Skip to content

L
libyuv
Commit b911428a authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

Adapt row interpolator to do YUV as well as ARGB without extrude so it can be used in I420Scale. 

BUG=237
TEST=Scale*
R=ryanpetrie@google.com

Review URL: https://webrtc-codereview.appspot.com/1587004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@710 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 8ffe78ab
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 321 additions and 219 deletions
README.chromium
View file @ b911428a
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 709
Version: 710
License: BSD
License File: LICENSE
......
include/libyuv/row.h
View file @ b911428a
......@@ -109,8 +109,8 @@ extern "C" {
#define HAS_ARGBBLENDROW_SSSE3
#define HAS_ARGBCOLORMATRIXROW_SSSE3
#define HAS_ARGBGRAYROW_SSSE3
#define HAS_ARGBINTERPOLATEROW_SSE2
#define HAS_ARGBINTERPOLATEROW_SSSE3
#define HAS_INTERPOLATEROW_SSE2
#define HAS_INTERPOLATEROW_SSSE3
#define HAS_ARGBMIRRORROW_SSSE3
#define HAS_ARGBMULTIPLYROW_SSE2
#define HAS_ARGBQUANTIZEROW_SSE2
......@@ -261,7 +261,6 @@ extern "C" {
#define HAS_ARGBBLENDROW_NEON
#define HAS_ARGBCOLORMATRIXROW_NEON
#define HAS_ARGBGRAYROW_NEON
#define HAS_ARGBINTERPOLATEROW_NEON
#define HAS_ARGBMIRRORROW_NEON
#define HAS_ARGBMULTIPLYROW_NEON
#define HAS_ARGBQUANTIZEROW_NEON
......@@ -272,6 +271,7 @@ extern "C" {
#define HAS_SOBELXYROW_NEON
#define HAS_SOBELXROW_NEON
#define HAS_SOBELYROW_NEON
#define HAS_INTERPOLATEROW_NEON
#endif
// The following are available on Mips platforms
......@@ -281,6 +281,7 @@ extern "C" {
#define HAS_I422TOABGRROW_MIPS_DSPR2
#define HAS_I422TOARGBROW_MIPS_DSPR2
#define HAS_I422TOBGRAROW_MIPS_DSPR2
#define HAS_INTERPOLATEROWS_MIPS_DSPR2
#define HAS_MIRRORROW_MIPS_DSPR2
#define HAS_MIRRORUVROW_MIPS_DSPR2
#define HAS_SPLITUVROW_MIPS_DSPR2
......@@ -1455,34 +1456,40 @@ LIBYUV_API
void ARGBAffineRow_SSE2(const uint8* src_argb, int src_argb_stride,
uint8* dst_argb, const float* uv_dudv, int width);
// Used for ARGBScale and ARGBInterpolate.
void ARGBInterpolateRow_C(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb,
int width, int source_y_fraction);
void ARGBInterpolateRow_SSE2(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb, int width,
// Used for I420Scale, ARGBScale, and ARGBInterpolate.
void InterpolateRow_C(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr,
int width, int source_y_fraction);
void InterpolateRow_SSE2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void InterpolateRow_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void InterpolateRow_NEON(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void InterpolateRows_MIPS_DSPR2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void InterpolateRow_Unaligned_SSE2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void InterpolateRow_Unaligned_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void InterpolateRow_Any_NEON(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void ARGBInterpolateRow_SSSE3(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb, int width,
int source_y_fraction);
void ARGBInterpolateRow_NEON(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb, int width,
void InterpolateRow_Any_SSE2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void ARGBInterpolateRow_Unaligned_SSE2(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb, int width,
int source_y_fraction);
void ARGBInterpolateRow_Unaligned_SSSE3(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb, int width,
int source_y_fraction);
void ARGBInterpolateRow_Any_NEON(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb, int width,
int source_y_fraction);
void ARGBInterpolateRow_Any_SSE2(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb, int width,
int source_y_fraction);
void ARGBInterpolateRow_Any_SSSE3(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride_argb, int width,
int source_y_fraction);
void InterpolateRow_Any_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
void InterpolateRows_Any_MIPS_DSPR2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride_ptr, int width,
int source_y_fraction);
// Sobel images.
void SobelXRow_C(const uint8* src_y0, const uint8* src_y1, const uint8* src_y2,
......
include/libyuv/version.h
View file @ b911428a
......@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 709
#define LIBYUV_VERSION 710
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/planar_functions.cc
View file @ b911428a
......@@ -1617,7 +1617,7 @@ int ARGBShade(const uint8* src_argb, int src_stride_argb,
// Interpolate 2 ARGB images by specified amount (0 to 255).
