ARM: use prefetch in nearest scaled 'src_0565_0565'

Benchmark on ARM Cortex-A8 r1p3 @500MHz, 32-bit LPDDR @166MHz:
 Microbenchmark (scaling 2000x2000 image with scale factor close to 1x):
  before: op=1, src=10020565, dst=10020565, speed=75.02 MPix/s
  after:  op=1, src=10020565, dst=10020565, speed=73.63 MPix/s

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=176.12 MPix/s
  after:  op=1, src=10020565, dst=10020565, speed=267.50 MPix/s

diff --git a/pixman/pixman-arm-simd-asm.S b/pixman/pixman-arm-simd-asm.S
index 7567700..dd1366d 100644 (file)
--- a/pixman/pixman-arm-simd-asm.S
+++ b/pixman/pixman-arm-simd-asm.S
@@ -348,6 +348,7 @@ pixman_asm_function pixman_scaled_nearest_scanline_0565_0565_SRC_asm_armv6
        TMP1    .req    r4
        TMP2    .req    r5
        VXMASK  .req    r6
+       PF_OFFS .req    r7
 
        ldr     UNIT_X, [sp]
        push    {r4, r5, r6, r7}
@@ -366,12 +367,33 @@ pixman_asm_function pixman_scaled_nearest_scanline_0565_0565_SRC_asm_armv6
                strh    TMP2, [DST], #2
        .endm
 
+       /*
+        * stop prefetch before reaching the end of scanline (a good behaving
+        * value selected based on some benchmarks with short scanlines)
+        */
+       #define PREFETCH_BRAKING_DISTANCE 32
+
        /* now do the scaling */
        and     TMP1, VXMASK, VX, lsr #15
        add     VX, VX, UNIT_X
-       subs    W, #4
+       subs    W, #(8 + PREFETCH_BRAKING_DISTANCE)
+       blt     2f
+       /* set prefetch distance to 80 pixels ahead */
+       add     PF_OFFS, VX, UNIT_X, lsl #6
+       add     PF_OFFS, PF_OFFS, UNIT_X, lsl #4
+1:     /* main loop, process 8 pixels per iteration with prefetch */
+       subs    W, W, #8
+       add     PF_OFFS, UNIT_X, lsl #3
+       scale_2_pixels
+       scale_2_pixels
+       scale_2_pixels
+       scale_2_pixels
+       pld     [SRC, PF_OFFS, lsr #15]
+       bge     1b
+2:
+       subs    W, #(4 - 8 - PREFETCH_BRAKING_DISTANCE)
        blt     2f
-1: /* main loop, process 4 pixels per iteration */
+1:     /* process the remaining pixels */
        scale_2_pixels
        scale_2_pixels
        subs    W, W, #4
@@ -394,6 +416,7 @@ pixman_asm_function pixman_scaled_nearest_scanline_0565_0565_SRC_asm_armv6
        .unreq  TMP1
        .unreq  TMP2
        .unreq  VXMASK
+       .unreq  PF_OFFS
        /* return */
        pop     {r4, r5, r6, r7}
        bx      lr