ARM assembly optimizations for 'encode_one_block'

Almost 2x faster than original C variant.

1 files changed, 362 insertions, 0 deletions

diff --git a/jchuff.c b/jchuff.c
index b05c8e7..5fbc53d 100644
--- a/jchuff.c
+++ b/jchuff.c
@@ -2,6 +2,11 @@
  * jchuff.c
  *
  * Copyright (C) 1991-1997, Thomas G. Lane.
+ *
+ * ARM optimizations
+ * Copyright (C) 2010 Nokia Corporation and/or its subsidiary(-ies). All rights reserved.
+ * Contact: Alexander Bokovoy <alexander.bokovoy@nokia.com>
+ *
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -505,6 +510,349 @@ flush_bits (working_state * state)
   return TRUE;
 }
 
+/*********************************/
+
+#ifdef __arm__
+
+/*
+#!/usr/bin/env ruby
+
+require 'generator'
+
+$jpeg_natural_order = Generator.new(
+[
+      1,  8, 16,  9,  2,  3, 10,
+ 17, 24, 32, 25, 18, 11,  4,  5,
+ 12, 19, 26, 33, 40, 48, 41, 34,
+ 27, 20, 13,  6,  7, 14, 21, 28,
+ 35, 42, 49, 56, 57, 50, 43, 36,
+ 29, 22, 15, 23, 30, 37, 44, 51,
+ 58, 59, 52, 45, 38, 31, 39, 46,
+ 53, 60, 61, 54, 47, 55, 62, 63
+])
+
+$ldr_reg = Generator.new {|g|
+    while true do
+	["%[tmp1]", "%[tmp2]", "%[tmp3]"].each {|x| g.yield x}
+    end
+}
+
+$cmp_reg = Generator.new {|g|
+    while true do
+	["%[tmp1]", "%[tmp2]", "%[tmp3]"].each {|x| g.yield x}
+    end
+}
+
+$idx = Generator.new(1.upto(1000))
+
+def load
+    printf("\t\t\"ldrh  %s, [%%[blk], #%d]\\n\"\n",
+	$ldr_reg.next, $jpeg_natural_order.next * 2)
+end
+
+def store
+    r = $cmp_reg.next
+    i = $idx.next
+    printf("\t\t\"cmp   %s, #0\\n\"\n", r)
+    printf("\t\t\"strh  %s, [%%[out], #%d]\\n\"\n", r, i * 2)
+    printf("\t\t\"movne %%[n], #%d\\n\"\n", i)
+end
+
+load
+load
+while $jpeg_natural_order.next? do
+    load
+    store
+end
+store
+store
+*/
+
+/*
+ * Find last nonzero coefficient and produce output in natural order,
+ * instructions are scheduled to make use of ARM Cortex-A8 dual-issue
+ * capability
+ */
+LOCAL(int)
+find_last_nonzero_index (JCOEFPTR block, JCOEFPTR out)
+{
+  int tmp1, tmp2, tmp3, n = 0;
+  asm volatile (
+    "ldrh  %[tmp1], [%[blk], #2]\n"
+    "ldrh  %[tmp2], [%[blk], #16]\n"
+    "ldrh  %[tmp3], [%[blk], #32]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #2]\n"
+    "movne %[n], #1\n"
+    "ldrh  %[tmp1], [%[blk], #18]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #4]\n"
+    "movne %[n], #2\n"
+    "ldrh  %[tmp2], [%[blk], #4]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #6]\n"
+    "movne %[n], #3\n"
+    "ldrh  %[tmp3], [%[blk], #6]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #8]\n"
+    "movne %[n], #4\n"
+    "ldrh  %[tmp1], [%[blk], #20]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #10]\n"
+    "movne %[n], #5\n"
+    "ldrh  %[tmp2], [%[blk], #34]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #12]\n"
+    "movne %[n], #6\n"
+    "ldrh  %[tmp3], [%[blk], #48]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #14]\n"
+    "movne %[n], #7\n"
+    "ldrh  %[tmp1], [%[blk], #64]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #16]\n"
+    "movne %[n], #8\n"
+    "ldrh  %[tmp2], [%[blk], #50]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #18]\n"
+    "movne %[n], #9\n"
+    "ldrh  %[tmp3], [%[blk], #36]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #20]\n"
+    "movne %[n], #10\n"
+    "ldrh  %[tmp1], [%[blk], #22]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #22]\n"
+    "movne %[n], #11\n"
+    "ldrh  %[tmp2], [%[blk], #8]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #24]\n"
+    "movne %[n], #12\n"
+    "ldrh  %[tmp3], [%[blk], #10]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #26]\n"
+    "movne %[n], #13\n"
+    "ldrh  %[tmp1], [%[blk], #24]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #28]\n"
+    "movne %[n], #14\n"
+    "ldrh  %[tmp2], [%[blk], #38]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #30]\n"
+    "movne %[n], #15\n"
+    "ldrh  %[tmp3], [%[blk], #52]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #32]\n"
+    "movne %[n], #16\n"
+    "ldrh  %[tmp1], [%[blk], #66]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #34]\n"
+    "movne %[n], #17\n"
+    "ldrh  %[tmp2], [%[blk], #80]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #36]\n"
+    "movne %[n], #18\n"
+    "ldrh  %[tmp3], [%[blk], #96]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #38]\n"
+    "movne %[n], #19\n"
+    "ldrh  %[tmp1], [%[blk], #82]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #40]\n"
+    "movne %[n], #20\n"
+    "ldrh  %[tmp2], [%[blk], #68]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #42]\n"
+    "movne %[n], #21\n"
+    "ldrh  %[tmp3], [%[blk], #54]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #44]\n"
+    "movne %[n], #22\n"
+    "ldrh  %[tmp1], [%[blk], #40]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #46]\n"
+    "movne %[n], #23\n"
+    "ldrh  %[tmp2], [%[blk], #26]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #48]\n"
+    "movne %[n], #24\n"
+    "ldrh  %[tmp3], [%[blk], #12]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #50]\n"
+    "movne %[n], #25\n"
+    "ldrh  %[tmp1], [%[blk], #14]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #52]\n"
+    "movne %[n], #26\n"
+    "ldrh  %[tmp2], [%[blk], #28]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #54]\n"
+    "movne %[n], #27\n"
+    "ldrh  %[tmp3], [%[blk], #42]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #56]\n"
+    "movne %[n], #28\n"
+    "ldrh  %[tmp1], [%[blk], #56]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #58]\n"
+    "movne %[n], #29\n"
+    "ldrh  %[tmp2], [%[blk], #70]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #60]\n"
+    "movne %[n], #30\n"
+    "ldrh  %[tmp3], [%[blk], #84]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #62]\n"
+    "movne %[n], #31\n"
+    "ldrh  %[tmp1], [%[blk], #98]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #64]\n"
+    "movne %[n], #32\n"
+    "ldrh  %[tmp2], [%[blk], #112]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #66]\n"
+    "movne %[n], #33\n"
+    "ldrh  %[tmp3], [%[blk], #114]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #68]\n"
+    "movne %[n], #34\n"
+    "ldrh  %[tmp1], [%[blk], #100]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #70]\n"
+    "movne %[n], #35\n"
+    "ldrh  %[tmp2], [%[blk], #86]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #72]\n"
+    "movne %[n], #36\n"
+    "ldrh  %[tmp3], [%[blk], #72]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #74]\n"
+    "movne %[n], #37\n"
+    "ldrh  %[tmp1], [%[blk], #58]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #76]\n"
+    "movne %[n], #38\n"
+    "ldrh  %[tmp2], [%[blk], #44]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #78]\n"
+    "movne %[n], #39\n"
+    "ldrh  %[tmp3], [%[blk], #30]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #80]\n"
+    "movne %[n], #40\n"
+    "ldrh  %[tmp1], [%[blk], #46]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #82]\n"
+    "movne %[n], #41\n"
+    "ldrh  %[tmp2], [%[blk], #60]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #84]\n"
+    "movne %[n], #42\n"
+    "ldrh  %[tmp3], [%[blk], #74]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #86]\n"
+    "movne %[n], #43\n"
+    "ldrh  %[tmp1], [%[blk], #88]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #88]\n"
+    "movne %[n], #44\n"
+    "ldrh  %[tmp2], [%[blk], #102]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #90]\n"
+    "movne %[n], #45\n"
+    "ldrh  %[tmp3], [%[blk], #116]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #92]\n"
+    "movne %[n], #46\n"
+    "ldrh  %[tmp1], [%[blk], #118]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #94]\n"
+    "movne %[n], #47\n"
+    "ldrh  %[tmp2], [%[blk], #104]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #96]\n"
+    "movne %[n], #48\n"
+    "ldrh  %[tmp3], [%[blk], #90]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #98]\n"
+    "movne %[n], #49\n"
+    "ldrh  %[tmp1], [%[blk], #76]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #100]\n"
+    "movne %[n], #50\n"
+    "ldrh  %[tmp2], [%[blk], #62]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #102]\n"
+    "movne %[n], #51\n"
+    "ldrh  %[tmp3], [%[blk], #78]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #104]\n"
+    "movne %[n], #52\n"
+    "ldrh  %[tmp1], [%[blk], #92]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #106]\n"
+    "movne %[n], #53\n"
+    "ldrh  %[tmp2], [%[blk], #106]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #108]\n"
+    "movne %[n], #54\n"
+    "ldrh  %[tmp3], [%[blk], #120]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #110]\n"
+    "movne %[n], #55\n"
+    "ldrh  %[tmp1], [%[blk], #122]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #112]\n"
+    "movne %[n], #56\n"
+    "ldrh  %[tmp2], [%[blk], #108]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #114]\n"
+    "movne %[n], #57\n"
+    "ldrh  %[tmp3], [%[blk], #94]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #116]\n"
+    "movne %[n], #58\n"
+    "ldrh  %[tmp1], [%[blk], #110]\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #118]\n"
+    "movne %[n], #59\n"
+    "ldrh  %[tmp2], [%[blk], #124]\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #120]\n"
+    "movne %[n], #60\n"
+    "ldrh  %[tmp3], [%[blk], #126]\n"
+    "cmp   %[tmp1], #0\n"
+    "strh  %[tmp1], [%[out], #122]\n"
+    "movne %[n], #61\n"
+    "cmp   %[tmp2], #0\n"
+    "strh  %[tmp2], [%[out], #124]\n"
+    "movne %[n], #62\n"
+    "cmp   %[tmp3], #0\n"
+    "strh  %[tmp3], [%[out], #126]\n"
+    "movne %[n], #63\n"
+    : [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3),
+      [n] "+&r" (n)
+    : [blk] "r" (block), [out] "r" (out)
+    : "memory", "cc");
+  return n;
+}
+
+/*
+LOCAL(int)
+find_last_nonzero_index (JCOEFPTR block, JCOEFPTR out)
+{
+  int tmp, i, n = 0;
+  for (i = 1; i < DCTSIZE2; i++) {
+    if ((tmp = block[jpeg_natural_order[i]]) != 0)
+      n = i;
+    out[i] = tmp;
+  }
+  return n;
+}
+*/
+
+#endif
+
 /* Encode a single block's worth of coefficients */
 
 LOCAL(boolean)
@@ -518,6 +866,10 @@ encode_one_block (working_state * state, JCOEFPTR block, int last_dc_val,
   size_t put_buffer;  int put_bits;
   int code_0xf0 = actbl->ehufco[0xf0], size_0xf0 = actbl->ehufsi[0xf0];
   size_t bytes, bytestocopy;  int localbuf = 0;
+#ifdef __arm__
+  int last_nonzero_index, k;
+  JCOEF workspace[DCTSIZE2];
+#endif
 
   put_buffer = state->cur.put_buffer;
   put_bits = state->cur.put_bits;
@@ -552,6 +904,15 @@ encode_one_block (working_state * state, JCOEFPTR block, int last_dc_val,
     r = 0;  \
   }}
 
+#ifdef __arm__
+  last_nonzero_index = find_last_nonzero_index(block, workspace) * 2;
+  block = &workspace[0];
+  for (k = 2; k <= last_nonzero_index; k += 2) {
+      innerloop(k);
+  }
+  /* If the last coef(s) were zero, emit an end-of-block code */
+  if (k < DCTSIZE2 * 2) DUMP_SINGLE_VALUE(actbl, 0x0)
+#else
   innerloop(2*1);   innerloop(2*8);   innerloop(2*16);  innerloop(2*9);
   innerloop(2*2);   innerloop(2*3);   innerloop(2*10);  innerloop(2*17);
   innerloop(2*24);  innerloop(2*32);  innerloop(2*25);  innerloop(2*18);
@@ -571,6 +932,7 @@ encode_one_block (working_state * state, JCOEFPTR block, int last_dc_val,
 
   /* If the last coef(s) were zero, emit an end-of-block code */
   if (r > 0) DUMP_SINGLE_VALUE(actbl, 0x0)
+#endif
 
   state->cur.put_buffer = put_buffer;
   state->cur.put_bits = put_bits;