// TODO(fbarchard): Consider selecting a specialization for interpolation so
// row function doesn't need to check interpolation on each row.
// row function doesn't need to check interpolation on each row.
LIBYUV_API
int ARGBInterpolate(const uint8* src_argb0, int src_stride_argb0,
const uint8* src_argb1, int src_stride_argb1,
......@@ -1642,46 +1642,55 @@ int ARGBInterpolate(const uint8* src_argb0, int src_stride_argb0,
width * height, 1,
interpolation);
}
void (*ARGBInterpolateRow)(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = ARGBInterpolateRow_C;
#if defined(HAS_ARGBINTERPOLATEROW_SSE2)
void (*InterpolateRow)(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_C;
#if defined(HAS_INTERPOLATEROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && width >= 4) {
ARGBInterpolateRow = ARGBInterpolateRow_Any_SSE2;
InterpolateRow = InterpolateRow_Any_SSE2;
if (IS_ALIGNED(width, 4)) {
ARGBInterpolateRow = ARGBInterpolateRow_Unaligned_SSE2;
InterpolateRow = InterpolateRow_Unaligned_SSE2;
if (IS_ALIGNED(src_argb0, 16) && IS_ALIGNED(src_stride_argb0, 16) &&
IS_ALIGNED(src_argb1, 16) && IS_ALIGNED(src_stride_argb1, 16) &&
IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
ARGBInterpolateRow = ARGBInterpolateRow_SSE2;
InterpolateRow = InterpolateRow_SSE2;
}
}
}
#endif
#if defined(HAS_ARGBINTERPOLATEROW_SSSE3)
#if defined(HAS_INTERPOLATEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width >= 4) {
ARGBInterpolateRow = ARGBInterpolateRow_Any_SSSE3;
InterpolateRow = InterpolateRow_Any_SSSE3;
if (IS_ALIGNED(width, 4)) {
ARGBInterpolateRow = ARGBInterpolateRow_Unaligned_SSSE3;
InterpolateRow = InterpolateRow_Unaligned_SSSE3;
if (IS_ALIGNED(src_argb0, 16) && IS_ALIGNED(src_stride_argb0, 16) &&
IS_ALIGNED(src_argb1, 16) && IS_ALIGNED(src_stride_argb1, 16) &&
IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
ARGBInterpolateRow = ARGBInterpolateRow_SSSE3;
InterpolateRow = InterpolateRow_SSSE3;
}
}
}
#endif
#if defined(HAS_ARGBINTERPOLATEROW_NEON)
#if defined(HAS_INTERPOLATEROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && width >= 4) {
ARGBInterpolateRow = ARGBInterpolateRow_Any_NEON;
InterpolateRow = InterpolateRow_Any_NEON;
if (IS_ALIGNED(width, 4)) {
ARGBInterpolateRow = ARGBInterpolateRow_NEON;
InterpolateRow = InterpolateRow_NEON;
}
}
#endif
#if defined(HAS_INTERPOLATEROWS_MIPS_DSPR2)
if (TestCpuFlag(kCpuHasMIPS_DSPR2) && width >= 1 &&
IS_ALIGNED(src_argb0, 4) && IS_ALIGNED(src_stride_argb0, 4) &&
IS_ALIGNED(src_argb1, 4) && IS_ALIGNED(src_stride_argb1, 4) &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) {
ScaleARGBFilterRows = InterpolateRow_MIPS_DSPR2;
}
#endif
for (int y = 0; y < height; ++y) {
ARGBInterpolateRow(dst_argb, src_argb0, src_argb1 - src_argb0,
width, interpolation);
InterpolateRow(dst_argb, src_argb0, src_argb1 - src_argb0,
width * 4, interpolation);
src_argb0 += src_stride_argb0;
src_argb1 += src_stride_argb1;
dst_argb += dst_stride_argb;
......
source/row_any.cc
View file @ b911428a
......@@ -483,29 +483,33 @@ YANY(ARGBShuffleRow_Any_NEON, ARGBShuffleRow_NEON,
#undef YANY
// Interpolate may want to work in place, so last16 method can not be used.
