< prev index next >
src/java.desktop/share/native/libmlib_image/mlib_c_ImageAffine_BC.c
Print this page
rev 59383 : [mq]: final
*** 1,7 ****
/*
! * Copyright (c) 1998, 2003, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation. Oracle designates this
--- 1,7 ----
/*
! * Copyright (c) 1998, 2020, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation. Oracle designates this
*** 69,600 ****
#define FUN_NAME(CHAN) mlib_ImageAffine_u8_##CHAN##_bc
#define FILTER_BITS 8
/***************************************************************/
! #ifdef __sparc /* for SPARC, using floating-point multiplies is faster */
!
! #undef FILTER_ELEM_BITS
! #define FILTER_ELEM_BITS 4
!
! #ifdef MLIB_USE_FTOI_CLAMPING
!
! #define SAT8(DST) \
! DST = ((mlib_s32)(val0 - sat) >> 24) ^ 0x80
!
! #else
!
! #define SAT8(DST) \
! val0 -= sat; \
! if (val0 >= MLIB_S32_MAX) \
! DST = MLIB_U8_MAX; \
! else if (val0 <= MLIB_S32_MIN) \
! DST = MLIB_U8_MIN; \
! else \
! DST = ((mlib_s32)val0 >> 24) ^ 0x80
!
! #endif /* MLIB_USE_FTOI_CLAMPING */
!
! /***************************************************************/
! mlib_status FUN_NAME(1ch)(mlib_affine_param *param)
! {
! DECLAREVAR_BC();
! DTYPE *dstLineEnd;
! mlib_d64 sat = (mlib_d64) 0x7F800000;
! const mlib_f32 *mlib_filters_table;
!
! if (filter == MLIB_BICUBIC) {
! mlib_filters_table = mlib_filters_u8f_bc;
! }
! else {
! mlib_filters_table = mlib_filters_u8f_bc2;
! }
!
! for (j = yStart; j <= yFinish; j++) {
! mlib_d64 xf0, xf1, xf2, xf3;
! mlib_d64 yf0, yf1, yf2, yf3;
! mlib_d64 c0, c1, c2, c3, val0;
! mlib_s32 filterpos;
! mlib_f32 *fptr;
! mlib_u8 s0, s1, s2, s3;
!
! CLIP(1);
! dstLineEnd = (DTYPE *) dstData + xRight;
!
! filterpos = (X >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! xf0 = fptr[0];
! xf1 = fptr[1];
! xf2 = fptr[2];
! xf3 = fptr[3];
!
! filterpos = (Y >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! yf0 = fptr[0];
! yf1 = fptr[1];
! yf2 = fptr[2];
! yf3 = fptr[3];
!
! xSrc = (X >> MLIB_SHIFT) - 1;
! ySrc = (Y >> MLIB_SHIFT) - 1;
!
! srcPixelPtr = ((DTYPE **) lineAddr)[ySrc] + xSrc;
! s0 = srcPixelPtr[0];
! s1 = srcPixelPtr[1];
! s2 = srcPixelPtr[2];
! s3 = srcPixelPtr[3];
!
! #ifdef __SUNPRO_C
! #pragma pipeloop(0)
! #endif /* __SUNPRO_C */
! for (; dstPixelPtr <= (dstLineEnd - 1); dstPixelPtr++) {
! X += dX;
! Y += dY;
!
! c0 = (mlib_U82D64[s0] * xf0 + mlib_U82D64[s1] * xf1 +
! mlib_U82D64[s2] * xf2 + mlib_U82D64[s3] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c1 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[1]] * xf1 +
! mlib_U82D64[srcPixelPtr[2]] * xf2 + mlib_U82D64[srcPixelPtr[3]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c2 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[1]] * xf1 +
! mlib_U82D64[srcPixelPtr[2]] * xf2 + mlib_U82D64[srcPixelPtr[3]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c3 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[1]] * xf1 +
! mlib_U82D64[srcPixelPtr[2]] * xf2 + mlib_U82D64[srcPixelPtr[3]] * xf3);
!
