2003-02-01 Richard Henderson <rth@redhat.com>

	* optabs.c (expand_unop): Use word_mode for outmode of bit scaners.
	* libgcc2.c (__ffsdi2, __clzsi2, __clzdi2, __ctzsi2, __ctzdi2,
	__popcountsi2, __popcountdi2, __paritysi2 __paritydi2): Change
	return type to Wtype.

	* libgcc-std.ver (GCC_3.4): Fix inheritance.

	* config/i386/i386.md (ffssi2): Use nonimmediate_operand for
	expander input constraint.

2003-02-01  Falk Hueffner  <falk.hueffner@student.uni-tuebingen.de>

        * optabs.h (optab_index): Add OTI_clz, OTI_ctz, OTI_popcount and
        OTI_parity.
        (clz_optab, ctz_optab, popcount_optab, parity_optab): New.
        * optabs.c (widen_clz, expand_parity): New.
        (expand_unop): Handle clz and parity.  Hardcode SImode as outmode
        for libcalls to clz, ctz, popcount, and parity.
        (init_optabs): Init clz_optab, ctz_optab, popcount_optab and
        parity_optab, and set up libfunc handlers.
        * libgcc2.c (__clzsi2, __clzdi2, __ctzsi2, __ctzdi2,
        __popcountsi2, __popcountdi2, __paritysi2 __paritydi2,
        __popcount_tab): New.
        * libgcc2.h: Declare them.
        * libgcc-std.ver (GCC_3.4): Add new functions from libgcc2.c.
        * genopinit.c (optabs): Add clz_optab, ctz_optab, popcount_optab
        and parity_optab.
        * builtin-types.def (BT_FN_INT_LONG, BT_FN_INT_LONGLONG): New.
        * builtins.def (BUILT_IN_CLZ, BUILT_IN_CTZ, BUILT_IN_POPCOUNT,
        BUILT_IN_PARITY, BUILT_IN_FFSL, BUILT_IN_CLZL, BUILT_IN_CTZL,
        BUILT_IN_POPCOUNTL, BUILT_IN_PARITYL, BUILT_IN_FFSLL,
        BUILT_IN_CLZLL, BUILT_IN_CTZLL, BUILT_IN_POPCOUNTLL,
        BUILT_IN_PARITYLL): New.
        * builtins.c (expand_builtin_unop): Rename from expand_builtin_ffs
        and add optab argument.
        (expand_builtin): Expand BUILT_IN_{FFS,CLZ,POPCOUNT,PARITY}*.
        * tree.def (CLZ_EXPR, CTZ_EXPR, POPCOUNT_EXPR, PARITY_EXPR): New.
        * expr.c (expand_expr): Handle them.
        * fold-const.c (tree_expr_nonnegative_p): Likewise.
        * rtl.def (CLZ, CTZ, POPCOUNT, PARITY): New.
        * reload1.c (eliminate_regs): Handle them.
        (elimination_effects): Likewise.
        * function.c (instantiate_virtual_regs_1): Likewise
        * genattrtab.c (check_attr_value): Likewise.
        * simplify-rtx.c (simplify_unary_operation): Likewise.
        * c-common.c (c_common_truthvalue_conversion): Handle POPCOUNT_EXPR.
        * combine.c (combine_simplify_rtx): Handle POPCOUNT and PARITY.
        (nonzero_bits): Handle CLZ, CTZ, POPCOUNT and PARITY.
        * config/alpha/alpha.md (clzdi2, ctzdi2, popcountdi2): New.
        * config/arm/arm.c (arm_init_builtins): Rename __builtin_clz to
        __builtin_arm_clz.
        * Makefile.in (LIB2FUNCS_1, LIB2FUNCS_2): Move...
        * mklibgcc.in (lib2funcs): ...here and merge.  Add new members.
        * doc/extend.texi (Other Builtins): Add new builtins.
        * doc/md.texi (Standard Names): Add new patterns.


git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@62252 138bc75d-0d04-0410-961f-82ee72b054a4

1 files changed, 151 insertions, 1 deletions

diff --git a/gcc/libgcc2.c b/gcc/libgcc2.c
index deb772fa629..5de1edecd8c 100644
--- a/gcc/libgcc2.c
+++ b/gcc/libgcc2.c
@@ -331,7 +331,7 @@ __ashrdi3 (DWtype u, word_type b)
 #endif
 
 #ifdef L_ffsdi2
-DWtype
+Wtype
 __ffsdi2 (DWtype u)
 {
   DWunion uu;
@@ -495,6 +495,11 @@ __udiv_w_sdiv (UWtype *rp __attribute__ ((__unused__)),
 #define L_udivmoddi4
 #endif
 