#define NANY(NAMEANY, ARGBTERP_SIMD, ARGBTERP_C, SBPP, BPP, MASK) \
void NAMEANY(uint8* dst_argb, const uint8* src_argb, \
ptrdiff_t src_stride_argb, int width, \
#define NANY(NAMEANY, TERP_SIMD, TERP_C, SBPP, BPP, MASK) \
void NAMEANY(uint8* dst_ptr, const uint8* src_ptr, \
ptrdiff_t src_stride_ptr, int width, \
int source_y_fraction) { \
int n = width & ~MASK; \
ARGBTERP_SIMD(dst_argb, src_argb, src_stride_argb, \
n, source_y_fraction); \
ARGBTERP_C(dst_argb + n * BPP, \
src_argb + n * SBPP, src_stride_argb, \
width & MASK, source_y_fraction); \
TERP_SIMD(dst_ptr, src_ptr, src_stride_ptr, \
n, source_y_fraction); \
TERP_C(dst_ptr + n * BPP, \
src_ptr + n * SBPP, src_stride_ptr, \
width & MASK, source_y_fraction); \
}
#ifdef HAS_ARGBINTERPOLATEROW_SSSE3
NANY(ARGBInterpolateRow_Any_SSSE3, ARGBInterpolateRow_Unaligned_SSSE3,
ARGBInterpolateRow_C, 4, 4, 3)
#ifdef HAS_INTERPOLATEROW_SSSE3
NANY(InterpolateRow_Any_SSSE3, InterpolateRow_Unaligned_SSSE3,
InterpolateRow_C, 1, 1, 15)
#endif
#ifdef HAS_ARGBINTERPOLATEROW_SSE2
NANY(ARGBInterpolateRow_Any_SSE2, ARGBInterpolateRow_Unaligned_SSE2,
ARGBInterpolateRow_C, 4, 4, 3)
#ifdef HAS_INTERPOLATEROW_SSE2
NANY(InterpolateRow_Any_SSE2, InterpolateRow_Unaligned_SSE2,
InterpolateRow_C, 1, 1, 15)
#endif
#ifdef HAS_ARGBINTERPOLATEROW_NEON
NANY(ARGBInterpolateRow_Any_NEON, ARGBInterpolateRow_NEON,
ARGBInterpolateRow_C, 4, 4, 3)
#ifdef HAS_INTERPOLATEROW_NEON
NANY(InterpolateRow_Any_NEON, InterpolateRow_NEON,
InterpolateRow_C, 1, 1, 15)
#endif
#ifdef HAS_INTERPOLATEROW_MIPS_DSPR2
NANY(InterpolateRow_Any_MIPS_DSPR2, InterpolateRow_MIPS_DSPR2,
InterpolateRow_C, 1, 1, 3)
#endif
#undef NANY
......
source/row_common.cc
View file @ b911428a
......@@ -1775,9 +1775,9 @@ void ARGBAffineRow_C(const uint8* src_argb, int src_argb_stride,
}
// C version 2x2 -> 2x1.
void ARGBInterpolateRow_C(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride,
int width, int source_y_fraction) {
void InterpolateRow_C(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride,
int width, int source_y_fraction) {
int y1_fraction = source_y_fraction;
int y0_fraction = 256 - y1_fraction;
const uint8* src_ptr1 = src_ptr + src_stride;
......@@ -1785,21 +1785,12 @@ void ARGBInterpolateRow_C(uint8* dst_ptr, const uint8* src_ptr,
for (int x = 0; x < width - 1; x += 2) {
dst_ptr[0] = (src_ptr[0] * y0_fraction + src_ptr1[0] * y1_fraction) >> 8;
dst_ptr[1] = (src_ptr[1] * y0_fraction + src_ptr1[1] * y1_fraction) >> 8;
dst_ptr[2] = (src_ptr[2] * y0_fraction + src_ptr1[2] * y1_fraction) >> 8;
dst_ptr[3] = (src_ptr[3] * y0_fraction + src_ptr1[3] * y1_fraction) >> 8;
dst_ptr[4] = (src_ptr[4] * y0_fraction + src_ptr1[4] * y1_fraction) >> 8;
dst_ptr[5] = (src_ptr[5] * y0_fraction + src_ptr1[5] * y1_fraction) >> 8;
dst_ptr[6] = (src_ptr[6] * y0_fraction + src_ptr1[6] * y1_fraction) >> 8;
dst_ptr[7] = (src_ptr[7] * y0_fraction + src_ptr1[7] * y1_fraction) >> 8;
src_ptr += 8;
src_ptr1 += 8;
dst_ptr += 8;
src_ptr += 2;
src_ptr1 += 2;
dst_ptr += 2;
}
if (width & 1) {
dst_ptr[0] = (src_ptr[0] * y0_fraction + src_ptr1[0] * y1_fraction) >> 8;
dst_ptr[1] = (src_ptr[1] * y0_fraction + src_ptr1[1] * y1_fraction) >> 8;
dst_ptr[2] = (src_ptr[2] * y0_fraction + src_ptr1[2] * y1_fraction) >> 8;
dst_ptr[3] = (src_ptr[3] * y0_fraction + src_ptr1[3] * y1_fraction) >> 8;
}
}
......
