michael@0: 
michael@0: #include <arm_neon.h>
michael@0: 
michael@0: 
michael@0: #define SCALE_NOFILTER_NAME     MAKENAME(_nofilter_scale)
michael@0: #define SCALE_FILTER_NAME       MAKENAME(_filter_scale)
michael@0: #define AFFINE_NOFILTER_NAME    MAKENAME(_nofilter_affine)
michael@0: #define AFFINE_FILTER_NAME      MAKENAME(_filter_affine)
michael@0: #define PERSP_NOFILTER_NAME     MAKENAME(_nofilter_persp)
michael@0: #define PERSP_FILTER_NAME       MAKENAME(_filter_persp)
michael@0: 
michael@0: #define PACK_FILTER_X_NAME  MAKENAME(_pack_filter_x)
michael@0: #define PACK_FILTER_Y_NAME  MAKENAME(_pack_filter_y)
michael@0: #define PACK_FILTER_X4_NAME MAKENAME(_pack_filter_x4)
michael@0: #define PACK_FILTER_Y4_NAME MAKENAME(_pack_filter_y4)
michael@0: 
michael@0: #ifndef PREAMBLE
michael@0:     #define PREAMBLE(state)
michael@0:     #define PREAMBLE_PARAM_X
michael@0:     #define PREAMBLE_PARAM_Y
michael@0:     #define PREAMBLE_ARG_X
michael@0:     #define PREAMBLE_ARG_Y
michael@0: #endif
michael@0: 
michael@0: static void SCALE_NOFILTER_NAME(const SkBitmapProcState& s,
michael@0:                                 uint32_t xy[], int count, int x, int y) {
michael@0:     SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
michael@0:                              SkMatrix::kScale_Mask)) == 0);
michael@0: 
michael@0:     PREAMBLE(s);
michael@0: 
michael@0:     // we store y, x, x, x, x, x
michael@0:     const unsigned maxX = s.fBitmap->width() - 1;
michael@0:     SkFractionalInt fx;
michael@0:     {
michael@0:         SkPoint pt;
michael@0:         s.fInvProc(s.fInvMatrix, SkIntToScalar(x) + SK_ScalarHalf,
michael@0:                                  SkIntToScalar(y) + SK_ScalarHalf, &pt);
michael@0:         fx = SkScalarToFractionalInt(pt.fY);
michael@0:         const unsigned maxY = s.fBitmap->height() - 1;
michael@0:         *xy++ = TILEY_PROCF(SkFractionalIntToFixed(fx), maxY);
michael@0:         fx = SkScalarToFractionalInt(pt.fX);
michael@0:     }
michael@0: 
michael@0:     if (0 == maxX) {
michael@0:         // all of the following X values must be 0
michael@0:         memset(xy, 0, count * sizeof(uint16_t));
michael@0:         return;
michael@0:     }
michael@0: 
michael@0:     const SkFractionalInt dx = s.fInvSxFractionalInt;
michael@0: 
michael@0: #ifdef CHECK_FOR_DECAL
michael@0:     // test if we don't need to apply the tile proc
michael@0:     if (can_truncate_to_fixed_for_decal(fx, dx, count, maxX)) {
michael@0:         decal_nofilter_scale_neon(xy, SkFractionalIntToFixed(fx),
michael@0:                              SkFractionalIntToFixed(dx), count);
michael@0:         return;
michael@0:     }
michael@0: #endif
michael@0: 
michael@0:     if (count >= 8) {
michael@0:         SkFractionalInt dx2 = dx+dx;
michael@0:         SkFractionalInt dx4 = dx2+dx2;
michael@0:         SkFractionalInt dx8 = dx4+dx4;
michael@0: 
michael@0:         // now build fx/fx+dx/fx+2dx/fx+3dx
michael@0:         SkFractionalInt fx1, fx2, fx3;
michael@0:         int32x4_t lbase, hbase;
michael@0:         int16_t *dst16 = (int16_t *)xy;
michael@0: 
michael@0:         fx1 = fx+dx;
michael@0:         fx2 = fx1+dx;
