, const_alpha);
|
||||
}
|
||||
|
||||
|
||||
QT_END_NAMESPACE
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@ -52,6 +52,48 @@ extern "C" void qt_memfill32_asm_mips_dsp(quint32 *dest, quint32 value, int coun
|
||||
|
||||
extern "C" void comp_func_SourceOver_asm_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_solid_DestinationOver_dsp_asm_x2(uint *dest, int length, uint color);
|
||||
|
||||
extern "C" void comp_func_solid_Source_dsp_asm_x2(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_DestinationOver_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_solid_SourceIn_dsp_asm_x2(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_SourceIn_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_solid_DestinationIn_dsp_asm_x2(uint *dest, int length, uint a);
|
||||
|
||||
extern "C" void comp_func_DestinationIn_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_DestinationOut_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_solid_SourceAtop_dsp_asm_x2(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_SourceAtop_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_solid_DestinationAtop_dsp_asm_x2(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_DestinationAtop_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_solid_XOR_dsp_asm_x2(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_XOR_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_solid_SourceOut_dsp_asm_x2(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_SourceOut_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void comp_func_Source_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void qt_blend_argb32_on_argb32_mips_dsp_asm_x2(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
extern "C" void qt_blend_argb32_on_argb32_const_alpha_256_mips_dsp_asm(uint *dest, const uint *src, int length);
|
||||
|
||||
extern "C" uint * destfetchARGB32_asm_mips_dsp(uint *buffer, const uint *data, int length);
|
||||
|
||||
extern "C" uint * qt_destStoreARGB32_asm_mips_dsp(uint *buffer, const uint *data, int length);
|
||||
|
||||
void qt_blend_argb32_on_argb32_mips_dsp(uchar *destPixels, int dbpl,
|
||||
const uchar *srcPixels, int sbpl,
|
||||
int w, int h,
|
||||
@ -71,6 +113,46 @@ uint * QT_FASTCALL qt_destFetchARGB32_mips_dsp(uint *buffer,
|
||||
void QT_FASTCALL qt_destStoreARGB32_mips_dsp(QRasterBuffer *rasterBuffer, int x, int y,
|
||||
const uint *buffer, int length);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_Source_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_SourceOver_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationOver_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_SourceOver_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationOver_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_DestinationOver_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_SourceIn_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_SourceIn_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationIn_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_DestinationIn_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationOut_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_DestinationOut_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_SourceAtop_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_SourceAtop_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_DestinationAtop_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_DestinationAtop_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_XOR_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_solid_SourceOut_mips_dsp(uint *dest, int length, uint color, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_SourceOut_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