+#if (defined (L_clzsi2) || defined (L_clzdi2) || \
+     defined (L_ctzsi2) || defined (L_ctzdi2))
+extern const UQItype __clz_tab[];
+#endif
+
 #ifdef L_clz
 const UQItype __clz_tab[] =
 {
@@ -508,6 +513,151 @@ const UQItype __clz_tab[] =
   8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
 };
 #endif
+
+#ifdef L_clzsi2
+Wtype
+__clzsi2 (USItype x)
+{
+  Wtype a;
+
+  /* Note that we've already verified that BITS_PER_UNIT == 8, and
+     thus SItype is 32 bits wide.  */
+  if (x < (1 << 2 * 8))
+    if (x < (1 << 1 * 8))
+      a = 0 * 8;
+    else
+      a = 1 * 8;
+  else
+    if (x < (1 << 3 * 8))
+      a = 2 * 8;
+    else
+      a = 3 * 8;
+
+  return 32 - (__clz_tab[x >> a] + a);
+}
+#endif
+
+#ifdef L_clzdi2
+Wtype
+__clzdi2 (UDItype x)
+{
+  Wtype a;
+
+  /* Note that we've already verified that BITS_PER_UNIT == 8, and
+     thus DItype is 64 bits wide.  */
+  for (a = 64 - 8; a > 0; a -= 8)
+    if (((x >> a) & 0xff) != 0)
+      break;
+
+  return 64 - (__clz_tab[x >> a] + a);
+}
+#endif
+
+#ifdef L_ctzsi2
+Wtype
+__ctzsi2 (USItype x)
+{
+  Wtype a;
+
+  x = x & -x;
+
+  /* Note that we've already verified that BITS_PER_UNIT == 8, and
+     thus SItype is 32 bits wide.  */
+  if (x < (1 << 2 * 8))
+    if (x < (1 << 1 * 8))
+      a = 0 * 8;
+    else
+      a = 1 * 8;
+  else
+    if (x < (1 << 3 * 8))
+      a = 2 * 8;
+    else
+      a = 3 * 8;
+
+  return __clz_tab[x >> a] + a - 1;
+}
+#endif
+
+#ifdef L_ctzdi2
+Wtype
+__ctzdi2 (UDItype x)
+{
+  Wtype a;
+
+  x = x & -x;
+  for (a = 64 - 8; a > 0; a -= 8)
+    if (((x >> a) & 0xff) != 0)
+      break;
+
+  return __clz_tab[x >> a] + a - 1;
+}
+#endif
+
+#if (defined (L_popcountsi2) || defined (L_popcountdi2) || \
+     defined (L_paritysi2) || defined (L_paritydi2))
+extern const UQItype __popcount_tab[];
+#endif
+
+#ifdef L_popcount_tab
+const UQItype __popcount_tab[] =
+{
+    0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,1,2,2,3,2,3,3,4,2,3,3,4,3,4,4,5,
+    1,2,2,3,2,3,3,4,2,3,3,4,3,4,4,5,2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,
+    1,2,2,3,2,3,3,4,2,3,3,4,3,4,4,5,2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,
+    2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,3,4,4,5,4,5,5,6,4,5,5,6,5,6,6,7,
+    1,2,2,3,2,3,3,4,2,3,3,4,3,4,4,5,2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,
+    2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,3,4,4,5,4,5,5,6,4,5,5,6,5,6,6,7,
+    2,3,3,4,3,4,4,5,3,4,4,5,4,5,5,6,3,4,4,5,4,5,5,6,4,5,5,6,5,6,6,7,
+    3,4,4,5,4,5,5,6,4,5,5,6,5,6,6,7,4,5,5,6,5,6,6,7,5,6,6,7,6,7,7,8,
+};
+#endif
+
+#ifdef L_popcountsi2
+Wtype
+__popcountsi2 (USItype x)
+{
+  return __popcount_tab[(x >>  0) & 0xff]
+       + __popcount_tab[(x >>  8) & 0xff]
+       + __popcount_tab[(x >> 16) & 0xff]
+       + __popcount_tab[(x >> 24) & 0xff];
+}
+#endif
+
+#ifdef L_popcountdi2
+Wtype
+__popcountdi2 (UDItype x)
+{
+  return __popcount_tab[(x >>  0) & 0xff]
+       + __popcount_tab[(x >>  8) & 0xff]
+       + __popcount_tab[(x >> 16) & 0xff]
+       + __popcount_tab[(x >> 24) & 0xff]
+       + __popcount_tab[(x >> 32) & 0xff]
+       + __popcount_tab[(x >> 40) & 0xff]
+       + __popcount_tab[(x >> 48) & 0xff]
+       + __popcount_tab[(x >> 56) & 0xff];
+}
+#endif
+
+#ifdef L_paritysi2
+Wtype
+__paritysi2 (USItype x)
+{
+  x ^= x >> 16;
+  x ^= x >> 8;
+  return __popcount_tab[x & 0xff] & 1;
+}
+#endif
+
+#ifdef L_paritydi2
+Wtype
+__paritydi2 (UDItype x)
+{
+  Wtype nx = x ^ (x >> 32);
+  nx ^= nx >> 16;
+  nx ^= nx >> 8;
+  return __popcount_tab[nx & 0xff] & 1;
+}
+#endif
 
 #ifdef L_udivmoddi4