michael@0:         fx3 = fx2+dx;
michael@0: 
michael@0:         lbase = vdupq_n_s32(SkFractionalIntToFixed(fx));
michael@0:         lbase = vsetq_lane_s32(SkFractionalIntToFixed(fx1), lbase, 1);
michael@0:         lbase = vsetq_lane_s32(SkFractionalIntToFixed(fx2), lbase, 2);
michael@0:         lbase = vsetq_lane_s32(SkFractionalIntToFixed(fx3), lbase, 3);
michael@0:         hbase = vaddq_s32(lbase, vdupq_n_s32(SkFractionalIntToFixed(dx4)));
michael@0: 
michael@0:         // store & bump
michael@0:         while (count >= 8) {
michael@0: 
michael@0:             int16x8_t fx8;
michael@0: 
michael@0:             fx8 = TILEX_PROCF_NEON8(lbase, hbase, maxX);
michael@0: 
michael@0:             vst1q_s16(dst16, fx8);
michael@0: 
michael@0:             // but preserving base & on to the next
michael@0:             lbase = vaddq_s32 (lbase, vdupq_n_s32(SkFractionalIntToFixed(dx8)));
michael@0:             hbase = vaddq_s32 (hbase, vdupq_n_s32(SkFractionalIntToFixed(dx8)));
michael@0:             dst16 += 8;
michael@0:             count -= 8;
michael@0:             fx += dx8;
michael@0:         };
michael@0:         xy = (uint32_t *) dst16;
michael@0:     }
michael@0: 
michael@0:     uint16_t* xx = (uint16_t*)xy;
michael@0:     for (int i = count; i > 0; --i) {
michael@0:         *xx++ = TILEX_PROCF(SkFractionalIntToFixed(fx), maxX);
michael@0:         fx += dx;
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void AFFINE_NOFILTER_NAME(const SkBitmapProcState& s,
michael@0:                                  uint32_t xy[], int count, int x, int y) {
michael@0:     SkASSERT(s.fInvType & SkMatrix::kAffine_Mask);
michael@0:     SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
michael@0:                              SkMatrix::kScale_Mask |
michael@0:                              SkMatrix::kAffine_Mask)) == 0);
michael@0: 
michael@0:     PREAMBLE(s);
michael@0:     SkPoint srcPt;
michael@0:     s.fInvProc(s.fInvMatrix,
michael@0:                SkIntToScalar(x) + SK_ScalarHalf,
michael@0:                SkIntToScalar(y) + SK_ScalarHalf, &srcPt);
michael@0: 
michael@0:     SkFractionalInt fx = SkScalarToFractionalInt(srcPt.fX);
michael@0:     SkFractionalInt fy = SkScalarToFractionalInt(srcPt.fY);
michael@0:     SkFractionalInt dx = s.fInvSxFractionalInt;
michael@0:     SkFractionalInt dy = s.fInvKyFractionalInt;
michael@0:     int maxX = s.fBitmap->width() - 1;
michael@0:     int maxY = s.fBitmap->height() - 1;
michael@0: 
michael@0:     if (count >= 8) {
michael@0:         SkFractionalInt dx4 = dx * 4;
michael@0:         SkFractionalInt dy4 = dy * 4;
michael@0:         SkFractionalInt dx8 = dx * 8;
michael@0:         SkFractionalInt dy8 = dy * 8;
michael@0: 
michael@0:         int32x4_t xbase, ybase;
michael@0:         int32x4_t x2base, y2base;
michael@0:         int16_t *dst16 = (int16_t *) xy;
michael@0: 
michael@0:         // now build fx, fx+dx, fx+2dx, fx+3dx
michael@0:         xbase = vdupq_n_s32(SkFractionalIntToFixed(fx));
michael@0:         xbase = vsetq_lane_s32(SkFractionalIntToFixed(fx+dx), xbase, 1);
michael@0:         xbase = vsetq_lane_s32(SkFractionalIntToFixed(fx+dx+dx), xbase, 2);