void QT_FASTCALL comp_func_XOR_mips_dsp(uint *dest, const uint *src, int length, uint const_alpha);
|
||||
|
||||
#endif // QT_COMPILER_SUPPORTS_MIPS_DSP
|
||||
|
||||
|
||||
|
||||
@ -41,59 +41,6 @@
|
||||
|
||||
#include "qt_mips_asm_dsp.h"
|
||||
|
||||
LEAF_MIPS_DSPR2(INTERPOLATE_PIXEL_255_asm_mips_dspr2)
|
||||
/*
|
||||
* a0 - uint x (First value to multiply)
|
||||
* a1 - uint a (Multiplicator byte for first value)
|
||||
* a2 - uint y (Second value to multiply)
|
||||
* a3 - uint b (Multiplicator byte for second value)
|
||||
*/
|
||||
|
||||
.set reorder
|
||||
replv.ph a1, a1
|
||||
replv.ph a3, a3
|
||||
li t8, 8388736
|
||||
muleu_s.ph.qbl t0, a0, a1
|
||||
muleu_s.ph.qbl t1, a2, a3
|
||||
muleu_s.ph.qbr t2, a0, a1
|
||||
muleu_s.ph.qbr t3, a2, a3
|
||||
addu.ph t4, t0, t1
|
||||
addu.ph t5, t2, t3
|
||||
preceu.ph.qbla t0, t4
|
||||
addu t1, t0, t8
|
||||
addu t1, t4, t1
|
||||
preceu.ph.qbla t6, t5
|
||||
addu t7, t6, t8
|
||||
addu t7, t5, t7
|
||||
precrq.qb.ph t2, t1, t7
|
||||
move v0, t2
|
||||
j ra
|
||||
|
||||
END(INTERPOLATE_PIXEL_255_asm_mips_dspr2)
|
||||
|
||||
LEAF_MIPS_DSPR2(BYTE_MUL_asm_mips_dspr2)
|
||||
/*
|
||||
* a0 - uint x (Value to multiply)
|
||||
* a1 - uint a (Multiplicator byte)
|
||||
*/
|
||||
|
||||
.set reorder
|
||||
replv.ph a1, a1 /* a1 = 0x00a00a */
|
||||
li t4, 8388736 /* t4 = 0x800080 */
|
||||
muleu_s.ph.qbl t0, a0, a1
|
||||
muleu_s.ph.qbr t2, a0, a1
|
||||
preceu.ph.qbla t1, t0
|
||||
addu t0, t0, t1
|
||||
addu t0, t0, t4
|
||||
preceu.ph.qbla t3, t2
|
||||
addu t2, t2, t3
|
||||
addu t2, t2, t4
|
||||
precrq.qb.ph t4, t0, t2
|
||||
move v0, t4
|
||||
j ra
|
||||
|
||||
END(BYTE_MUL_asm_mips_dspr2)
|
||||
|
||||
LEAF_MIPS_DSPR2(qConvertRgb16To32_asm_mips_dspr2)
|
||||
/*
|
||||
* a0 - dst (a8r8g8b8)
|
||||
|
||||
@ -110,4 +110,140 @@ LEAF_MIPS32R2(symbol) \
|
||||
.end function; \
|
||||
.size function,.-function
|
||||
|
||||
/*
|
||||
* BYTE_MUL operation on two pixels (in_1 and in_2) with two
|
||||
* multiplicator bytes, repl_a1 and repl_a2, which should be
|
||||
* prepered with:
|
||||
* replv.ph repl_a1, a1
|
||||
* replv.ph repl_a2, a2
|
||||
* to became such as:
|
||||
* repl_a1 = | 00 | a1 | 00 | a1 |
|
||||
* repl_a2 = | 00 | a2 | 00 | a2 |
|
||||
*
|
||||
* rounding_factor must have following value:
|
||||
* li rounding_factor, 0x00800080
|
||||
*
|
||||
* scratch(n) - temporary registers
|
||||
*
|
||||
* in_const: 1 -> (default) causes that in_1, in_2
|
||||
* registers will remain unchanged after usage
|
||||
* 0 -> (or anything different then 1) causes
|
||||
* that registers repl_a1, repl_a2 remain
|
||||
* unchanged after usage
|
||||
*/
|
||||
.macro BYTE_MUL_x2 in_1, in_2, out_1, out_2 \
|
||||
repl_a1, repl_a2, rounding_factor, \
|
||||
scratch1, scratch2, scratch3, scratch4, \
|
||||
in_const = 1
|
||||
muleu_s.ph.qbl \scratch1, \in_1, \repl_a1
|
||||
muleu_s.ph.qbr \scratch2, \in_1, \repl_a1
|
||||
muleu_s.ph.qbl \scratch3, \in_2, \repl_a2
|
||||
muleu_s.ph.qbr \scratch4, \in_2, \repl_a2
|
||||
|
||||
.if \in_const == 1
|
||||
preceu.ph.qbla \repl_a1, \scratch1
|
||||
preceu.ph.qbla \repl_a2, \scratch2
|
||||
preceu.ph.qbla \out_1, \scratch3
