ARM: assembly optimized nearest scaled 'src_8888_8888'
authorSiarhei Siamashka <siarhei.siamashka@nokia.com>
Sun, 6 Mar 2011 20:16:32 +0000 (22:16 +0200)
committerSiarhei Siamashka <siarhei.siamashka@nokia.com>
Sat, 12 Mar 2011 19:26:05 +0000 (21:26 +0200)
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=20028888, dst=20028888, speed=44.36 MPix/s
  after:  op=1, src=20028888, dst=20028888, speed=39.79 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=20028888, dst=20028888, speed=102.36 MPix/s
  after:  op=1, src=20028888, dst=20028888, speed=163.12 MPix/s

pixman/pixman-arm-simd-asm.S
pixman/pixman-arm-simd.c

index a9775e2..858c690 100644 (file)
@@ -433,3 +433,6 @@ pixman_asm_function fname
 
 generate_nearest_scanline_func \
     pixman_scaled_nearest_scanline_0565_0565_SRC_asm_armv6, 1, h, 80, 32
+
+generate_nearest_scanline_func \
+    pixman_scaled_nearest_scanline_8888_8888_SRC_asm_armv6, 2,  , 48, 32
index 6bbc109..a66f8df 100644 (file)
@@ -389,6 +389,8 @@ PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, armv6, over_n_8_8888,
 
 PIXMAN_ARM_BIND_SCALED_NEAREST_SRC_DST (armv6, 0565_0565, SRC,
                                         uint16_t, uint16_t)
+PIXMAN_ARM_BIND_SCALED_NEAREST_SRC_DST (armv6, 8888_8888, SRC,
+                                        uint32_t, uint32_t)
 
 static const pixman_fast_path_t arm_simd_fast_paths[] =
 {
@@ -411,6 +413,13 @@ static const pixman_fast_path_t arm_simd_fast_paths[] =
     PIXMAN_ARM_SIMPLE_NEAREST_FAST_PATH (SRC, r5g6b5, r5g6b5, armv6_0565_0565),
     PIXMAN_ARM_SIMPLE_NEAREST_FAST_PATH (SRC, b5g6r5, b5g6r5, armv6_0565_0565),
 
+    PIXMAN_ARM_SIMPLE_NEAREST_FAST_PATH (SRC, a8r8g8b8, a8r8g8b8, armv6_8888_8888),
+    PIXMAN_ARM_SIMPLE_NEAREST_FAST_PATH (SRC, a8r8g8b8, x8r8g8b8, armv6_8888_8888),
+    PIXMAN_ARM_SIMPLE_NEAREST_FAST_PATH (SRC, x8r8g8b8, x8r8g8b8, armv6_8888_8888),
+    PIXMAN_ARM_SIMPLE_NEAREST_FAST_PATH (SRC, a8b8g8r8, a8b8g8r8, armv6_8888_8888),
+    PIXMAN_ARM_SIMPLE_NEAREST_FAST_PATH (SRC, a8b8g8r8, x8b8g8r8, armv6_8888_8888),
+    PIXMAN_ARM_SIMPLE_NEAREST_FAST_PATH (SRC, x8b8g8r8, x8b8g8r8, armv6_8888_8888),
+
     { PIXMAN_OP_NONE },
 };