michael@0:         xbase = vsetq_lane_s32(SkFractionalIntToFixed(fx+dx+dx+dx), xbase, 3);
michael@0: 
michael@0:         // same for fy
michael@0:         ybase = vdupq_n_s32(SkFractionalIntToFixed(fy));
michael@0:         ybase = vsetq_lane_s32(SkFractionalIntToFixed(fy+dy), ybase, 1);
michael@0:         ybase = vsetq_lane_s32(SkFractionalIntToFixed(fy+dy+dy), ybase, 2);
michael@0:         ybase = vsetq_lane_s32(SkFractionalIntToFixed(fy+dy+dy+dy), ybase, 3);
michael@0: 
michael@0:         x2base = vaddq_s32(xbase, vdupq_n_s32(SkFractionalIntToFixed(dx4)));
michael@0:         y2base = vaddq_s32(ybase, vdupq_n_s32(SkFractionalIntToFixed(dy4)));
michael@0: 
michael@0:         // store & bump
michael@0:         do {
michael@0:             int16x8x2_t hi16;
michael@0: 
michael@0:             hi16.val[0] = TILEX_PROCF_NEON8(xbase, x2base, maxX);
michael@0:             hi16.val[1] = TILEY_PROCF_NEON8(ybase, y2base, maxY);
michael@0: 
michael@0:             vst2q_s16(dst16, hi16);
michael@0: 
michael@0:             // moving base and on to the next
michael@0:             xbase = vaddq_s32(xbase, vdupq_n_s32(SkFractionalIntToFixed(dx8)));
michael@0:             ybase = vaddq_s32(ybase, vdupq_n_s32(SkFractionalIntToFixed(dy8)));
michael@0:             x2base = vaddq_s32(x2base, vdupq_n_s32(SkFractionalIntToFixed(dx8)));
michael@0:             y2base = vaddq_s32(y2base, vdupq_n_s32(SkFractionalIntToFixed(dy8)));
michael@0: 
michael@0:             dst16 += 16; // 8x32 aka 16x16
michael@0:             count -= 8;
michael@0:             fx += dx8;
michael@0:             fy += dy8;
michael@0:         } while (count >= 8);
michael@0:         xy = (uint32_t *) dst16;
michael@0:     }
michael@0: 
michael@0:     for (int i = count; i > 0; --i) {
michael@0:         *xy++ = (TILEY_PROCF(SkFractionalIntToFixed(fy), maxY) << 16) |
michael@0:                  TILEX_PROCF(SkFractionalIntToFixed(fx), maxX);
michael@0:         fx += dx; fy += dy;
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void PERSP_NOFILTER_NAME(const SkBitmapProcState& s,
michael@0:                                 uint32_t* SK_RESTRICT xy,
michael@0:                                 int count, int x, int y) {
michael@0:     SkASSERT(s.fInvType & SkMatrix::kPerspective_Mask);
michael@0: 
michael@0:     PREAMBLE(s);
michael@0:     // max{X,Y} are int here, but later shown/assumed to fit in 16 bits
michael@0:     int maxX = s.fBitmap->width() - 1;
michael@0:     int maxY = s.fBitmap->height() - 1;
michael@0: 
michael@0:     SkPerspIter iter(s.fInvMatrix,
michael@0:                      SkIntToScalar(x) + SK_ScalarHalf,
michael@0:                      SkIntToScalar(y) + SK_ScalarHalf, count);
michael@0: 
michael@0:     while ((count = iter.next()) != 0) {
michael@0:         const SkFixed* SK_RESTRICT srcXY = iter.getXY();
michael@0: 
michael@0:         if (count >= 8) {
michael@0:             int32_t *mysrc = (int32_t *) srcXY;
michael@0:             int16_t *mydst = (int16_t *) xy;
michael@0:             do {
michael@0:                 int16x8x2_t hi16;
michael@0:                 int32x4x2_t xy1, xy2;
michael@0: 
