ARM: NEON scanline functions for bilinear scaling

General fetch->combine->store based bilinear scanline functions.
Need further optimizations and eventually will be replaced with optimal
functions one by one.
General functions should be located in pixman-arm-neon-asm-bilinear.S and
optimal functions in pixman-arm-neon-asm.S

Following general bilinear scanline functions are implemented
    over_8888_8888
    add_8888_8888
    src_8888_8_8888
    src_8888_8_0565
    src_0565_8_x888
    src_0565_8_0565
    over_8888_8_8888
    add_8888_8_8888

2 files changed, 769 insertions, 0 deletions

diff --git a/pixman/Makefile.am b/pixman/Makefile.am
index d016e9f2..be082668 100644
--- a/pixman/Makefile.am
+++ b/pixman/Makefile.am
@@ -115,6 +115,7 @@ libpixman_arm_neon_la_SOURCES = \
         pixman-arm-neon.c	\
         pixman-arm-common.h	\
         pixman-arm-neon-asm.S	\
+		pixman-arm-neon-asm-bilinear.S \
         pixman-arm-neon-asm.h
 libpixman_arm_neon_la_CFLAGS = $(DEP_CFLAGS)
 libpixman_arm_neon_la_LIBADD = $(DEP_LIBS)
diff --git a/pixman/pixman-arm-neon-asm-bilinear.S b/pixman/pixman-arm-neon-asm-bilinear.S
new file mode 100644
index 00000000..9a4a1ffb
--- /dev/null
+++ b/pixman/pixman-arm-neon-asm-bilinear.S
@@ -0,0 +1,768 @@
+/*
+ * Copyright © 2011 SCore Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Author:  Siarhei Siamashka (siarhei.siamashka@nokia.com)
+ * Author:  Taekyun Kim (tkq.kim@samsung.com)
+ */
+
+/*
+ * This file contains scaled bilinear scanline functions implemented
+ * using older siarhei's bilinear macro template.
+ *
+ * << General scanline function procedures >>
+ *  1. bilinear interpolate source pixels
+ *  2. load mask pixels
+ *  3. load destination pixels
+ *  4. duplicate mask to fill whole register
+ *  5. interleave source & destination pixels
+ *  6. apply mask to source pixels
+ *  7. combine source & destination pixels
+ *  8, Deinterleave final result
+ *  9. store destination pixels
+ *
+ * All registers with single number (i.e. src0, tmp0) are 64-bits registers.
+ * Registers with double numbers(src01, dst01) are 128-bits registers.
+ * All temp registers can be used freely outside the code block.
+ * Assume that symbol(register .req) OUT and MASK are defined at caller of these macro blocks.
+ *
+ * TODOs
+ *  Support 0565 pixel format
+ *  Optimization for two and last pixel cases
+ *
+ * Remarks
+ *  There can be lots of pipeline stalls inside code block and between code blocks.
+ *  Further optimizations will be done by new macro templates using head/tail_head/tail scheme.
+ */
+
+/* Prevent the stack from becoming executable for no reason... */
+#if defined(__linux__) && defined (__ELF__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+.text
+.fpu neon
+.arch armv7a
+.object_arch armv4
+.eabi_attribute 10, 0
+.eabi_attribute 12, 0
+.arm
+.altmacro
+
+#include "pixman-arm-neon-asm.h"
+
+/*
+ * Bilinear macros from pixman-arm-neon-asm.S
+ */
+
+/* Supplementary macro for setting function attributes */
+.macro pixman_asm_function fname
+    .func fname
+    .global fname
+#ifdef __ELF__
+    .hidden fname
+    .type fname, %function
+#endif
+fname:
+.endm
+
+/*
+ * Bilinear scaling support code which tries to provide pixel fetching, color
+ * format conversion, and interpolation as separate macros which can be used
+ * as the basic building blocks for constructing bilinear scanline functions.