source/row_mips.cc
View file @ b911428a
......@@ -909,6 +909,63 @@ void I422ToBGRARow_MIPS_DSPR2(const uint8* y_buf,
"s4", "s5", "s6"
);
}
// Bilinear filter 8x2 -> 8x1
void InterpolateRows_MIPS_DSPR2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) {
int y0_fraction = 256 - source_y_fraction;
const uint8* src_ptr1 = src_ptr + src_stride;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"replv.ph $t0, %[y0_fraction] \n"
"replv.ph $t1, %[source_y_fraction] \n"
"1: \n"
"lw $t2, 0(%[src_ptr]) \n"
"lw $t3, 0(%[src_ptr1]) \n"
"lw $t4, 4(%[src_ptr]) \n"
"lw $t5, 4(%[src_ptr1]) \n"
"muleu_s.ph.qbl $t6, $t2, $t0 \n"
"muleu_s.ph.qbr $t7, $t2, $t0 \n"
"muleu_s.ph.qbl $t8, $t3, $t1 \n"
"muleu_s.ph.qbr $t9, $t3, $t1 \n"
"muleu_s.ph.qbl $t2, $t4, $t0 \n"
"muleu_s.ph.qbr $t3, $t4, $t0 \n"
"muleu_s.ph.qbl $t4, $t5, $t1 \n"
"muleu_s.ph.qbr $t5, $t5, $t1 \n"
"addq.ph $t6, $t6, $t8 \n"
"addq.ph $t7, $t7, $t9 \n"
"addq.ph $t2, $t2, $t4 \n"
"addq.ph $t3, $t3, $t5 \n"
"shra.ph $t6, $t6, 8 \n"
"shra.ph $t7, $t7, 8 \n"
"shra.ph $t2, $t2, 8 \n"
"shra.ph $t3, $t3, 8 \n"
"precr.qb.ph $t6, $t6, $t7 \n"
"precr.qb.ph $t2, $t2, $t3 \n"
"addiu %[src_ptr], %[src_ptr], 8 \n"
"addiu %[src_ptr1], %[src_ptr1], 8 \n"
"addiu %[dst_width], %[dst_width], -8 \n"
"sw $t6, 0(%[dst_ptr]) \n"
"sw $t2, 4(%[dst_ptr]) \n"
"bgtz %[dst_width], 1b \n"
" addiu %[dst_ptr], %[dst_ptr], 8 \n"
".set pop \n"
: [dst_ptr] "+r" (dst_ptr),
[src_ptr1] "+r" (src_ptr1),
[src_ptr] "+r" (src_ptr),
[dst_width] "+r" (dst_width)
: [source_y_fraction] "r" (source_y_fraction),
[y0_fraction] "r" (y0_fraction),
[src_stride] "r" (src_stride)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
}
#endif // __mips_dsp_rev >= 2
#ifdef __cplusplus
......
source/row_neon.cc
View file @ b911428a
......@@ -2161,11 +2161,10 @@ void RAWToYRow_NEON(const uint8* src_raw, uint8* dst_y, int pix) {
);
}
// 4x2 -> 4x1
// Same as ScaleARGBFilterRows_NEON but with last pixel not duplicated.
void ARGBInterpolateRow_NEON(uint8* dst_ptr,
const uint8* src_ptr, ptrdiff_t src_stride,
int dst_width, int source_y_fraction) {
// Bilinear filter 16x2 -> 16x1
void InterpolateRow_NEON(uint8* dst_ptr,
const uint8* src_ptr, ptrdiff_t src_stride,
int dst_width, int source_y_fraction) {
asm volatile (
"cmp %4, #0 \n"
"beq 100f \n"
......@@ -2184,7 +2183,7 @@ void ARGBInterpolateRow_NEON(uint8* dst_ptr,
"1: \n"
"vld1.u8 {q0}, [%1]! \n"
"vld1.u8 {q1}, [%2]! \n"
"subs %3, #4 \n"
"subs %3, %3, #16 \n"
"vmull.u8 q13, d0, d4 \n"
"vmull.u8 q14, d1, d4 \n"
"vmlal.u8 q13, d2, d5 \n"
......@@ -2199,7 +2198,7 @@ void ARGBInterpolateRow_NEON(uint8* dst_ptr,
"25: \n"
"vld1.u8 {q0}, [%1]! \n"
"vld1.u8 {q1}, [%2]! \n"
"subs %3, #4 \n"
"subs %3, %3, #16 \n"
"vrhadd.u8 q0, q1 \n"
"vrhadd.u8 q0, q1 \n"
"vst1.u8 {q0}, [%0]! \n"
......@@ -2210,7 +2209,7 @@ void ARGBInterpolateRow_NEON(uint8* dst_ptr,
"50: \n"
"vld1.u8 {q0}, [%1]! \n"
"vld1.u8 {q1}, [%2]! \n"
"subs %3, #4 \n"
"subs %3, %3, #16 \n"
"vrhadd.u8 q0, q1 \n"
"vst1.u8 {q0}, [%0]! \n"
"bgt 50b \n"
......@@ -2220,7 +2219,7 @@ void ARGBInterpolateRow_NEON(uint8* dst_ptr,
"75: \n"
"vld1.u8 {q1}, [%1]! \n"
"vld1.u8 {q0}, [%2]! \n"
"subs %3, #4 \n"
"subs %3, %3, #16 \n"
"vrhadd.u8 q0, q1 \n"
"vrhadd.u8 q0, q1 \n"
"vst1.u8 {q0}, [%0]! \n"
......@@ -2230,7 +2229,7 @@ void ARGBInterpolateRow_NEON(uint8* dst_ptr,
// Blend 100 / 0 - Copy row unchanged.