michael@0:                 xy1 = vld2q_s32(mysrc);
michael@0:                 xy2 = vld2q_s32(mysrc+8);
michael@0: 
michael@0:                 hi16.val[0] = TILEX_PROCF_NEON8(xy1.val[0], xy2.val[0], maxX);
michael@0:                 hi16.val[1] = TILEY_PROCF_NEON8(xy1.val[1], xy2.val[1], maxY);
michael@0: 
michael@0:                 vst2q_s16(mydst, hi16);
michael@0: 
michael@0:                 count -= 8;  // 8 iterations
michael@0:                 mysrc += 16; // 16 longs
michael@0:                 mydst += 16; // 16 shorts, aka 8 longs
michael@0:             } while (count >= 8);
michael@0:             // get xy and srcXY fixed up
michael@0:             srcXY = (const SkFixed *) mysrc;
michael@0:             xy = (uint32_t *) mydst;
michael@0:         }
michael@0: 
michael@0:         while (--count >= 0) {
michael@0:             *xy++ = (TILEY_PROCF(srcXY[1], maxY) << 16) |
michael@0:                      TILEX_PROCF(srcXY[0], maxX);
michael@0:             srcXY += 2;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: static inline uint32_t PACK_FILTER_Y_NAME(SkFixed f, unsigned max,
michael@0:                                           SkFixed one PREAMBLE_PARAM_Y) {
michael@0:     unsigned i = TILEY_PROCF(f, max);
michael@0:     i = (i << 4) | TILEY_LOW_BITS(f, max);
michael@0:     return (i << 14) | (TILEY_PROCF((f + one), max));
michael@0: }
michael@0: 
michael@0: static inline uint32_t PACK_FILTER_X_NAME(SkFixed f, unsigned max,
michael@0:                                           SkFixed one PREAMBLE_PARAM_X) {
michael@0:     unsigned i = TILEX_PROCF(f, max);
michael@0:     i = (i << 4) | TILEX_LOW_BITS(f, max);
michael@0:     return (i << 14) | (TILEX_PROCF((f + one), max));
michael@0: }
michael@0: 
michael@0: static inline int32x4_t PACK_FILTER_X4_NAME(int32x4_t f, unsigned max,
michael@0:                                           SkFixed one PREAMBLE_PARAM_X) {
michael@0:     int32x4_t ret, res, wide_one;
michael@0: 
michael@0:     // Prepare constants
michael@0:     wide_one = vdupq_n_s32(one);
michael@0: 
michael@0:     // Step 1
michael@0:     res = TILEX_PROCF_NEON4(f, max);
michael@0: 
michael@0:     // Step 2
michael@0:     ret = TILEX_LOW_BITS_NEON4(f, max);
michael@0:     ret = vsliq_n_s32(ret, res, 4);
michael@0: 
michael@0:     // Step 3
michael@0:     res = TILEX_PROCF_NEON4(f + wide_one, max);
michael@0:     ret = vorrq_s32(vshlq_n_s32(ret, 14), res);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static inline int32x4_t PACK_FILTER_Y4_NAME(int32x4_t f, unsigned max,
michael@0:                                           SkFixed one PREAMBLE_PARAM_X) {
michael@0:     int32x4_t ret, res, wide_one;
michael@0: 
michael@0:     // Prepare constants
michael@0:     wide_one = vdupq_n_s32(one);
michael@0: 
michael@0:     // Step 1
michael@0:     res = TILEY_PROCF_NEON4(f, max);
michael@0: 
michael@0:     // Step 2
michael@0:     ret = TILEY_LOW_BITS_NEON4(f, max);
michael@0:     ret = vsliq_n_s32(ret, res, 4);
michael@0: 
michael@0:     // Step 3
michael@0:     res = TILEY_PROCF_NEON4(f + wide_one, max);
michael@0:     ret = vorrq_s32(vshlq_n_s32(ret, 14), res);
michael@0: 
