ARMv6: Add fast path for over_reverse_n_8888

Benchmark results, "before" is upstream commit
c343846 lowlevel-blt-bench: add in_reverse_8888_8888 test
and "after" is with this patch only added on top.

lowlevel-blt-bench, over_reverse_n_8888, 100 iterations:

       Before          After
      Mean StdDev     Mean StdDev   Confidence   Change
L1    15.1    0.1    274.5    2.3    100.00%   +1718.9%
L2    12.8    0.3    181.8    0.7    100.00%   +1315.5%
M     10.8    0.0     77.9    0.0    100.00%    +621.2%
HT     9.7    0.0     29.4    0.2    100.00%    +204.9%
VT     9.5    0.0     26.7    0.1    100.00%    +179.3%
R      9.3    0.0     25.3    0.1    100.00%    +173.6%
RT     6.0    0.1     11.0    0.2    100.00%     +82.9%

At most 16 outliers rejected per case per set.

cairo-perf-trace with trimmed traces, 30 iterations:

                                    Before          After
                                   Mean StdDev     Mean StdDev   Confidence   Change
t-poppler.trace                    12.9    0.1      9.7    0.0    100.00%     +32.6%
t-firefox-talos-gfx.trace          33.2    0.7     32.9    0.4     95.23%      +0.9%  (insignificant)
t-firefox-particles.trace          27.4    0.1     27.3    0.2     99.65%      +0.4%
t-firefox-canvas-alpha.trace       20.5    0.3     20.5    0.3     57.51%      +0.3%  (insignificant)
t-poppler-reseau.trace             22.4    0.1     22.4    0.1     95.69%      +0.3%  (insignificant)
t-firefox-fishtank.trace           13.2    0.0     13.2    0.0     99.84%      +0.1%
t-swfdec-giant-steps.trace         14.9    0.0     14.9    0.0     87.68%      +0.1%  (insignificant)
t-swfdec-youtube.trace              7.8    0.0      7.8    0.0     35.22%      +0.1%  (insignificant)
t-firefox-planet-gnome.trace       11.5    0.0     11.5    0.0     29.37%      +0.0%  (insignificant)
t-firefox-fishbowl.trace           21.2    0.0     21.2    0.0     18.09%      +0.0%  (insignificant)
t-grads-heat-map.trace              4.4    0.0      4.4    0.0      1.84%      +0.0%  (insignificant)
t-firefox-paintball.trace          18.0    0.0     18.0    0.0     33.43%      -0.0%  (insignificant)
t-firefox-talos-svg.trace          20.5    0.0     20.5    0.1     68.56%      -0.1%  (insignificant)
t-midori-zoomed.trace               8.0    0.0      8.0    0.0     99.98%      -0.1%
t-firefox-canvas-swscroll.trace    32.1    0.1     32.1    0.1     85.27%      -0.1%  (insignificant)
t-gnome-system-monitor.trace       17.2    0.0     17.2    0.0     99.97%      -0.2%
t-firefox-chalkboard.trace         36.5    0.0     36.6    0.0    100.00%      -0.2%
t-firefox-asteroids.trace          11.1    0.0     11.1    0.0    100.00%      -0.2%
t-firefox-canvas.trace             17.9    0.0     18.0    0.0    100.00%      -0.3%
t-chromium-tabs.trace               4.9    0.0      4.9    0.0     97.95%      -0.3%  (insignificant)
t-xfce4-terminal-a1.trace           4.8    0.0      4.8    0.0    100.00%      -0.4%
t-firefox-scrolling.trace          31.1    0.1     31.2    0.1    100.00%      -0.5%
t-evolution.trace                  13.7    0.1     13.8    0.1     99.99%      -0.6%
t-gnome-terminal-vim.trace         22.0    0.2     22.2    0.1     99.99%      -0.7%
t-gvim.trace                       33.2    0.2     33.5    0.2    100.00%      -0.8%

At most 6 outliers rejected per case per set.

Cairo perf reports the running time, but the change is computed for
operations per second instead (inverse of running time).

Changes in the order of +/- 1% can be accounted for measurement errors,
even if they are deemed to be statistically significant. This claim is
based on comparing two 30-iteration identical "before" runs using the
exact same binaries, and observing changes from -0.4% to +0.5% with
>=99% confidence.

Confidence is based on Welch's t-test.

v4, Pekka Paalanen <pekka.paalanen@collabora.co.uk> :
	Rebased, re-benchmarked on Raspberry Pi, commit message.