|
||||
preceu.ph.qbla \out_2, \scratch4
|
||||
|
||||
addu \scratch1, \repl_a1, \scratch1
|
||||
addu \scratch2, \repl_a2, \scratch2
|
||||
.else
|
||||
preceu.ph.qbla \in_1, \scratch1
|
||||
preceu.ph.qbla \in_2, \scratch2
|
||||
preceu.ph.qbla \out_1, \scratch3
|
||||
preceu.ph.qbla \out_2, \scratch4
|
||||
|
||||
addu \scratch1, \in_1, \scratch1
|
||||
addu \scratch2, \in_2, \scratch2
|
||||
.endif
|
||||
|
||||
addu \out_1, \out_1, \scratch3
|
||||
addu \out_2, \out_2, \scratch4
|
||||
|
||||
addu \scratch1, \scratch1, \rounding_factor
|
||||
addu \scratch2, \scratch2, \rounding_factor
|
||||
addu \scratch3, \out_1, \rounding_factor
|
||||
addu \scratch4, \out_2, \rounding_factor
|
||||
|
||||
precrq.qb.ph \out_1, \scratch1, \scratch2
|
||||
precrq.qb.ph \out_2, \scratch3, \scratch4
|
||||
|
||||
.endm
|
||||
|
||||
/*
|
||||
* BYTE_MUL operation on one pixel (in_1) with
|
||||
* multiplicator byte, repl_a1, which should be
|
||||
* prepered with:
|
||||
* replv.ph repl_a1, a1
|
||||
* to became such as:
|
||||
* repl_a1 = | 00 | a1 | 00 | a1 |
|
||||
*
|
||||
* rounding_factor must have following value:
|
||||
* li rounding_factor, 0x00800080
|
||||
*
|
||||
* scratch(n) - temporary registers
|
||||
*/
|
||||
.macro BYTE_MUL in_1, out_1, \
|
||||
repl_a1, rounding_factor, \
|
||||
scratch1, scratch2, scratch3, scratch4
|
||||
muleu_s.ph.qbl \scratch1, \in_1, \repl_a1
|
||||
muleu_s.ph.qbr \scratch2, \in_1, \repl_a1
|
||||
|
||||
preceu.ph.qbla \scratch3, \scratch1
|
||||
preceu.ph.qbla \scratch4, \scratch2
|
||||
|
||||
addu \scratch1, \scratch1, \scratch3
|
||||
addu \scratch1, \scratch1, \rounding_factor
|
||||
|
||||
addu \scratch2, \scratch2, \scratch4
|
||||
addu \scratch2, \scratch2, \rounding_factor
|
||||
|
||||
precrq.qb.ph \out_1, \scratch1, \scratch2
|
||||
|
||||
.endm
|
||||
|
||||
/*
|
||||
* macro for INTERPOLATE_PIXEL_255 operation
|
||||
* in_1 - First value to multiply
|
||||
* mul_1 - Multiplicator byte for first value
|
||||
* in_2 - Second value to multiply
|
||||
* mul_2 - Multiplicator byte for second value
|
||||
* rounding_factor and andi_factor should be prepared
|
||||
* as:
|
||||
* li rounding_factor, 0x00800080
|
||||
* li andi_factor, 0xff00ff00
|
||||
* scratch(n) - temporary registers
|
||||
*/
|
||||
.macro INTERPOLATE_PIXEL_255 in_1, mul_1, \
|
||||
in_2, mul_2, \
|
||||
out_1, \
|
||||
rounding_factor, andi_factor \
|
||||
scratch1, scratch2, scratch3, scratch4
|
||||
# x part
|
||||
preceu.ph.qbra \scratch1, \in_1
|
||||
preceu.ph.qbra \scratch2, \in_2
|
||||
mul \scratch1, \scratch1, \mul_1
|
||||
mul \scratch2, \scratch2, \mul_2
|
||||
# x>>8 part
|
||||
preceu.ph.qbla \scratch3, \in_1
|
||||
preceu.ph.qbla \scratch4, \in_2
|
||||
mul \scratch3, \scratch3, \mul_1
|
||||
mul \scratch4, \scratch4, \mul_2
|
||||
# x part
|
||||
addu \scratch1, \scratch1, \scratch2
|
||||
preceu.ph.qbla \scratch2, \scratch1
|
||||
addu \scratch1, \scratch1, \scratch2
|
||||
addu \scratch1, \scratch1, \rounding_factor
|
||||
preceu.ph.qbla \scratch1, \scratch1
|
||||
# x>>8 part
|
||||
addu \scratch3, \scratch3, \scratch4
|
||||
preceu.ph.qbla \scratch4, \scratch3
|
||||
addu \scratch3, \scratch3, \scratch4
|
||||
addu \scratch3, \scratch3, \rounding_factor
|
||||
and \scratch3, \scratch3, \andi_factor
|
||||
|
||||
or \out_1, \scratch1, \scratch3
|
||||
.endm
|
||||
|
||||
#endif //QT_MIPS_DSP_H__
|
||||
|
||||
Loading…
Reference in New Issue
Block a user