michael@0:     return ret;
michael@0: }
michael@0: 
michael@0: static void SCALE_FILTER_NAME(const SkBitmapProcState& s,
michael@0:                               uint32_t xy[], int count, int x, int y) {
michael@0:     SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
michael@0:                              SkMatrix::kScale_Mask)) == 0);
michael@0:     SkASSERT(s.fInvKy == 0);
michael@0: 
michael@0:     PREAMBLE(s);
michael@0: 
michael@0:     const unsigned maxX = s.fBitmap->width() - 1;
michael@0:     const SkFixed one = s.fFilterOneX;
michael@0:     const SkFractionalInt dx = s.fInvSxFractionalInt;
michael@0:     SkFractionalInt fx;
michael@0: 
michael@0:     {
michael@0:         SkPoint pt;
michael@0:         s.fInvProc(s.fInvMatrix, SkIntToScalar(x) + SK_ScalarHalf,
michael@0:                                  SkIntToScalar(y) + SK_ScalarHalf, &pt);
michael@0:         const SkFixed fy = SkScalarToFixed(pt.fY) - (s.fFilterOneY >> 1);
michael@0:         const unsigned maxY = s.fBitmap->height() - 1;
michael@0:         // compute our two Y values up front
michael@0:         *xy++ = PACK_FILTER_Y_NAME(fy, maxY, s.fFilterOneY PREAMBLE_ARG_Y);
michael@0:         // now initialize fx
michael@0:         fx = SkScalarToFractionalInt(pt.fX) - (SkFixedToFractionalInt(one) >> 1);
michael@0:     }
michael@0: 
michael@0: #ifdef CHECK_FOR_DECAL
michael@0:     // test if we don't need to apply the tile proc
michael@0:     if (can_truncate_to_fixed_for_decal(fx, dx, count, maxX)) {
michael@0:         decal_filter_scale_neon(xy, SkFractionalIntToFixed(fx),
michael@0:                              SkFractionalIntToFixed(dx), count);
michael@0:         return;
michael@0:     }
michael@0: #endif
michael@0:     {
michael@0: 
michael@0:     if (count >= 4) {
michael@0:         int32x4_t wide_fx;
michael@0: 
michael@0:         wide_fx = vdupq_n_s32(SkFractionalIntToFixed(fx));
michael@0:         wide_fx = vsetq_lane_s32(SkFractionalIntToFixed(fx+dx), wide_fx, 1);
michael@0:         wide_fx = vsetq_lane_s32(SkFractionalIntToFixed(fx+dx+dx), wide_fx, 2);
michael@0:         wide_fx = vsetq_lane_s32(SkFractionalIntToFixed(fx+dx+dx+dx), wide_fx, 3);
michael@0: 
michael@0:         while (count >= 4) {
michael@0:             int32x4_t res;
michael@0: 
michael@0:             res = PACK_FILTER_X4_NAME(wide_fx, maxX, one PREAMBLE_ARG_X);
michael@0: 
michael@0:             vst1q_u32(xy, vreinterpretq_u32_s32(res));
michael@0: 
michael@0:             wide_fx += vdupq_n_s32(SkFractionalIntToFixed(dx+dx+dx+dx));
michael@0:             fx += dx+dx+dx+dx;
michael@0:             xy += 4;
michael@0:             count -= 4;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     while (--count >= 0) {
michael@0:         *xy++ = PACK_FILTER_X_NAME(SkFractionalIntToFixed(fx), maxX, one PREAMBLE_ARG_X);
michael@0:         fx += dx;
michael@0:     }
michael@0: 
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void AFFINE_FILTER_NAME(const SkBitmapProcState& s,
michael@0:                                uint32_t xy[], int count, int x, int y) {
michael@0:     SkASSERT(s.fInvType & SkMatrix::kAffine_Mask);