"100: \n"
"vld1.u8 {q0}, [%1]! \n"
"subs %3, #4 \n"
"subs %3, %3, #16 \n"
"vst1.u8 {q0}, [%0]! \n"
"bgt 100b \n"
......
source/row_posix.cc
View file @ b911428a
This diff is collapsed.
source/row_win.cc
View file @ b911428a
This diff is collapsed.
source/scale.cc
View file @ b911428a
This diff is collapsed.
source/scale_argb.cc
View file @ b911428a
......@@ -824,41 +824,51 @@ static void ScaleARGBBilinearDown(int src_height,
int xr = (dx >= 0) ? xlast : x;
xl = (xl >> 16) & ~3; // Left edge aligned.
xr = (xr >> 16) + 1; // Right most pixel used.
int clip_src_width = ((xr - xl) + 1 + 3) & ~3; // Width aligned to 4.
int clip_src_width = (((xr - xl) + 1 + 3) & ~3) * 4; // Width aligned to 4.
src_argb += xl * 4;
x -= (xl << 16);
assert(clip_src_width * 4 <= kMaxStride);
assert(clip_src_width <= kMaxStride);
// TODO(fbarchard): Remove clip_src_width alignment checks.
SIMD_ALIGNED(uint8 row[kMaxStride + 16]);
void (*ScaleARGBFilterRows)(uint8* dst_argb, const uint8* src_argb,
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
ARGBInterpolateRow_C;
#if defined(HAS_ARGBINTERPOLATEROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && clip_src_width >= 4) {
ScaleARGBFilterRows = ARGBInterpolateRow_Any_SSE2;
if (IS_ALIGNED(clip_src_width, 4)) {
ScaleARGBFilterRows = ARGBInterpolateRow_Unaligned_SSE2;
InterpolateRow_C;
#if defined(HAS_INTERPOLATEROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && clip_src_width >= 16) {
InterpolateRow = InterpolateRow_Any_SSE2;
if (IS_ALIGNED(clip_src_width, 16)) {
InterpolateRow = InterpolateRow_Unaligned_SSE2;
if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16)) {
ScaleARGBFilterRows = ARGBInterpolateRow_SSE2;
InterpolateRow = InterpolateRow_SSE2;
}
}
}
#endif
#if defined(HAS_ARGBINTERPOLATEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && clip_src_width >= 4) {
ScaleARGBFilterRows = ARGBInterpolateRow_Any_SSSE3;
if (IS_ALIGNED(clip_src_width, 4)) {
ScaleARGBFilterRows = ARGBInterpolateRow_Unaligned_SSSE3;
#if defined(HAS_INTERPOLATEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && clip_src_width >= 16) {
InterpolateRow = InterpolateRow_Any_SSSE3;
if (IS_ALIGNED(clip_src_width, 16)) {
InterpolateRow = InterpolateRow_Unaligned_SSSE3;
if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16)) {
ScaleARGBFilterRows = ARGBInterpolateRow_SSSE3;
InterpolateRow = InterpolateRow_SSSE3;
}
}
}
#endif
#if defined(HAS_ARGBINTERPOLATEROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && clip_src_width >= 4) {
ScaleARGBFilterRows = ARGBInterpolateRow_Any_NEON;
#if defined(HAS_INTERPOLATEROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && clip_src_width >= 16) {
InterpolateRow = InterpolateRow_Any_NEON;
if (IS_ALIGNED(clip_src_width, 16)) {
InterpolateRow = InterpolateRow_NEON;
}
}
#endif
#if defined(HAS_INTERPOLATEROWS_MIPS_DSPR2)
if (TestCpuFlag(kCpuHasMIPS_DSPR2) && clip_src_width >= 4 &&
IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4)) {
InterpolateRow = InterpolateRow_Any_MIPS_DSPR2;
if (IS_ALIGNED(clip_src_width, 4)) {
ScaleARGBFilterRows = ARGBInterpolateRow_NEON;
InterpolateRow = InterpolateRow_MIPS_DSPR2;
}
}
#endif
......@@ -877,7 +887,7 @@ static void ScaleARGBBilinearDown(int src_height,
int yi = y >> 16;
int yf = (y >> 8) & 255;
const uint8* src = src_argb + yi * src_stride;
ScaleARGBFilterRows(row, src, src_stride, clip_src_width, yf);
InterpolateRow(row, src, src_stride, clip_src_width, yf);