+ */
+
+.macro bilinear_load_8888 reg1, reg2, tmp
+    mov       TMP2, X, asr #16
+    add       X, X, UX
+    add       TMP1, TOP, TMP2, asl #2
+    add       TMP2, BOTTOM, TMP2, asl #2
+    vld1.32   {reg1}, [TMP1]
+    vld1.32   {reg2}, [TMP2]
+.endm
+
+.macro bilinear_load_0565 reg1, reg2, tmp
+    mov       TMP2, X, asr #16
+    add       X, X, UX
+    add       TMP1, TOP, TMP2, asl #1
+    add       TMP2, BOTTOM, TMP2, asl #1
+    vld1.32   {reg2[0]}, [TMP1]
+    vld1.32   {reg2[1]}, [TMP2]
+    convert_four_0565_to_x888_packed reg2, reg1, reg2, tmp
+.endm
+
+.macro bilinear_load_and_vertical_interpolate_two_8888 \
+                    acc1, acc2, reg1, reg2, reg3, reg4, tmp1, tmp2
+
+    bilinear_load_8888 reg1, reg2, tmp1
+    vmull.u8  acc1, reg1, d28
+    vmlal.u8  acc1, reg2, d29
+    bilinear_load_8888 reg3, reg4, tmp2
+    vmull.u8  acc2, reg3, d28
+    vmlal.u8  acc2, reg4, d29
+.endm
+
+.macro bilinear_load_and_vertical_interpolate_four_8888 \
+                xacc1, xacc2, xreg1, xreg2, xreg3, xreg4, xacc2lo, xacc2hi \
+                yacc1, yacc2, yreg1, yreg2, yreg3, yreg4, yacc2lo, yacc2hi
+
+    bilinear_load_and_vertical_interpolate_two_8888 \
+                xacc1, xacc2, xreg1, xreg2, xreg3, xreg4, xacc2lo, xacc2hi
+    bilinear_load_and_vertical_interpolate_two_8888 \
+                yacc1, yacc2, yreg1, yreg2, yreg3, yreg4, yacc2lo, yacc2hi
+.endm
+
+.macro bilinear_load_and_vertical_interpolate_two_0565 \
+                acc1, acc2, reg1, reg2, reg3, reg4, acc2lo, acc2hi
+
+    mov       TMP2, X, asr #16
+    add       X, X, UX
+    mov       TMP4, X, asr #16
+    add       X, X, UX
+    add       TMP1, TOP, TMP2, asl #1
+    add       TMP2, BOTTOM, TMP2, asl #1
+    add       TMP3, TOP, TMP4, asl #1
+    add       TMP4, BOTTOM, TMP4, asl #1
+    vld1.32   {acc2lo[0]}, [TMP1]
+    vld1.32   {acc2hi[0]}, [TMP3]
+    vld1.32   {acc2lo[1]}, [TMP2]
+    vld1.32   {acc2hi[1]}, [TMP4]
+    convert_0565_to_x888 acc2, reg3, reg2, reg1
+    vzip.u8   reg1, reg3
+    vzip.u8   reg2, reg4
+    vzip.u8   reg3, reg4
+    vzip.u8   reg1, reg2
+    vmull.u8  acc1, reg1, d28
+    vmlal.u8  acc1, reg2, d29
+    vmull.u8  acc2, reg3, d28
+    vmlal.u8  acc2, reg4, d29
+.endm
+
+.macro bilinear_load_and_vertical_interpolate_four_0565 \
+                xacc1, xacc2, xreg1, xreg2, xreg3, xreg4, xacc2lo, xacc2hi \
+                yacc1, yacc2, yreg1, yreg2, yreg3, yreg4, yacc2lo, yacc2hi
+
+    mov       TMP2, X, asr #16
+    add       X, X, UX
+    mov       TMP4, X, asr #16
+    add       X, X, UX
+    add       TMP1, TOP, TMP2, asl #1
+    add       TMP2, BOTTOM, TMP2, asl #1
