Reorganize optimization defines

Enable x86_64 to skip checking for SSE2
Enable deflate_medium on all archs

diff --git a/arch/x86/crc_folding.c b/arch/x86/crc_folding.c
index 87f8c9dcde1fe363ac0f648a0131f079ef4ec7dd..9f4a2baf26a33def1fd56e576455346624c072c6 100644 (file)
--- a/arch/x86/crc_folding.c
+++ b/arch/x86/crc_folding.c
@@ -16,7 +16,7 @@
  * For conditions of distribution and use, see copyright notice in zlib.h
  */
 
-#ifdef HAVE_PCLMULQDQ
+#ifdef X86_PCLMULQDQ_CRC
 
 #include "deflate.h"
 

diff --git a/arch/x86/fill_window_sse.c b/arch/x86/fill_window_sse.c
index bb4f521b24acbe98cff051e1cec6a2f166eebe4c..ef666e2e218d21064ab8115e0bff6bb4ddbe6d66 100644 (file)
--- a/arch/x86/fill_window_sse.c
+++ b/arch/x86/fill_window_sse.c
@@ -8,7 +8,7 @@
  *
  * For conditions of distribution and use, see copyright notice in zlib.h
  */
-#ifdef HAVE_SSE2
+#ifdef X86_SSE2_FILL_WINDOW
 
 #include <immintrin.h>
 #include "deflate.h"

diff --git a/configure b/configure
index 28a1657782f9255c0d8f80e4cb5733c1d5ec1bd7..718676e3d5b18f22310f5020c4a25331387ab168 100755 (executable)
--- a/configure
+++ b/configure
@@ -634,6 +634,12 @@ else
     HAVE_PCLMULQDQ_INTRIN=0
 fi
 
+# Enable deflate_medium at level 4-6
+if test $without_new_strategies -eq 0; then
+    CFLAGS="${CFLAGS} -DMEDIUM_STRATEGY"
+    SFLAGS="${SFLAGS} -DMEDIUM_STRATEGY"
+fi
+
 ARCHDIR=""
 ARCH_STATIC_OBJS=""
 ARCH_SHARED_OBJS=""
@@ -645,8 +651,8 @@ case "${ARCH}" in
 
         case "${ARCH}" in
             x86_64)
-                CFLAGS="${CFLAGS} -DX86_64"
-                SFLAGS="${SFLAGS} -DX86_64"
+                CFLAGS="${CFLAGS} -DX86_64 -DX86_NOCHECK_SSE2"
+                SFLAGS="${SFLAGS} -DX86_64 -DX86_NOCHECK_SSE2"
             ;;
             i386 | i486 | i586 | i686)
                 CFLAGS="${CFLAGS} -DX86"
@@ -654,8 +660,8 @@ case "${ARCH}" in
             ;;
         esac
 
-        CFLAGS="${CFLAGS} -DUNALIGNED_OK -DADLER32_UNROLL_LESS -DCRC32_UNROLL_LESS"
-        SFLAGS="${SFLAGS} -DUNALIGNED_OK -DADLER32_UNROLL_LESS -DCRC32_UNROLL_LESS"
+        CFLAGS="${CFLAGS} -DUNALIGNED_OK -DADLER32_UNROLL_LESS -DCRC32_UNROLL_LESS -DX86_CPUID"
+        SFLAGS="${SFLAGS} -DUNALIGNED_OK -DADLER32_UNROLL_LESS -DCRC32_UNROLL_LESS -DX86_CPUID"
 
        # Enable arch-specific optimizations?
        if test $without_optimizations -eq 0; then
@@ -663,34 +669,36 @@ case "${ARCH}" in
             ARCH_SHARED_OBJS="${ARCH_SHARED_OBJS} x86.lo"
 
             if test ${HAVE_SSE2_INTRIN} -eq 1; then
-                CFLAGS="${CFLAGS} -UCHECK_SSE2 -DHAVE_SSE2"
-                SFLAGS="${SFLAGS} -UCHECK_SSE2 -DHAVE_SSE2"
+                CFLAGS="${CFLAGS} -DX86_SSE2_FILL_WINDOW"
+                SFLAGS="${SFLAGS} -DX86_SSE2_FILL_WINDOW"
                 ARCH_STATIC_OBJS="${ARCH_STATIC_OBJS} fill_window_sse.o"
                 ARCH_SHARED_OBJS="${ARCH_SHARED_OBJS} fill_window_sse.lo"
             fi
 
-            CFLAGS="${CFLAGS} -DUSE_SSE4_2_CRC_HASH"
-            SFLAGS="${SFLAGS} -DUSE_SSE4_2_CRC_HASH"
+            CFLAGS="${CFLAGS} -DX86_SSE4_2_CRC_HASH"
+            SFLAGS="${SFLAGS} -DX86_SSE4_2_CRC_HASH"
 
             if test ${HAVE_PCLMULQDQ_INTRIN} -eq 1; then
-                CFLAGS="${CFLAGS} -DHAVE_PCLMULQDQ"
-                SFLAGS="${SFLAGS} -DHAVE_PCLMULQDQ"
+                CFLAGS="${CFLAGS} -DX86_PCLMULQDQ_CRC"
+                SFLAGS="${SFLAGS} -DX86_PCLMULQDQ_CRC"
                 ARCH_STATIC_OBJS="${ARCH_STATIC_OBJS} crc_folding.o"
                 ARCH_SHARED_OBJS="${ARCH_SHARED_OBJS} crc_folding.lo"
-           fi
-       fi
+            fi
 
-       # Enable deflate_quick at level 1?
-       if test $without_new_strategies -eq 0; then
-            CFLAGS="${CFLAGS} -DUSE_QUICK -DUSE_MEDIUM"
-            SFLAGS="${SFLAGS} -DUSE_QUICK -DUSE_MEDIUM"
+           # Enable deflate_quick at level 1?
+            if test $without_new_strategies -eq 0; then
+                CFLAGS="${CFLAGS} -DX86_QUICK_STRATEGY"
+                SFLAGS="${SFLAGS} -DX86_QUICK_STRATEGY"
 
-            ARCH_STATIC_OBJS="${ARCH_STATIC_OBJS} deflate_quick.o"
-            ARCH_SHARED_OBJS="${ARCH_SHARED_OBJS} deflate_quick.lo"
+                ARCH_STATIC_OBJS="${ARCH_STATIC_OBJS} deflate_quick.o"
+                ARCH_SHARED_OBJS="${ARCH_SHARED_OBJS} deflate_quick.lo"
+            fi
         fi
     ;;
 esac
 
+
+
 # show the results in the log
 echo >> configure.log
 echo ALL = $ALL >> configure.log
@@ -722,12 +730,9 @@ echo mandir = $mandir >> configure.log
 echo prefix = $prefix >> configure.log
 echo sharedlibdir = $sharedlibdir >> configure.log
 echo uname = $uname >> configure.log
-echo FILL_WINDOW_SSE_o = ${FILL_WINDOW_SSE_o} >> configure.log
-echo FILL_WINDOW_SSE_lo= ${FILL_WINDOW_SSE_lo} >> configure.log
-echo CRC_FOLDING_o = ${CRC_FOLDING_o} >> configure.log
-echo CRC_FOLDING_lo= ${CRC_FOLDING_lo} >> configure.log
-echo DEFLATE_QUICK_o=${DEFLATE_QUICK_o} >> configure.log
-echo DEFLATE_QUICK_lo=${DEFLATE_QUICK_lo} >> configure.log
+echo ARCHDIR = ${ARCHDIR} >> configure.log
+echo ARCH_STATIC_OBJS = ${ARCH_STATIC_OBJS} >> configure.log
+echo ARCH_SHARED_OBJS = ${ARCH_SHARED_OBJS} >> configure.log
 
 # udpate Makefile with the configure results
 sed < Makefile.in "

diff --git a/crc32.c b/crc32.c
index 92499d57cbf305cf8e1cfc77284f8c85bde04dd4..4b2a386949cb6393c3bd29a4e8d74493a99e7599 100644 (file)
--- a/crc32.c
+++ b/crc32.c
@@ -437,7 +437,7 @@ uLong ZEXPORT crc32_combine64(crc1, crc2, len2)
 
 #include "deflate.h"
 
-#ifdef HAVE_PCLMULQDQ
+#ifdef X86_PCLMULQDQ_CRC
 #include "arch/x86/x86.h"
 extern void ZLIB_INTERNAL crc_fold_init(deflate_state *z_const s);
 extern void ZLIB_INTERNAL crc_fold_copy(deflate_state *z_const s,
@@ -447,7 +447,7 @@ extern unsigned ZLIB_INTERNAL crc_fold_512to32(deflate_state *z_const s);
 
 ZLIB_INTERNAL void crc_reset(deflate_state *const s)
 {
-#ifdef HAVE_PCLMULQDQ
+#ifdef X86_PCLMULQDQ_CRC
     if (x86_cpu_has_pclmulqdq) {
         crc_fold_init(s);
         return;
@@ -458,7 +458,7 @@ ZLIB_INTERNAL void crc_reset(deflate_state *const s)
 
 ZLIB_INTERNAL void crc_finalize(deflate_state *const s)
 {
-#ifdef HAVE_PCLMULQDQ
+#ifdef X86_PCLMULQDQ_CRC
     if (x86_cpu_has_pclmulqdq)
         s->strm->adler = crc_fold_512to32(s);
 #endif
@@ -466,7 +466,7 @@ ZLIB_INTERNAL void crc_finalize(deflate_state *const s)
 
 ZLIB_INTERNAL void copy_with_crc(z_streamp strm, Bytef *dst, long size)
 {
-#ifdef HAVE_PCLMULQDQ
+#ifdef X86_PCLMULQDQ_CRC
     if (x86_cpu_has_pclmulqdq) {
         crc_fold_copy(strm->state, dst, strm->next_in, size);
         return;

diff --git a/deflate.c b/deflate.c
index 002fca4aab982116769abd4aabb938298b715b33..5bfc7d25377857468f48b636ff032cf3bdc2a162 100644 (file)
--- a/deflate.c
+++ b/deflate.c
@@ -51,7 +51,7 @@
 
 #include "deflate.h"
 
-#if defined(CHECK_SSE2) || defined(USE_SSE4_2_CRC_HASH) || defined(USE_QUICK)
+#if defined(X86_CPUID)
 #include "arch/x86/x86.h"
 #endif
 
@@ -126,17 +126,18 @@ typedef struct config_s {
 local const config configuration_table[10] = {
 /*      good lazy nice chain */
 /* 0 */ {0,    0,  0,    0, deflate_stored},  /* store only */
-#ifdef USE_QUICK
+
+#ifdef X86_QUICK_STRATEGY
 /* 1 */ {4,    4,  8,    4, deflate_quick},
 /* 2 */ {4,    4,  8,    4, deflate_fast}, /* max speed, no lazy matches */
-/* 3 */ {4,    6, 32,   32, deflate_fast},
 #else
 /* 1 */ {4,    4,  8,    4, deflate_fast}, /* max speed, no lazy matches */
 /* 2 */ {4,    5, 16,    8, deflate_fast},
-/* 3 */ {4,    6, 32,   32, deflate_fast},
 #endif
 
-#ifdef USE_MEDIUM
+/* 3 */ {4,    6, 32,   32, deflate_fast},
+
+#ifdef MEDIUM_STRATEGY
 /* 4 */ {4,    4, 16,   16, deflate_medium},  /* lazy matches */
 /* 5 */ {8,   16, 32,   32, deflate_medium},
 /* 6 */ {8,   16, 128, 128, deflate_medium},
@@ -174,7 +175,7 @@ struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
  *    input characters and the first MIN_MATCH bytes of str are valid
  *    (except for the last MIN_MATCH-1 bytes of the input file).
  */
-#ifdef USE_SSE4_2_CRC_HASH
+#ifdef X86_SSE4_2_CRC_HASH
 local inline Pos insert_string_sse(deflate_state *z_const s, z_const Pos str)
 {
     Pos ret;
@@ -212,7 +213,7 @@ local inline Pos insert_string_c(deflate_state *z_const s, z_const Pos str)
 
 local inline Pos insert_string(deflate_state *z_const s, z_const Pos str)
 {
-#ifdef USE_SSE4_2_CRC_HASH
+#ifdef X86_SSE4_2_CRC_HASH
     if (x86_cpu_has_sse42)
         return insert_string_sse(s, str);
 #endif
@@ -272,7 +273,7 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
      * output size for (length,distance) codes is <= 24 bits.
      */
 
-#if defined(CHECK_SSE2) || defined(USE_SSE4_2_CRC_HASH)
+#if defined(X86_SSE2_FILL_WINDOW) || defined(X86_SSE4_2_CRC_HASH)
     x86_check_features();
 #endif
 
@@ -317,7 +318,7 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
     }
     if (windowBits == 8) windowBits = 9;  /* until 256-byte window bug fixed */
 
-#ifdef USE_QUICK
+#ifdef X86_QUICK_STRATEGY
     if (level == 1)
         windowBits = 13;
 #endif
@@ -333,7 +334,7 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
     s->w_size = 1 << s->w_bits;
     s->w_mask = s->w_size - 1;
 
-#ifdef USE_SSE4_2_CRC_HASH
+#ifdef X86_SSE4_2_CRC_HASH
     if (x86_cpu_has_sse42)
         s->hash_bits = 15;
     else
@@ -344,7 +345,7 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
     s->hash_mask = s->hash_size - 1;
     s->hash_shift =  ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
 
-#ifdef HAVE_PCLMULQDQ
+#ifdef X86_PCLMULQDQ_CRC
     window_padding = 8;
 #endif
 
@@ -952,7 +953,7 @@ int ZEXPORT deflate (strm, flush)
         (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
         block_state bstate;
 
-#ifdef USE_QUICK
+#ifdef X86_QUICK_STRATEGY
         if (s->level == 1 && !x86_cpu_has_sse42) 
             bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
                     (s->strategy == Z_RLE ? deflate_rle(s, flush) :
@@ -1224,21 +1225,21 @@ local void check_match(s, start, match, length)
  *    performed for at least two bytes (required for the zip translate_eol
  *    option -- not supported here).
  */
-#ifdef HAVE_SSE2
+#ifdef X86_SSE2_FILL_WINDOW
 extern void fill_window_sse(deflate_state *s);
 #endif
 local void fill_window_c(deflate_state *s);
 
 local void fill_window(deflate_state *s)
 {
-#ifdef HAVE_SSE2
-#ifdef CHECK_SSE2
+#ifdef X86_SSE2_FILL_WINDOW
+#ifndef X86_NOCHECK_SSE2
     if (x86_cpu_has_sse2) {
 #endif
         fill_window_sse(s);
         return;
-#ifdef CHECK_SSE2
-    }
+#ifndef X86_NOCHECK_SSE2
+  }
 #endif
 #endif
     
@@ -1609,7 +1610,7 @@ local block_state deflate_fast(s, flush)
 }
 
 
-#ifdef USE_MEDIUM
+#ifdef MEDIUM_STRATEGY
 #include "deflate_medium.c"
 #endif
 

diff --git a/deflate.h b/deflate.h
index 8ba824f6456ce09c53749b5fac50bdf112f02594..e1f44b403d0a5ccb72f6918a07f0434ab197fa73 100644 (file)
--- a/deflate.h
+++ b/deflate.h
@@ -110,7 +110,7 @@ typedef struct internal_state {
     Byte  method;        /* can only be DEFLATED */
     int   last_flush;    /* value of flush param for previous deflate call */
 
-#ifdef HAVE_PCLMULQDQ
+#ifdef X86_PCLMULQDQ_CRC
     unsigned __attribute__((aligned(16))) crc0[4 * 5];
 #endif
 
@@ -386,7 +386,7 @@ void ZLIB_INTERNAL bi_windup (deflate_state *s);
  *    input characters, so that a running hash key can be computed from the
  *    previous key instead of complete recalculation each time.
  */
-#ifdef USE_SSE4_2_CRC_HASH
+#ifdef X86_SSE4_2_CRC_HASH
 #define UPDATE_HASH(s,h,i) (\
     {\
         if (s->level < 6) \