michael@0:     SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
michael@0:                              SkMatrix::kScale_Mask |
michael@0:                              SkMatrix::kAffine_Mask)) == 0);
michael@0: 
michael@0:     PREAMBLE(s);
michael@0:     SkPoint srcPt;
michael@0:     s.fInvProc(s.fInvMatrix,
michael@0:                SkIntToScalar(x) + SK_ScalarHalf,
michael@0:                SkIntToScalar(y) + SK_ScalarHalf, &srcPt);
michael@0: 
michael@0:     SkFixed oneX = s.fFilterOneX;
michael@0:     SkFixed oneY = s.fFilterOneY;
michael@0:     SkFixed fx = SkScalarToFixed(srcPt.fX) - (oneX >> 1);
michael@0:     SkFixed fy = SkScalarToFixed(srcPt.fY) - (oneY >> 1);
michael@0:     SkFixed dx = s.fInvSx;
michael@0:     SkFixed dy = s.fInvKy;
michael@0:     unsigned maxX = s.fBitmap->width() - 1;
michael@0:     unsigned maxY = s.fBitmap->height() - 1;
michael@0: 
michael@0:     if (count >= 4) {
michael@0:         int32x4_t wide_fy, wide_fx;
michael@0: 
michael@0:         wide_fx = vdupq_n_s32(fx);
michael@0:         wide_fx = vsetq_lane_s32(fx+dx, wide_fx, 1);
michael@0:         wide_fx = vsetq_lane_s32(fx+dx+dx, wide_fx, 2);
michael@0:         wide_fx = vsetq_lane_s32(fx+dx+dx+dx, wide_fx, 3);
michael@0: 
michael@0:         wide_fy = vdupq_n_s32(fy);
michael@0:         wide_fy = vsetq_lane_s32(fy+dy, wide_fy, 1);
michael@0:         wide_fy = vsetq_lane_s32(fy+dy+dy, wide_fy, 2);
michael@0:         wide_fy = vsetq_lane_s32(fy+dy+dy+dy, wide_fy, 3);
michael@0: 
michael@0:         while (count >= 4) {
michael@0:             int32x4x2_t vxy;
michael@0: 
michael@0:             // do the X side, then the Y side, then interleave them
michael@0:             vxy.val[0] = PACK_FILTER_Y4_NAME(wide_fy, maxY, oneY PREAMBLE_ARG_Y);
michael@0:             vxy.val[1] = PACK_FILTER_X4_NAME(wide_fx, maxX, oneX PREAMBLE_ARG_X);
michael@0: 
michael@0:             // interleave as YXYXYXYX as part of the storing
michael@0:             vst2q_s32((int32_t*)xy, vxy);
michael@0: 
michael@0:             // prepare next iteration
michael@0:             wide_fx += vdupq_n_s32(dx+dx+dx+dx);
michael@0:             fx += dx + dx + dx + dx;
michael@0:             wide_fy += vdupq_n_s32(dy+dy+dy+dy);
michael@0:             fy += dy+dy+dy+dy;
michael@0:             xy += 8; // 4 x's, 4 y's
michael@0:             count -= 4;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     while (--count >= 0) {
michael@0:         // NB: writing Y/X
michael@0:         *xy++ = PACK_FILTER_Y_NAME(fy, maxY, oneY PREAMBLE_ARG_Y);
michael@0:         fy += dy;
michael@0:         *xy++ = PACK_FILTER_X_NAME(fx, maxX, oneX PREAMBLE_ARG_X);
michael@0:         fx += dx;
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void PERSP_FILTER_NAME(const SkBitmapProcState& s,
michael@0:                               uint32_t* SK_RESTRICT xy, int count,
michael@0:                               int x, int y) {
michael@0:     SkASSERT(s.fInvType & SkMatrix::kPerspective_Mask);
michael@0: 
michael@0:     PREAMBLE(s);
michael@0:     unsigned maxX = s.fBitmap->width() - 1;
michael@0:     unsigned maxY = s.fBitmap->height() - 1;
michael@0:     SkFixed oneX = s.fFilterOneX;