ScaleARGBFilterCols(dst_argb, row, dst_width, x, dx);
dst_argb += dst_stride;
y += dy;
......@@ -895,38 +905,44 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
assert(dst_width > 0);
assert(dst_height > 0);
assert(dst_width * 4 <= kMaxStride);
void (*ScaleARGBFilterRows)(uint8* dst_argb, const uint8* src_argb,
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
ARGBInterpolateRow_C;
#if defined(HAS_ARGBINTERPOLATEROW_SSE2)
InterpolateRow_C;
#if defined(HAS_INTERPOLATEROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && dst_width >= 4) {
ScaleARGBFilterRows = ARGBInterpolateRow_Any_SSE2;
InterpolateRow = InterpolateRow_Any_SSE2;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBFilterRows = ARGBInterpolateRow_Unaligned_SSE2;
InterpolateRow = InterpolateRow_Unaligned_SSE2;
if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
ScaleARGBFilterRows = ARGBInterpolateRow_SSE2;
InterpolateRow = InterpolateRow_SSE2;
}
}
}
#endif
#if defined(HAS_ARGBINTERPOLATEROW_SSSE3)
#if defined(HAS_INTERPOLATEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && dst_width >= 4) {
ScaleARGBFilterRows = ARGBInterpolateRow_Any_SSSE3;
InterpolateRow = InterpolateRow_Any_SSSE3;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBFilterRows = ARGBInterpolateRow_Unaligned_SSSE3;
InterpolateRow = InterpolateRow_Unaligned_SSSE3;
if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
ScaleARGBFilterRows = ARGBInterpolateRow_SSSE3;
InterpolateRow = InterpolateRow_SSSE3;
}
}
}
#endif
#if defined(HAS_ARGBINTERPOLATEROW_NEON)
#if defined(HAS_INTERPOLATEROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && dst_width >= 4) {
ScaleARGBFilterRows = ARGBInterpolateRow_Any_NEON;
InterpolateRow = InterpolateRow_Any_NEON;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBFilterRows = ARGBInterpolateRow_NEON;
InterpolateRow = InterpolateRow_NEON;
}
}
#endif
#if defined(HAS_INTERPOLATEROWS_MIPS_DSPR2)
if (TestCpuFlag(kCpuHasMIPS_DSPR2) && dst_width >= 1 &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) {
InterpolateRow = InterpolateRow_MIPS_DSPR2;
}
#endif
void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) = ScaleARGBFilterCols_C;
......@@ -965,7 +981,7 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
}
}
int yf = (y >> 8) & 255;
ScaleARGBFilterRows(dst_argb, rowptr, rowstride, dst_width, yf);
InterpolateRow(dst_argb, rowptr, rowstride, dst_width * 4, yf);
dst_argb += dst_stride;
y += dy;
}
......@@ -1024,24 +1040,23 @@ static void ScaleARGBAnySize(int src_width, int src_height,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
FilterMode filtering) {
if (!filtering ||
(src_width * 4 > kMaxStride && dst_width * 4 > kMaxStride)) {
ScaleARGBSimple(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src_argb, dst_argb,
x, dx, y, dy);
if (filtering && dy < 65536 && dst_width * 4 <= kMaxStride) {
ScaleARGBBilinearUp(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src_argb, dst_argb,
x, dx, y, dy);
return;
}
if (dy >= 65536 || dst_width * 4 > kMaxStride) {
if (filtering && src_width * 4 < kMaxStride) {
ScaleARGBBilinearDown(src_height,
clip_width, clip_height,
src_stride, dst_stride, src_argb, dst_argb,
x, dx, y, dy);
} else {
ScaleARGBBilinearUp(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src_argb, dst_argb,
x, dx, y, dy);
return;
}
ScaleARGBSimple(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src_argb, dst_argb,
x, dx, y, dy);
}
// ScaleARGB a ARGB.