2 files changed, 84 insertions, 0 deletions

diff --git a/pixman/pixman-arm-simd-asm.S b/pixman/pixman-arm-simd-asm.S
index c209688..dd6f788 100644
--- a/pixman/pixman-arm-simd-asm.S
+++ b/pixman/pixman-arm-simd-asm.S
@@ -611,3 +611,81 @@ generate_composite_function \
 
 /******************************************************************************/
 
+.macro over_reverse_n_8888_init
+        ldr     SRC, [sp, #ARGS_STACK_OFFSET]
+        ldr     MASK, =0x00800080
+        /* Split source pixel into RB/AG parts */
+        uxtb16  STRIDE_S, SRC
+        uxtb16  STRIDE_M, SRC, ror #8
+        /* Set GE[3:0] to 0101 so SEL instructions do what we want */
+        uadd8   SCRATCH, MASK, MASK
+        line_saved_regs  STRIDE_D, ORIG_W
+.endm
+
+.macro over_reverse_n_8888_newline
+        mov     STRIDE_D, #0xFF
+.endm
+
+.macro over_reverse_n_8888_process_head  cond, numbytes, firstreg, unaligned_src, unaligned_mask, preload
+        pixld   , numbytes, firstreg, DST, 0
+.endm
+
+.macro over_reverse_n_8888_1pixel  d, is_only
+        teq     WK&d, #0
+        beq     8f       /* replace with source */
+        bics    ORIG_W, STRIDE_D, WK&d, lsr #24
+ .if is_only == 1
+        beq     49f      /* skip store */
+ .else
+        beq     9f       /* write same value back */
+ .endif
+        mla     SCRATCH, STRIDE_S, ORIG_W, MASK /* red/blue */
+        mla     ORIG_W, STRIDE_M, ORIG_W, MASK  /* alpha/green */
+        uxtab16 SCRATCH, SCRATCH, SCRATCH, ror #8
+        uxtab16 ORIG_W, ORIG_W, ORIG_W, ror #8
+        mov     SCRATCH, SCRATCH, ror #8
+        sel     ORIG_W, SCRATCH, ORIG_W
+        uqadd8  WK&d, WK&d, ORIG_W
+        b       9f
+8:      mov     WK&d, SRC
+9:
+.endm
+
+.macro over_reverse_n_8888_tail  numbytes, reg1, reg2, reg3, reg4
+ .if numbytes == 4
+        over_reverse_n_8888_1pixel  reg1, 1
+ .else
+        and     SCRATCH, WK&reg1, WK&reg2
+  .if numbytes == 16
+        and     SCRATCH, SCRATCH, WK&reg3
+        and     SCRATCH, SCRATCH, WK&reg4
+  .endif
+        mvns    SCRATCH, SCRATCH, asr #24
+        beq     49f /* skip store if all opaque */
+        over_reverse_n_8888_1pixel  reg1, 0
+        over_reverse_n_8888_1pixel  reg2, 0
+  .if numbytes == 16
+        over_reverse_n_8888_1pixel  reg3, 0
+        over_reverse_n_8888_1pixel  reg4, 0
+  .endif
+ .endif
+        pixst   , numbytes, reg1, DST
+49:
+.endm
+
+.macro over_reverse_n_8888_process_tail  cond, numbytes, firstreg
+        over_reverse_n_8888_tail  numbytes, firstreg, %(firstreg+1), %(firstreg+2), %(firstreg+3)
+.endm
+
+generate_composite_function \
+    pixman_composite_over_reverse_n_8888_asm_armv6, 0, 0, 32 \
+    FLAG_DST_READWRITE | FLAG_BRANCH_OVER | FLAG_PROCESS_CORRUPTS_PSR | FLAG_PROCESS_DOES_STORE | FLAG_SPILL_LINE_VARS | FLAG_PROCESS_CORRUPTS_SCRATCH, \
+    3, /* prefetch distance */ \
+    over_reverse_n_8888_init, \
+    over_reverse_n_8888_newline, \
+    nop_macro, /* cleanup */ \
+    over_reverse_n_8888_process_head, \
+    over_reverse_n_8888_process_tail
+
+/******************************************************************************/
+
diff --git a/pixman/pixman-arm-simd.c b/pixman/pixman-arm-simd.c
index af062e1..8fbc439 100644
--- a/pixman/pixman-arm-simd.c
+++ b/pixman/pixman-arm-simd.c
@@ -47,6 +47,9 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (armv6, add_8_8,
 PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (armv6, over_8888_8888,
                                    uint32_t, 1, uint32_t, 1)
 
+PIXMAN_ARM_BIND_FAST_PATH_N_DST (0, armv6, over_reverse_n_8888,
+                                 uint32_t, 1)
+
 PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (SKIP_ZERO_MASK, armv6, over_8888_n_8888,
                                      uint32_t, 1, uint32_t, 1)
 
@@ -225,6 +228,9 @@ static const pixman_fast_path_t arm_simd_fast_paths[] =
     PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, solid, a8b8g8r8, armv6_composite_over_8888_n_8888),
     PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, solid, x8b8g8r8, armv6_composite_over_8888_n_8888),
 
+    PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8r8g8b8, armv6_composite_over_reverse_n_8888),
+    PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8b8g8r8, armv6_composite_over_reverse_n_8888),
+
     PIXMAN_STD_FAST_PATH (ADD, a8, null, a8, armv6_composite_add_8_8),
 
     PIXMAN_STD_FAST_PATH (OVER, solid, a8, a8r8g8b8, armv6_composite_over_n_8_8888),