ARM: a bit faster NEON bilinear scaling for r5g6b5 source images

Instructions scheduling improved in the code responsible for fetching r5g6b5
pixels and converting them to the intermediate x8r8g8b8 color format used in
the interpolation part of code. Still a lot of NEON stalls are remaining,
which can be resolved later by the use of pipelining.

Benchmark on ARM Cortex-A8 r2p2 @1GHz, 32-bit LPDDR @200MHz:
 Microbenchmark (scaling 2000x2000 image with scale factor close to 1x):
  before: op=1, src=10020565, dst=10020565, speed=32.29 MPix/s
          op=1, src=10020565, dst=20020888, speed=36.82 MPix/s
  after:  op=1, src=10020565, dst=10020565, speed=41.35 MPix/s
          op=1, src=10020565, dst=20020888, speed=49.16 MPix/s

1 files changed, 100 insertions, 18 deletions

diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 2b6875b7..71b30ac5 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -2430,6 +2430,101 @@ fname:
     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]!
@@ -2477,12 +2572,8 @@ fname:
 .endm
 
 .macro bilinear_interpolate_two_pixels src_fmt, dst_fmt
-    bilinear_load_&src_fmt d0, d1, d2
-    vmull.u8  q1, d0, d28
-    vmlal.u8  q1, d1, d29
-    bilinear_load_&src_fmt d20, d21, d22
-    vmull.u8  q11, d20, d28
-    vmlal.u8  q11, d21, d29
+    bilinear_load_and_vertical_interpolate_two_&src_fmt \
+                q1, q11, d0, d1, d20, d21, d22, d23
     vshr.u16  q15, q12, #8
     vadd.u16  q12, q12, q13
     vshll.u16 q0, d2, #8
@@ -2498,18 +2589,9 @@ fname:
 .endm
 
 .macro bilinear_interpolate_four_pixels src_fmt, dst_fmt
-    bilinear_load_&src_fmt d0, d1, d2
-    vmull.u8  q1, d0, d28
-    vmlal.u8  q1, d1, d29
-    bilinear_load_&src_fmt d20, d21, d22
-    vmull.u8  q11, d20, d28
-    vmlal.u8  q11, d21, d29
-    bilinear_load_&src_fmt d4, d5, d6
-    vmull.u8  q3, d4, d28
-    vmlal.u8  q3, d5, d29
-    bilinear_load_&src_fmt d16, d17, d18
-    vmull.u8  q9, d16, d28
-    vmlal.u8  q9, d17, d29
+    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