! filterpos = (X >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! xf0 = fptr[0];
! xf1 = fptr[1];
! xf2 = fptr[2];
! xf3 = fptr[3];
!
! val0 = (c0 * yf0 + c1 * yf1 + c2 * yf2 + c3 * yf3);
!
! filterpos = (Y >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! yf0 = fptr[0];
! yf1 = fptr[1];
! yf2 = fptr[2];
! yf3 = fptr[3];
!
! SAT8(dstPixelPtr[0]);
!
! xSrc = (X >> MLIB_SHIFT) - 1;
! ySrc = (Y >> MLIB_SHIFT) - 1;
!
! srcPixelPtr = ((DTYPE **) lineAddr)[ySrc] + xSrc;
! s0 = srcPixelPtr[0];
! s1 = srcPixelPtr[1];
! s2 = srcPixelPtr[2];
! s3 = srcPixelPtr[3];
! }
!
! c0 = (mlib_U82D64[s0] * xf0 + mlib_U82D64[s1] * xf1 +
! mlib_U82D64[s2] * xf2 + mlib_U82D64[s3] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c1 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[1]] * xf1 +
! mlib_U82D64[srcPixelPtr[2]] * xf2 + mlib_U82D64[srcPixelPtr[3]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c2 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[1]] * xf1 +
! mlib_U82D64[srcPixelPtr[2]] * xf2 + mlib_U82D64[srcPixelPtr[3]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c3 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[1]] * xf1 +
! mlib_U82D64[srcPixelPtr[2]] * xf2 + mlib_U82D64[srcPixelPtr[3]] * xf3);
!
! val0 = (c0 * yf0 + c1 * yf1 + c2 * yf2 + c3 * yf3);
!
! SAT8(dstPixelPtr[0]);
! }
!
! return MLIB_SUCCESS;
! }
!
! /***************************************************************/
! mlib_status FUN_NAME(2ch)(mlib_affine_param *param)
! {
! DECLAREVAR_BC();
! DTYPE *dstLineEnd;
! mlib_d64 sat = (mlib_d64) 0x7F800000;
! const mlib_f32 *mlib_filters_table;
!
! if (filter == MLIB_BICUBIC) {
! mlib_filters_table = mlib_filters_u8f_bc;
! }
! else {
! mlib_filters_table = mlib_filters_u8f_bc2;
! }
!
! for (j = yStart; j <= yFinish; j++) {
! mlib_d64 xf0, xf1, xf2, xf3;
! mlib_d64 yf0, yf1, yf2, yf3;
! mlib_d64 c0, c1, c2, c3, val0;
! mlib_s32 filterpos, k;
! mlib_f32 *fptr;
! mlib_u8 s0, s1, s2, s3;
!
! CLIP(2);
! dstLineEnd = (DTYPE *) dstData + 2 * xRight;
!
! for (k = 0; k < 2; k++) {
! mlib_s32 X1 = X;
! mlib_s32 Y1 = Y;
! DTYPE *dPtr = dstPixelPtr + k;
!
! filterpos = (X1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! xf0 = fptr[0];
! xf1 = fptr[1];
! xf2 = fptr[2];
! xf3 = fptr[3];
!
! filterpos = (Y1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! yf0 = fptr[0];
! yf1 = fptr[1];
! yf2 = fptr[2];
! yf3 = fptr[3];
!
! xSrc = (X1 >> MLIB_SHIFT) - 1;
! ySrc = (Y1 >> MLIB_SHIFT) - 1;
!
! srcPixelPtr = ((DTYPE **) lineAddr)[ySrc] + 2 * xSrc + k;
! s0 = srcPixelPtr[0];
! s1 = srcPixelPtr[2];
! s2 = srcPixelPtr[4];
! s3 = srcPixelPtr[6];
!
! #ifdef __SUNPRO_C
! #pragma pipeloop(0)
! #endif /* __SUNPRO_C */
! for (; dPtr <= (dstLineEnd - 1); dPtr += 2) {
! X1 += dX;
! Y1 += dY;
!
! c0 = (mlib_U82D64[s0] * xf0 + mlib_U82D64[s1] * xf1 +
! mlib_U82D64[s2] * xf2 + mlib_U82D64[s3] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c1 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[2]] * xf1 +
! mlib_U82D64[srcPixelPtr[4]] * xf2 + mlib_U82D64[srcPixelPtr[6]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c2 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[2]] * xf1 +
! mlib_U82D64[srcPixelPtr[4]] * xf2 + mlib_U82D64[srcPixelPtr[6]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c3 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[2]] * xf1 +
! mlib_U82D64[srcPixelPtr[4]] * xf2 + mlib_U82D64[srcPixelPtr[6]] * xf3);
!
! filterpos = (X1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! xf0 = fptr[0];
! xf1 = fptr[1];
! xf2 = fptr[2];
! xf3 = fptr[3];
!
! val0 = (c0 * yf0 + c1 * yf1 + c2 * yf2 + c3 * yf3);
!
! filterpos = (Y1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! yf0 = fptr[0];
! yf1 = fptr[1];
! yf2 = fptr[2];
! yf3 = fptr[3];
!
! SAT8(dPtr[0]);
!
! xSrc = (X1 >> MLIB_SHIFT) - 1;
! ySrc = (Y1 >> MLIB_SHIFT) - 1;
!
! srcPixelPtr = ((DTYPE **) lineAddr)[ySrc] + 2 * xSrc + k;
! s0 = srcPixelPtr[0];
! s1 = srcPixelPtr[2];
! s2 = srcPixelPtr[4];
! s3 = srcPixelPtr[6];
! }
!
! c0 = (mlib_U82D64[s0] * xf0 + mlib_U82D64[s1] * xf1 +
! mlib_U82D64[s2] * xf2 + mlib_U82D64[s3] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c1 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[2]] * xf1 +
! mlib_U82D64[srcPixelPtr[4]] * xf2 + mlib_U82D64[srcPixelPtr[6]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c2 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[2]] * xf1 +
! mlib_U82D64[srcPixelPtr[4]] * xf2 + mlib_U82D64[srcPixelPtr[6]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c3 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[2]] * xf1 +
! mlib_U82D64[srcPixelPtr[4]] * xf2 + mlib_U82D64[srcPixelPtr[6]] * xf3);
!
! val0 = (c0 * yf0 + c1 * yf1 + c2 * yf2 + c3 * yf3);
!
! SAT8(dPtr[0]);
! }
! }
!
! return MLIB_SUCCESS;
! }
!
! /***************************************************************/
! mlib_status FUN_NAME(3ch)(mlib_affine_param *param)
! {
! DECLAREVAR_BC();
! DTYPE *dstLineEnd;
! mlib_d64 sat = (mlib_d64) 0x7F800000;
! const mlib_f32 *mlib_filters_table;
!
! if (filter == MLIB_BICUBIC) {
! mlib_filters_table = mlib_filters_u8f_bc;
! }
! else {
! mlib_filters_table = mlib_filters_u8f_bc2;
! }
!
! for (j = yStart; j <= yFinish; j++) {
! mlib_d64 xf0, xf1, xf2, xf3;
! mlib_d64 yf0, yf1, yf2, yf3;
! mlib_d64 c0, c1, c2, c3, val0;
! mlib_s32 filterpos, k;
! mlib_f32 *fptr;
! mlib_u8 s0, s1, s2, s3;
!
! CLIP(3);
! dstLineEnd = (DTYPE *) dstData + 3 * xRight;
!
! for (k = 0; k < 3; k++) {
! mlib_s32 X1 = X;
! mlib_s32 Y1 = Y;
! DTYPE *dPtr = dstPixelPtr + k;
!
! filterpos = (X1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! xf0 = fptr[0];
! xf1 = fptr[1];
! xf2 = fptr[2];
! xf3 = fptr[3];
!
! filterpos = (Y1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! yf0 = fptr[0];
! yf1 = fptr[1];
! yf2 = fptr[2];
! yf3 = fptr[3];
!
! xSrc = (X1 >> MLIB_SHIFT) - 1;
! ySrc = (Y1 >> MLIB_SHIFT) - 1;
!
! srcPixelPtr = ((DTYPE **) lineAddr)[ySrc] + 3 * xSrc + k;
! s0 = srcPixelPtr[0];
! s1 = srcPixelPtr[3];
! s2 = srcPixelPtr[6];
! s3 = srcPixelPtr[9];
!
! #ifdef __SUNPRO_C
! #pragma pipeloop(0)
! #endif /* __SUNPRO_C */
! for (; dPtr <= (dstLineEnd - 1); dPtr += 3) {
! X1 += dX;
! Y1 += dY;
!
! c0 = (mlib_U82D64[s0] * xf0 + mlib_U82D64[s1] * xf1 +
! mlib_U82D64[s2] * xf2 + mlib_U82D64[s3] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c1 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[3]] * xf1 +
! mlib_U82D64[srcPixelPtr[6]] * xf2 + mlib_U82D64[srcPixelPtr[9]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c2 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[3]] * xf1 +
! mlib_U82D64[srcPixelPtr[6]] * xf2 + mlib_U82D64[srcPixelPtr[9]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c3 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[3]] * xf1 +
! mlib_U82D64[srcPixelPtr[6]] * xf2 + mlib_U82D64[srcPixelPtr[9]] * xf3);
!
! filterpos = (X1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! xf0 = fptr[0];
! xf1 = fptr[1];
! xf2 = fptr[2];
! xf3 = fptr[3];
!
! val0 = (c0 * yf0 + c1 * yf1 + c2 * yf2 + c3 * yf3);
!
! filterpos = (Y1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! yf0 = fptr[0];
! yf1 = fptr[1];
! yf2 = fptr[2];
! yf3 = fptr[3];
!
! SAT8(dPtr[0]);
!
! xSrc = (X1 >> MLIB_SHIFT) - 1;
! ySrc = (Y1 >> MLIB_SHIFT) - 1;
!
! srcPixelPtr = ((DTYPE **) lineAddr)[ySrc] + 3 * xSrc + k;
! s0 = srcPixelPtr[0];
! s1 = srcPixelPtr[3];
! s2 = srcPixelPtr[6];
! s3 = srcPixelPtr[9];
! }
!
! c0 = (mlib_U82D64[s0] * xf0 + mlib_U82D64[s1] * xf1 +
! mlib_U82D64[s2] * xf2 + mlib_U82D64[s3] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c1 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[3]] * xf1 +
! mlib_U82D64[srcPixelPtr[6]] * xf2 + mlib_U82D64[srcPixelPtr[9]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c2 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[3]] * xf1 +
! mlib_U82D64[srcPixelPtr[6]] * xf2 + mlib_U82D64[srcPixelPtr[9]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c3 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[3]] * xf1 +
! mlib_U82D64[srcPixelPtr[6]] * xf2 + mlib_U82D64[srcPixelPtr[9]] * xf3);
!
! val0 = (c0 * yf0 + c1 * yf1 + c2 * yf2 + c3 * yf3);
!
! SAT8(dPtr[0]);
! }
! }
!
! return MLIB_SUCCESS;
! }
!
! /***************************************************************/
! mlib_status FUN_NAME(4ch)(mlib_affine_param *param)
! {
! DECLAREVAR_BC();
! DTYPE *dstLineEnd;
! mlib_d64 sat = (mlib_d64) 0x7F800000;
! const mlib_f32 *mlib_filters_table;
!
! if (filter == MLIB_BICUBIC) {
! mlib_filters_table = mlib_filters_u8f_bc;
! }
! else {
! mlib_filters_table = mlib_filters_u8f_bc2;
! }
!
! for (j = yStart; j <= yFinish; j++) {
! mlib_d64 xf0, xf1, xf2, xf3;
! mlib_d64 yf0, yf1, yf2, yf3;
! mlib_d64 c0, c1, c2, c3, val0;
! mlib_s32 filterpos, k;
! mlib_f32 *fptr;
! mlib_u8 s0, s1, s2, s3;
!
! CLIP(4);
! dstLineEnd = (DTYPE *) dstData + 4 * xRight;
!
! for (k = 0; k < 4; k++) {
! mlib_s32 X1 = X;
! mlib_s32 Y1 = Y;
! DTYPE *dPtr = dstPixelPtr + k;
!
! filterpos = (X1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! xf0 = fptr[0];
! xf1 = fptr[1];
! xf2 = fptr[2];
! xf3 = fptr[3];
!
! filterpos = (Y1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! yf0 = fptr[0];
! yf1 = fptr[1];
! yf2 = fptr[2];
! yf3 = fptr[3];
!
! xSrc = (X1 >> MLIB_SHIFT) - 1;
! ySrc = (Y1 >> MLIB_SHIFT) - 1;
!
! srcPixelPtr = ((DTYPE **) lineAddr)[ySrc] + 4 * xSrc + k;
! s0 = srcPixelPtr[0];
! s1 = srcPixelPtr[4];
! s2 = srcPixelPtr[8];
! s3 = srcPixelPtr[12];
!
! #ifdef __SUNPRO_C
! #pragma pipeloop(0)
! #endif /* __SUNPRO_C */
! for (; dPtr <= (dstLineEnd - 1); dPtr += 4) {
! X1 += dX;
! Y1 += dY;
!
! c0 = (mlib_U82D64[s0] * xf0 + mlib_U82D64[s1] * xf1 +
! mlib_U82D64[s2] * xf2 + mlib_U82D64[s3] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c1 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[4]] * xf1 +
! mlib_U82D64[srcPixelPtr[8]] * xf2 + mlib_U82D64[srcPixelPtr[12]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c2 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[4]] * xf1 +
! mlib_U82D64[srcPixelPtr[8]] * xf2 + mlib_U82D64[srcPixelPtr[12]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c3 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[4]] * xf1 +
! mlib_U82D64[srcPixelPtr[8]] * xf2 + mlib_U82D64[srcPixelPtr[12]] * xf3);
!
! filterpos = (X1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! xf0 = fptr[0];
! xf1 = fptr[1];
! xf2 = fptr[2];
! xf3 = fptr[3];
!
! val0 = (c0 * yf0 + c1 * yf1 + c2 * yf2 + c3 * yf3);
!
! filterpos = (Y1 >> FILTER_SHIFT) & FILTER_MASK;
! fptr = (mlib_f32 *) ((mlib_u8 *) mlib_filters_table + filterpos);
!
! yf0 = fptr[0];
! yf1 = fptr[1];
! yf2 = fptr[2];
! yf3 = fptr[3];
!
! SAT8(dPtr[0]);
!
! xSrc = (X1 >> MLIB_SHIFT) - 1;
! ySrc = (Y1 >> MLIB_SHIFT) - 1;
!
! srcPixelPtr = ((DTYPE **) lineAddr)[ySrc] + 4 * xSrc + k;
! s0 = srcPixelPtr[0];
! s1 = srcPixelPtr[4];
! s2 = srcPixelPtr[8];
! s3 = srcPixelPtr[12];
! }
!
! c0 = (mlib_U82D64[s0] * xf0 + mlib_U82D64[s1] * xf1 +
! mlib_U82D64[s2] * xf2 + mlib_U82D64[s3] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c1 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[4]] * xf1 +
! mlib_U82D64[srcPixelPtr[8]] * xf2 + mlib_U82D64[srcPixelPtr[12]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c2 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[4]] * xf1 +
! mlib_U82D64[srcPixelPtr[8]] * xf2 + mlib_U82D64[srcPixelPtr[12]] * xf3);
! srcPixelPtr = (DTYPE *) ((mlib_addr) srcPixelPtr + srcYStride);
! c3 = (mlib_U82D64[srcPixelPtr[0]] * xf0 + mlib_U82D64[srcPixelPtr[4]] * xf1 +
! mlib_U82D64[srcPixelPtr[8]] * xf2 + mlib_U82D64[srcPixelPtr[12]] * xf3);
!
! val0 = (c0 * yf0 + c1 * yf1 + c2 * yf2 + c3 * yf3);
!
! SAT8(dPtr[0]);
! }
! }
!
! return MLIB_SUCCESS;
! }
!
! #else /* for x86, using integer multiplies is faster */
#define SHIFT_X 12
#define ROUND_X 0 /* (1 << (SHIFT_X - 1)) */
#define SHIFT_Y (14 + 14 - SHIFT_X)
--- 69,79 ----
#define FUN_NAME(CHAN) mlib_ImageAffine_u8_##CHAN##_bc
#define FILTER_BITS 8
/***************************************************************/
! /* for x86, using integer multiplies is faster */
#define SHIFT_X 12
#define ROUND_X 0 /* (1 << (SHIFT_X - 1)) */
#define SHIFT_Y (14 + 14 - SHIFT_X)
*** 668,680 ****
s0 = srcPixelPtr[0];
s1 = srcPixelPtr[1];
s2 = srcPixelPtr[2];
s3 = srcPixelPtr[3];
- #ifdef __SUNPRO_C
- #pragma pipeloop(0)
- #endif /* __SUNPRO_C */
for (; dstPixelPtr <= (dstLineEnd - 1); dstPixelPtr++) {
X += dX;
Y += dY;
c0 = (s0 * xf0 + s1 * xf1 + s2 * xf2 + s3 * xf3 + ROUND_X) >> SHIFT_X;
--- 147,156 ----
*** 790,802 ****
s0 = srcPixelPtr[0];
s1 = srcPixelPtr[2];
s2 = srcPixelPtr[4];
s3 = srcPixelPtr[6];
- #ifdef __SUNPRO_C
- #pragma pipeloop(0)
- #endif /* __SUNPRO_C */
for (; dPtr <= (dstLineEnd - 1); dPtr += 2) {
X1 += dX;
Y1 += dY;
c0 = (s0 * xf0 + s1 * xf1 + s2 * xf2 + s3 * xf3 + ROUND_X) >> SHIFT_X;
--- 266,275 ----
*** 913,925 ****
s0 = srcPixelPtr[0];
s1 = srcPixelPtr[3];
s2 = srcPixelPtr[6];
s3 = srcPixelPtr[9];
- #ifdef __SUNPRO_C
- #pragma pipeloop(0)
- #endif /* __SUNPRO_C */
for (; dPtr <= (dstLineEnd - 1); dPtr += 3) {
X1 += dX;
Y1 += dY;
c0 = (s0 * xf0 + s1 * xf1 + s2 * xf2 + s3 * xf3 + ROUND_X) >> SHIFT_X;
--- 386,395 ----
*** 1036,1048 ****
s0 = srcPixelPtr[0];
s1 = srcPixelPtr[4];
s2 = srcPixelPtr[8];
s3 = srcPixelPtr[12];
- #ifdef __SUNPRO_C
- #pragma pipeloop(0)
- #endif /* __SUNPRO_C */
for (; dPtr <= (dstLineEnd - 1); dPtr += 4) {
X1 += dX;
Y1 += dY;
c0 = (s0 * xf0 + s1 * xf1 + s2 * xf2 + s3 * xf3 + ROUND_X) >> SHIFT_X;
--- 506,515 ----
*** 1104,1111 ****
}
return MLIB_SUCCESS;
}
- #endif /* __sparc ( for SPARC, using floating-point multiplies is faster ) */
-
/***************************************************************/
--- 571,576 ----
< prev index next >