......
source/scale_mips.cc
View file @ b911428a
......@@ -629,64 +629,6 @@ void ScaleRowDown38_3_Box_MIPS_DSPR2(const uint8* src_ptr,
);
}
void ScaleFilterRows_MIPS_DSPR2(unsigned char *dst_ptr,
const unsigned char* src_ptr,
ptrdiff_t src_stride,
int dst_width, int source_y_fraction) {
int y0_fraction = 256 - source_y_fraction;
const unsigned char* src_ptr1 = src_ptr + src_stride;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"replv.ph $t0, %[y0_fraction] \n"
"replv.ph $t1, %[source_y_fraction] \n"
"1: \n"
"lw $t2, 0(%[src_ptr]) \n"
"lw $t3, 0(%[src_ptr1]) \n"
"lw $t4, 4(%[src_ptr]) \n"
"lw $t5, 4(%[src_ptr1]) \n"
"muleu_s.ph.qbl $t6, $t2, $t0 \n"
"muleu_s.ph.qbr $t7, $t2, $t0 \n"
"muleu_s.ph.qbl $t8, $t3, $t1 \n"
"muleu_s.ph.qbr $t9, $t3, $t1 \n"
"muleu_s.ph.qbl $t2, $t4, $t0 \n"
"muleu_s.ph.qbr $t3, $t4, $t0 \n"
"muleu_s.ph.qbl $t4, $t5, $t1 \n"
"muleu_s.ph.qbr $t5, $t5, $t1 \n"
"addq.ph $t6, $t6, $t8 \n"
"addq.ph $t7, $t7, $t9 \n"
"addq.ph $t2, $t2, $t4 \n"
"addq.ph $t3, $t3, $t5 \n"
"shra.ph $t6, $t6, 8 \n"
"shra.ph $t7, $t7, 8 \n"
"shra.ph $t2, $t2, 8 \n"
"shra.ph $t3, $t3, 8 \n"
"precr.qb.ph $t6, $t6, $t7 \n"
"precr.qb.ph $t2, $t2, $t3 \n"
"addiu %[src_ptr], %[src_ptr], 8 \n"
"addiu %[src_ptr1], %[src_ptr1], 8 \n"
"addiu %[dst_width], %[dst_width], -8 \n"
"sw $t6, 0(%[dst_ptr]) \n"
"sw $t2, 4(%[dst_ptr]) \n"
"bgtz %[dst_width], 1b \n"
" addiu %[dst_ptr], %[dst_ptr], 8 \n"
"lbu $t0, -1(%[dst_ptr]) \n"
"sb $t0, 0(%[dst_ptr]) \n"
".set pop \n"
: [dst_ptr] "+r" (dst_ptr),
[src_ptr1] "+r" (src_ptr1),
[src_ptr] "+r" (src_ptr),
[dst_width] "+r" (dst_width)
: [source_y_fraction] "r" (source_y_fraction),
[y0_fraction] "r" (y0_fraction),
[src_stride] "r" (src_stride)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
}
#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
#ifdef __cplusplus
......
unit_test/scale_argb_test.cc
View file @ b911428a
......@@ -117,7 +117,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy2_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy2_Bilinear) {
......@@ -143,7 +143,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy1_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 0);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy1_Bilinear) {
......@@ -156,7 +156,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy1_Bilinear) {
dst_width, dst_height,
kFilterBilinear,
benchmark_iterations_);
EXPECT_LE(max_diff, 0);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy4_None) {
......@@ -169,7 +169,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy4_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy4_Bilinear) {
......@@ -195,7 +195,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy5_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy5_Bilinear) {
......@@ -221,7 +221,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy8_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy8_Bilinear) {
......@@ -247,7 +247,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy16_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy16_Bilinear) {
......@@ -263,6 +263,32 @@ TEST_F(libyuvTest, ARGBScaleDownBy16_Bilinear) {
EXPECT_LE(max_diff, 2);
}
TEST_F(libyuvTest, ARGBScaleDownBy23_None) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
const int dst_width = Abs(src_width) * 2 / 3;
const int dst_height = Abs(src_height) * 2 / 3;
int max_diff = ARGBTestFilter(src_width, src_height,
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy23_Bilinear) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
const int dst_width = Abs(src_width) * 2 / 3;
const int dst_height = Abs(src_height) * 2 / 3;
int max_diff = ARGBTestFilter(src_width, src_height,
dst_width, dst_height,
kFilterBilinear,
benchmark_iterations_);
EXPECT_LE(max_diff, 2);
}
TEST_F(libyuvTest, ARGBScaleDownBy34_None) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
......@@ -273,7 +299,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy34_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy34_Bilinear) {
......@@ -299,7 +325,7 @@ TEST_F(libyuvTest, ARGBScaleDownBy38_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleDownBy38_Bilinear) {
......@@ -325,7 +351,7 @@ TEST_F(libyuvTest, ARGBScaleTo1366x768_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleTo1366x768_Bilinear) {