michael@0:     SkFixed oneY = s.fFilterOneY;
michael@0: 
michael@0:     SkPerspIter iter(s.fInvMatrix,
michael@0:                      SkIntToScalar(x) + SK_ScalarHalf,
michael@0:                      SkIntToScalar(y) + SK_ScalarHalf, count);
michael@0: 
michael@0:     while ((count = iter.next()) != 0) {
michael@0:         const SkFixed* SK_RESTRICT srcXY = iter.getXY();
michael@0: 
michael@0:         while (count >= 4) {
michael@0:             int32x4_t wide_x, wide_y;
michael@0:             int32x4x2_t vxy, vresyx;
michael@0: 
michael@0:             // load src:  x-y-x-y-x-y-x-y
michael@0:             vxy = vld2q_s32(srcXY);
michael@0: 
michael@0:             // do the X side, then the Y side, then interleave them
michael@0:             wide_x = vsubq_s32(vxy.val[0], vdupq_n_s32(oneX>>1));
michael@0:             wide_y = vsubq_s32(vxy.val[1], vdupq_n_s32(oneY>>1));
michael@0: 
michael@0:             vresyx.val[0] = PACK_FILTER_Y4_NAME(wide_y, maxY, oneY PREAMBLE_ARG_Y);
michael@0:             vresyx.val[1] = PACK_FILTER_X4_NAME(wide_x, maxX, oneX PREAMBLE_ARG_X);
michael@0: 
michael@0:             // store interleaved as y-x-y-x-y-x-y-x (NB != read order)
michael@0:             vst2q_s32((int32_t*)xy, vresyx);
michael@0: 
michael@0:             // on to the next iteration
michael@0:             srcXY += 2*4;
michael@0:             count -= 4;
michael@0:             xy += 2*4;
michael@0:         }
michael@0: 
michael@0:         while (--count >= 0) {
michael@0:             // NB: we read x/y, we write y/x
michael@0:             *xy++ = PACK_FILTER_Y_NAME(srcXY[1] - (oneY >> 1), maxY,
michael@0:                                        oneY PREAMBLE_ARG_Y);
michael@0:             *xy++ = PACK_FILTER_X_NAME(srcXY[0] - (oneX >> 1), maxX,
michael@0:                                        oneX PREAMBLE_ARG_X);
michael@0:             srcXY += 2;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: const SkBitmapProcState::MatrixProc MAKENAME(_Procs)[] = {
michael@0:     SCALE_NOFILTER_NAME,
michael@0:     SCALE_FILTER_NAME,
michael@0:     AFFINE_NOFILTER_NAME,
michael@0:     AFFINE_FILTER_NAME,
michael@0:     PERSP_NOFILTER_NAME,
michael@0:     PERSP_FILTER_NAME
michael@0: };
michael@0: 
michael@0: #undef TILEX_PROCF_NEON8
michael@0: #undef TILEY_PROCF_NEON8
michael@0: #undef TILEX_PROCF_NEON4
michael@0: #undef TILEY_PROCF_NEON4
michael@0: #undef TILEX_LOW_BITS_NEON4
michael@0: #undef TILEY_LOW_BITS_NEON4
michael@0: 
michael@0: #undef MAKENAME
michael@0: #undef TILEX_PROCF
michael@0: #undef TILEY_PROCF
michael@0: #ifdef CHECK_FOR_DECAL
michael@0:     #undef CHECK_FOR_DECAL
michael@0: #endif
michael@0: 
michael@0: #undef SCALE_NOFILTER_NAME
michael@0: #undef SCALE_FILTER_NAME
michael@0: #undef AFFINE_NOFILTER_NAME
michael@0: #undef AFFINE_FILTER_NAME
michael@0: #undef PERSP_NOFILTER_NAME
michael@0: #undef PERSP_FILTER_NAME
michael@0: 
michael@0: #undef PREAMBLE
michael@0: #undef PREAMBLE_PARAM_X
michael@0: #undef PREAMBLE_PARAM_Y
michael@0: #undef PREAMBLE_ARG_X
michael@0: #undef PREAMBLE_ARG_Y
michael@0: 
michael@0: #undef TILEX_LOW_BITS
michael@0: #undef TILEY_LOW_BITS