+    add       TMP3, TOP, TMP4, asl #1
+    add       TMP4, BOTTOM, TMP4, asl #1
+    vld1.32   {xacc2lo[0]}, [TMP1]
+    vld1.32   {xacc2hi[0]}, [TMP3]
+    vld1.32   {xacc2lo[1]}, [TMP2]
+    vld1.32   {xacc2hi[1]}, [TMP4]
+    convert_0565_to_x888 xacc2, xreg3, xreg2, xreg1
+    mov       TMP2, X, asr #16
+    add       X, X, UX
+    mov       TMP4, X, asr #16
+    add       X, X, UX
+    add       TMP1, TOP, TMP2, asl #1
+    add       TMP2, BOTTOM, TMP2, asl #1
+    add       TMP3, TOP, TMP4, asl #1
+    add       TMP4, BOTTOM, TMP4, asl #1
+    vld1.32   {yacc2lo[0]}, [TMP1]
+    vzip.u8   xreg1, xreg3
+    vld1.32   {yacc2hi[0]}, [TMP3]
+    vzip.u8   xreg2, xreg4
+    vld1.32   {yacc2lo[1]}, [TMP2]
+    vzip.u8   xreg3, xreg4
+    vld1.32   {yacc2hi[1]}, [TMP4]
+    vzip.u8   xreg1, xreg2
+    convert_0565_to_x888 yacc2, yreg3, yreg2, yreg1
+    vmull.u8  xacc1, xreg1, d28
+    vzip.u8   yreg1, yreg3
+    vmlal.u8  xacc1, xreg2, d29
+    vzip.u8   yreg2, yreg4
+    vmull.u8  xacc2, xreg3, d28
+    vzip.u8   yreg3, yreg4
+    vmlal.u8  xacc2, xreg4, d29
+    vzip.u8   yreg1, yreg2
+    vmull.u8  yacc1, yreg1, d28
+    vmlal.u8  yacc1, yreg2, d29
+    vmull.u8  yacc2, yreg3, d28
+    vmlal.u8  yacc2, yreg4, d29
+.endm
+
+.macro bilinear_store_8888 numpix, tmp1, tmp2
+.if numpix == 4
+    vst1.32   {d0, d1}, [OUT]!
+.elseif numpix == 2
+    vst1.32   {d0}, [OUT]!
+.elseif numpix == 1
+    vst1.32   {d0[0]}, [OUT, :32]!
+.else
+    .error bilinear_store_8888 numpix is unsupported
+.endif
+.endm
+
+.macro bilinear_store_0565 numpix, tmp1, tmp2
+    vuzp.u8 d0, d1
+    vuzp.u8 d2, d3
+    vuzp.u8 d1, d3
+    vuzp.u8 d0, d2
+    convert_8888_to_0565 d2, d1, d0, q1, tmp1, tmp2
+.if numpix == 4
+    vst1.16   {d2}, [OUT]!
+.elseif numpix == 2
+    vst1.32   {d2[0]}, [OUT]!
+.elseif numpix == 1
+    vst1.16   {d2[0]}, [OUT]!
+.else
+    .error bilinear_store_0565 numpix is unsupported
+.endif
+.endm
+
+
+/*
+ * Macros for loading mask pixels into register 'mask'.
+ * vdup must be done in somewhere else.
+ */
+.macro bilinear_load_mask_x numpix, mask
+.endm
+
+.macro bilinear_load_mask_8 numpix, mask
+.if numpix == 4
+    vld1.32     {mask[0]}, [MASK]!
+.elseif numpix == 2
+    vld1.16     {mask[0]}, [MASK]!
+.elseif numpix == 1
+    vld1.8      {mask[0]}, [MASK]!
+.else
+    .error bilinear_load_mask_8 numpix is unsupported
+.endif
+.endm
+
+.macro bilinear_load_mask mask_fmt, numpix, mask
+    bilinear_load_mask_&mask_fmt numpix, mask
+.endm
+
+
+/*
+ * Macros for loading destination pixels into register 'dst0' and 'dst1'.
+ * Interleave should be done somewhere else.
+ */
+.macro bilinear_load_dst_0565_src numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_load_dst_8888_src numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_load_dst_8888 numpix, dst0, dst1, dst01
+.if numpix == 4
+    vld1.32     {dst0, dst1}, [OUT]
+.elseif numpix == 2
+    vld1.32     {dst0}, [OUT]
+.elseif numpix == 1
+    vld1.32     {dst0[0]}, [OUT]
+.else
+    .error bilinear_load_dst_8888 numpix is unsupported
+.endif
+.endm
+
+.macro bilinear_load_dst_8888_over numpix, dst0, dst1, dst01
+    bilinear_load_dst_8888 numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_load_dst_8888_add numpix, dst0, dst1, dst01
+    bilinear_load_dst_8888 numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_load_dst dst_fmt, op, numpix, dst0, dst1, dst01
+    bilinear_load_dst_&dst_fmt&_&op numpix, dst0, dst1, dst01
+.endm
+
+/*
+ * Macros for duplicating partially loaded mask to fill entire register.
+ * We will apply mask to interleaved source pixels, that is
+ *  (r0, r1, r2, r3, g0, g1, g2, g3) x (m0, m1, m2, m3, m0, m1, m2, m3)
+ *  (b0, b1, b2, b3, a0, a1, a2, a3) x (m0, m1, m2, m3, m0, m1, m2, m3)
+ * So, we need to duplicate loaded mask into whole register.
+ *
+ * For two pixel case
+ *  (r0, r1, x, x, g0, g1, x, x) x (m0, m1, m0, m1, m0, m1, m0, m1)
+ *  (b0, b1, x, x, a0, a1, x, x) x (m0, m1, m0, m1, m0, m1, m0, m1)
+ * We can do some optimizations for this including one pixel cases.
+ */
+.macro bilinear_duplicate_mask_x numpix, mask
+.endm
+
+.macro bilinear_duplicate_mask_8 numpix, mask
+.if numpix == 4
+    vdup.32     mask, mask[0]
+.elseif numpix == 2
+    vdup.16     mask, mask[0]
+.elseif numpix == 1
+    vdup.8      mask, mask[0]
+.else
+    .error bilinear_duplicate_mask_8 is unsupported
+.endif
+.endm
+
+.macro bilinear_duplicate_mask mask_fmt, numpix, mask
+    bilinear_duplicate_mask_&mask_fmt numpix, mask
+.endm
+
+/*
+ * Macros for interleaving src and dst pixels to rrrr gggg bbbb aaaa form.
+ * Interleave should be done when maks is enabled or operator is 'over'.
+ */
+.macro bilinear_interleave src0, src1, dst0, dst1
+    vuzp.8      src0, src1
+    vuzp.8      dst0, dst1
+    vuzp.8      src0, src1
+    vuzp.8      dst0, dst1
+.endm
+
+.macro bilinear_interleave_src_dst_x_src \
+                numpix, src0, src1, src01, dst0, dst1, dst01
+.endm
+
+.macro bilinear_interleave_src_dst_x_over \
+                numpix, src0, src1, src01, dst0, dst1, dst01
+
+    bilinear_interleave src0, src1, dst0, dst1
+.endm
+
+.macro bilinear_interleave_src_dst_x_add \
+                numpix, src0, src1, src01, dst0, dst1, dst01
+.endm
+
+.macro bilinear_interleave_src_dst_8_src \
+                numpix, src0, src1, src01, dst0, dst1, dst01
+
+    bilinear_interleave src0, src1, dst0, dst1
+.endm
+
+.macro bilinear_interleave_src_dst_8_over \
+                numpix, src0, src1, src01, dst0, dst1, dst01
+
+    bilinear_interleave src0, src1, dst0, dst1
+.endm
+
+.macro bilinear_interleave_src_dst_8_add \
+                numpix, src0, src1, src01, dst0, dst1, dst01
+
+    bilinear_interleave src0, src1, dst0, dst1
+.endm
+
+.macro bilinear_interleave_src_dst \
+                mask_fmt, op, numpix, src0, src1, src01, dst0, dst1, dst01
+
+    bilinear_interleave_src_dst_&mask_fmt&_&op \
+                numpix, src0, src1, src01, dst0, dst1, dst01
+.endm
+
+
+/*
+ * Macros for applying masks to src pixels. (see combine_mask_u() function)
+ * src, dst should be in interleaved form.
+ * mask register should be in form (m0, m1, m2, m3).
+ */
+.macro bilinear_apply_mask_to_src_x \
+                numpix, src0, src1, src01, mask, \
+                tmp01, tmp23, tmp45, tmp67
+.endm
+
+.macro bilinear_apply_mask_to_src_8 \
+                numpix, src0, src1, src01, mask, \
+                tmp01, tmp23, tmp45, tmp67
+
+    vmull.u8        tmp01, src0, mask
+    vmull.u8        tmp23, src1, mask
+    /* bubbles */
+    vrshr.u16       tmp45, tmp01, #8
+    vrshr.u16       tmp67, tmp23, #8
+    /* bubbles */
+    vraddhn.u16     src0, tmp45, tmp01
+    vraddhn.u16     src1, tmp67, tmp23
+.endm
+
+.macro bilinear_apply_mask_to_src \
+                mask_fmt, numpix, src0, src1, src01, mask, \
+                tmp01, tmp23, tmp45, tmp67
+
+    bilinear_apply_mask_to_src_&mask_fmt \
+                numpix, src0, src1, src01, mask, \
+                tmp01, tmp23, tmp45, tmp67
+.endm
+
+
+/*
+ * Macros for combining src and destination pixels.
+ * Interleave or not is depending on operator 'op'.
+ */
+.macro bilinear_combine_src \
+                numpix, src0, src1, src01, dst0, dst1, dst01, \
+                tmp01, tmp23, tmp45, tmp67, tmp8
+.endm
+
+.macro bilinear_combine_over \
+                numpix, src0, src1, src01, dst0, dst1, dst01, \
+                tmp01, tmp23, tmp45, tmp67, tmp8
+
+    vdup.32     tmp8, src1[1]
+    /* bubbles */
+    vmvn.8      tmp8, tmp8
+    /* bubbles */
+    vmull.u8    tmp01, dst0, tmp8
+    /* bubbles */
+    vmull.u8    tmp23, dst1, tmp8
+    /* bubbles */
+    vrshr.u16   tmp45, tmp01, #8
+    vrshr.u16   tmp67, tmp23, #8
+    /* bubbles */
+    vraddhn.u16 dst0, tmp45, tmp01
+    vraddhn.u16 dst1, tmp67, tmp23
+    /* bubbles */
+    vqadd.u8    src01, dst01, src01
+.endm
+
+.macro bilinear_combine_add \
+                numpix, src0, src1, src01, dst0, dst1, dst01, \
+                tmp01, tmp23, tmp45, tmp67, tmp8
+
+    vqadd.u8    src01, dst01, src01
+.endm
+
+.macro bilinear_combine \
+                op, numpix, src0, src1, src01, dst0, dst1, dst01, \
+                tmp01, tmp23, tmp45, tmp67, tmp8
+
+    bilinear_combine_&op \
+                numpix, src0, src1, src01, dst0, dst1, dst01, \
+                tmp01, tmp23, tmp45, tmp67, tmp8
+.endm
+
+/*
+ * Macros for final deinterleaving of destination pixels if needed.
+ */
+.macro bilinear_deinterleave numpix, dst0, dst1, dst01
+    vuzp.8      dst0, dst1
+    /* bubbles */
+    vuzp.8      dst0, dst1
+.endm
+
+.macro bilinear_deinterleave_dst_x_src numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_deinterleave_dst_x_over numpix, dst0, dst1, dst01
+    bilinear_deinterleave numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_deinterleave_dst_x_add numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_deinterleave_dst_8_src numpix, dst0, dst1, dst01
+    bilinear_deinterleave numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_deinterleave_dst_8_over numpix, dst0, dst1, dst01
+    bilinear_deinterleave numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_deinterleave_dst_8_add numpix, dst0, dst1, dst01
+    bilinear_deinterleave numpix, dst0, dst1, dst01
+.endm
+
+.macro bilinear_deinterleave_dst mask_fmt, op, numpix, dst0, dst1, dst01
+    bilinear_deinterleave_dst_&mask_fmt&_&op numpix, dst0, dst1, dst01
+.endm
+
+
+.macro bilinear_interpolate_last_pixel src_fmt, mask_fmt, dst_fmt, op
+    bilinear_load_&src_fmt d0, d1, d2
+    bilinear_load_mask mask_fmt, 1, d4
+    bilinear_load_dst dst_fmt, op, 1, d18, d19, q9
+    vmull.u8  q1, d0, d28
+    vmlal.u8  q1, d1, d29
+    vshr.u16  d30, d24, #8
+    /* 4 cycles bubble */
+    vshll.u16 q0, d2, #8
+    vmlsl.u16 q0, d2, d30
+    vmlal.u16 q0, d3, d30
+    /* 5 cycles bubble */
+    bilinear_duplicate_mask mask_fmt, 1, d4
+    vshrn.u32 d0, q0, #16
+    /* 3 cycles bubble */
+    vmovn.u16 d0, q0
+    /* 1 cycle bubble */
+    bilinear_interleave_src_dst \
+                mask_fmt, op, 1, d0, d1, q0, d18, d19, q9
+    bilinear_apply_mask_to_src \
+                mask_fmt, 1, d0, d1, q0, d4, \
+                q3, q8, q10, q11
+    bilinear_combine \
+                op, 1, d0, d1, q0, d18, d19, q9, \
+                q3, q8, q10, q11, d5
+    bilinear_deinterleave_dst mask_fmt, op, 1, d0, d1, q0
+    bilinear_store_&dst_fmt 1, q2, q3
+.endm
+
+.macro bilinear_interpolate_two_pixels src_fmt, mask_fmt, dst_fmt, op
+    bilinear_load_and_vertical_interpolate_two_&src_fmt \
+                q1, q11, d0, d1, d20, d21, d22, d23
+    bilinear_load_mask mask_fmt, 2, d4
+    bilinear_load_dst dst_fmt, op, 2, d18, d19, q9
+    vshr.u16  q15, q12, #8
+    vadd.u16  q12, q12, q13
+    vshll.u16 q0, d2, #8
+    vmlsl.u16 q0, d2, d30
+    vmlal.u16 q0, d3, d30
+    vshll.u16 q10, d22, #8
+    vmlsl.u16 q10, d22, d31
+    vmlal.u16 q10, d23, d31
+    vshrn.u32 d30, q0, #16
+    vshrn.u32 d31, q10, #16
+    bilinear_duplicate_mask mask_fmt, 2, d4
+    vmovn.u16 d0, q15
+    bilinear_interleave_src_dst \
+                mask_fmt, op, 2, d0, d1, q0, d18, d19, q9
+    bilinear_apply_mask_to_src \
+                mask_fmt, 2, d0, d1, q0, d4, \
+                q3, q8, q10, q11
+    bilinear_combine \
+                op, 2, d0, d1, q0, d18, d19, q9, \
+                q3, q8, q10, q11, d5
+    bilinear_deinterleave_dst mask_fmt, op, 2, d0, d1, q0
+    bilinear_store_&dst_fmt 2, q2, q3
+.endm
+
+.macro bilinear_interpolate_four_pixels src_fmt, mask_fmt, dst_fmt, op
+    bilinear_load_and_vertical_interpolate_four_&src_fmt \
+                q1, q11, d0, d1, d20, d21, d22, d23 \
+                q3, q9,  d4, d5, d16, d17, d18, d19
+    pld       [TMP1, PF_OFFS]
+    vshr.u16  q15, q12, #8
+    vadd.u16  q12, q12, q13
+    vshll.u16 q0, d2, #8
+    vmlsl.u16 q0, d2, d30
+    vmlal.u16 q0, d3, d30
+    vshll.u16 q10, d22, #8
+    vmlsl.u16 q10, d22, d31
+    vmlal.u16 q10, d23, d31
+    vshr.u16  q15, q12, #8
+    vshll.u16 q2, d6, #8
+    vmlsl.u16 q2, d6, d30
+    vmlal.u16 q2, d7, d30
+    vshll.u16 q8, d18, #8
+    bilinear_load_mask mask_fmt, 4, d30
+    bilinear_load_dst dst_fmt, op, 4, d2, d3, q1
+    pld       [TMP2, PF_OFFS]
+    vmlsl.u16 q8, d18, d31
+    vmlal.u16 q8, d19, d31
+    vadd.u16  q12, q12, q13
+    vshrn.u32 d0, q0, #16
+    vshrn.u32 d1, q10, #16
+    vshrn.u32 d4, q2, #16
+    vshrn.u32 d5, q8, #16
+    bilinear_duplicate_mask mask_fmt, 4, d30
+    vmovn.u16 d0, q0
+    vmovn.u16 d1, q2
+    bilinear_interleave_src_dst \
+                mask_fmt, op, 4, d0, d1, q0, d2, d3, q1
+    bilinear_apply_mask_to_src \
+                mask_fmt, 4, d0, d1, q0, d30, \
+                q3, q8, q9, q10
+    bilinear_combine \
+                op, 4, d0, d1, q0, d2, d3, q1, \
+                q3, q8, q9, q10, d22
+    bilinear_deinterleave_dst mask_fmt, op, 4, d0, d1, q0
+    bilinear_store_&dst_fmt 4, q2, q3
+.endm
+
+.macro generate_bilinear_scanline_func_src_dst \
+                fname, src_fmt, dst_fmt, op, \
+                bpp_shift, prefetch_distance
+
+pixman_asm_function fname
+    OUT       .req      r0
+    TOP       .req      r1
+    BOTTOM    .req      r2
+    WT        .req      r3
+    WB        .req      r4
+    X         .req      r5
+    UX        .req      r6
+    WIDTH     .req      ip
+    TMP1      .req      r3
+    TMP2      .req      r4
+    PF_OFFS   .req      r7
+    TMP3      .req      r8
+    TMP4      .req      r9
+
+    mov       ip, sp
+    push      {r4, r5, r6, r7, r8, r9}
+    mov       PF_OFFS, #prefetch_distance
+    ldmia     ip, {WB, X, UX, WIDTH}
+    mul       PF_OFFS, PF_OFFS, UX
+
+    cmp       WIDTH, #0
+    ble       3f
+
+    vdup.u16  q12, X
+    vdup.u16  q13, UX
+    vdup.u8   d28, WT
+    vdup.u8   d29, WB
+    vadd.u16  d25, d25, d26
+    vadd.u16  q13, q13, q13
+
+    subs      WIDTH, WIDTH, #4
+    blt       1f
+    mov       PF_OFFS, PF_OFFS, asr #(16 - bpp_shift)
+0:
+    bilinear_interpolate_four_pixels src_fmt, x, dst_fmt, op
+    subs      WIDTH, WIDTH, #4
+    bge       0b
+1:
+    tst       WIDTH, #2
+    beq       2f
+    bilinear_interpolate_two_pixels src_fmt, x, dst_fmt, op
+2:
+    tst       WIDTH, #1
+    beq       3f
+    bilinear_interpolate_last_pixel src_fmt, x, dst_fmt, op
+3:
+    pop       {r4, r5, r6, r7, r8, r9}
+    bx        lr
+
+    .unreq    OUT
+    .unreq    TOP
+    .unreq    BOTTOM
+    .unreq    WT
+    .unreq    WB
+    .unreq    X
+    .unreq    UX
+    .unreq    WIDTH
+    .unreq    TMP1
+    .unreq    TMP2
+    .unreq    PF_OFFS
+    .unreq    TMP3
+    .unreq    TMP4
+.endfunc
+
+.endm
+
+.macro generate_bilinear_scanline_func_src_a8_dst \
+                fname, src_fmt, dst_fmt, op, \
+                bpp_shift, prefetch_distance
+
+pixman_asm_function fname
+    OUT       .req      r0
+    MASK      .req      r1
+    TOP       .req      r2
+    BOTTOM    .req      r3
+    WT        .req      r4
+    WB        .req      r5
+    X         .req      r6
+    UX        .req      r7
+    WIDTH     .req      ip
+    TMP1      .req      r4
+    TMP2      .req      r5
+    PF_OFFS   .req      r8
+    TMP3      .req      r9
+    TMP4      .req      r10
+
+    mov       ip, sp
+    push      {r4, r5, r6, r7, r8, r9, r10, ip}
+    mov       PF_OFFS, #prefetch_distance
+    ldmia     ip, {WT, WB, X, UX, WIDTH}
+    mul       PF_OFFS, PF_OFFS, UX
+
+    cmp       WIDTH, #0
+    ble       3f
+
+    vdup.u16  q12, X
+    vdup.u16  q13, UX
+    vdup.u8   d28, WT
+    vdup.u8   d29, WB
+    vadd.u16  d25, d25, d26
+    vadd.u16  q13, q13, q13
+
+    subs      WIDTH, WIDTH, #4
+    blt       1f
+    mov       PF_OFFS, PF_OFFS, asr #(16 - bpp_shift)
+0:
+    bilinear_interpolate_four_pixels src_fmt, 8, dst_fmt, op
+    subs      WIDTH, WIDTH, #4
+    bge       0b
+1:
+    tst       WIDTH, #2
+    beq       2f
+    bilinear_interpolate_two_pixels src_fmt, 8, dst_fmt, op
+2:
+    tst       WIDTH, #1
+    beq       3f
+    bilinear_interpolate_last_pixel src_fmt, 8, dst_fmt, op
+3:
+    pop       {r4, r5, r6, r7, r8, r9, r10, ip}
+    bx        lr
+
+    .unreq    OUT
+    .unreq    TOP
+    .unreq    BOTTOM
+    .unreq    WT
+    .unreq    WB
+    .unreq    X
+    .unreq    UX
+    .unreq    WIDTH
+    .unreq    MASK
+    .unreq    TMP1
+    .unreq    TMP2
+    .unreq    PF_OFFS
+    .unreq    TMP3
+    .unreq    TMP4
+.endfunc
+
+.endm
+
+generate_bilinear_scanline_func_src_dst \
+    pixman_scaled_bilinear_scanline_8888_8888_OVER_asm_neon, \
+    8888, 8888, over, 2, 28
+
+generate_bilinear_scanline_func_src_dst \
+    pixman_scaled_bilinear_scanline_8888_8888_ADD_asm_neon, \
+    8888, 8888, add, 2, 28
+
+generate_bilinear_scanline_func_src_a8_dst \
+    pixman_scaled_bilinear_scanline_8888_8_8888_SRC_asm_neon, \
+    8888, 8888, src, 2, 28
+
+generate_bilinear_scanline_func_src_a8_dst \
+    pixman_scaled_bilinear_scanline_8888_8_0565_SRC_asm_neon, \
+    8888, 0565, src, 2, 28
+
+generate_bilinear_scanline_func_src_a8_dst \
+    pixman_scaled_bilinear_scanline_0565_8_x888_SRC_asm_neon, \
+    0565, 8888, src, 1, 28
+
+generate_bilinear_scanline_func_src_a8_dst \
+    pixman_scaled_bilinear_scanline_0565_8_0565_SRC_asm_neon, \
+    0565, 0565, src, 1, 28
+
+generate_bilinear_scanline_func_src_a8_dst \
+    pixman_scaled_bilinear_scanline_8888_8_8888_OVER_asm_neon, \
+    8888, 8888, over, 2, 28
+
+generate_bilinear_scanline_func_src_a8_dst \
+    pixman_scaled_bilinear_scanline_8888_8_8888_ADD_asm_neon, \
+    8888, 8888, add, 2, 28