......@@ -352,7 +378,7 @@ TEST_F(libyuvTest, ARGBScaleTo1280x720_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleTo1280x720_Bilinear) {
......@@ -378,7 +404,7 @@ TEST_F(libyuvTest, ARGBScaleTo853x480_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleTo853x480_Bilinear) {
......@@ -404,7 +430,7 @@ TEST_F(libyuvTest, ARGBScaleFrom640x360_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleFrom640x360_Bilinear) {
......@@ -675,6 +701,32 @@ TEST_F(libyuvTest, ARGBScaleClipDownBy16_Bilinear) {
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleClipDownBy23_None) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
const int dst_width = Abs(src_width) * 2 / 3;
const int dst_height = Abs(src_height) * 2 / 3;
int max_diff = ARGBClipTestFilter(src_width, src_height,
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleClipDownBy23_Bilinear) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
const int dst_width = Abs(src_width) * 2 / 3;
const int dst_height = Abs(src_height) * 2 / 3;
int max_diff = ARGBClipTestFilter(src_width, src_height,
dst_width, dst_height,
kFilterBilinear,
benchmark_iterations_);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ARGBScaleClipDownBy34_None) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
......
unit_test/scale_test.cc
View file @ b911428a
......@@ -175,7 +175,7 @@ TEST_F(libyuvTest, ScaleDownBy2_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleDownBy2_Bilinear) {
......@@ -214,7 +214,7 @@ TEST_F(libyuvTest, ScaleDownBy4_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 2); // This is the only scale factor with error of 2.
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleDownBy4_Bilinear) {
......@@ -253,7 +253,7 @@ TEST_F(libyuvTest, ScaleDownBy5_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleDownBy5_Bilinear) {
......@@ -292,7 +292,7 @@ TEST_F(libyuvTest, ScaleDownBy8_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleDownBy8_Bilinear) {
......@@ -331,7 +331,7 @@ TEST_F(libyuvTest, ScaleDownBy16_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleDownBy16_Bilinear) {
......@@ -344,7 +344,7 @@ TEST_F(libyuvTest, ScaleDownBy16_Bilinear) {
dst_width, dst_height,
kFilterBilinear,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_LE(max_diff, 2);
}
TEST_F(libyuvTest, ScaleDownBy16_Box) {
......@@ -360,6 +360,32 @@ TEST_F(libyuvTest, ScaleDownBy16_Box) {
EXPECT_LE(max_diff, 1);
}
TEST_F(libyuvTest, ScaleDownBy23_None) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
const int dst_width = Abs(src_width) * 2 / 3;
const int dst_height = Abs(src_height) * 2 / 3;
int max_diff = TestFilter(src_width, src_height,
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleDownBy23_Bilinear) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
const int dst_width = Abs(src_width) * 2 / 3;
const int dst_height = Abs(src_height) * 2 / 3;
int max_diff = TestFilter(src_width, src_height,
dst_width, dst_height,
kFilterBilinear,
benchmark_iterations_);
EXPECT_LE(max_diff, 2);
}
TEST_F(libyuvTest, ScaleDownBy34_None) {
const int src_width = benchmark_width_;
const int src_height = benchmark_height_;
......@@ -370,7 +396,7 @@ TEST_F(libyuvTest, ScaleDownBy34_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleDownBy34_Bilinear) {
......@@ -409,7 +435,7 @@ TEST_F(libyuvTest, ScaleDownBy38_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleDownBy38_Bilinear) {
......@@ -448,7 +474,7 @@ TEST_F(libyuvTest, ScaleTo1366x768_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleTo1366x768_Bilinear) {
......@@ -487,7 +513,7 @@ TEST_F(libyuvTest, ScaleTo1280x720_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleTo1280x720_Bilinear) {
......@@ -526,7 +552,7 @@ TEST_F(libyuvTest, ScaleTo853x480_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 1);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleTo853x480_Bilinear) {
......@@ -565,7 +591,7 @@ TEST_F(libyuvTest, ScaleFrom640x360_None) {
dst_width, dst_height,
kFilterNone,
benchmark_iterations_);
EXPECT_LE(max_diff, 2);
EXPECT_EQ(0, max_diff);
}
TEST_F(libyuvTest, ScaleFrom640x360_Bilinear) {
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment