* elf64-sparc.c (sparc64_elf_relax_section): New.

	(sparc64_elf_relocate_section): Optimize tail call into branch always
	if possible.
	* elf32-sparc.c (elf32_sparc_relocate_section): Likewise.
	(elf32_sparc_relax_section): New.

3 files changed, 235 insertions, 1 deletions

diff --git a/bfd/ChangeLog b/bfd/ChangeLog
index 8a87d01..52e4190 100644
--- a/bfd/ChangeLog
+++ b/bfd/ChangeLog
@@ -1,5 +1,13 @@
 2000-05-26  Jakub Jelinek  <jakub@redhat.com>
 
+	* elf64-sparc.c (sparc64_elf_relax_section): New.
+	(sparc64_elf_relocate_section): Optimize tail call into branch always
+	if possible.
+	* elf32-sparc.c (elf32_sparc_relocate_section): Likewise.
+	(elf32_sparc_relax_section): New.
+
+2000-05-26  Jakub Jelinek  <jakub@redhat.com>
+
 	* elf64-sparc.c: Add ATTRIBUTE_UNUSED to unused function parameters.
 	Remove unusued variables.
 	(sparc64_elf_relocate_section): Change r_symndx type to unsigned long.
diff --git a/bfd/elf32-sparc.c b/bfd/elf32-sparc.c
index 1c8d0ba..70d65a5 100644
--- a/bfd/elf32-sparc.c
+++ b/bfd/elf32-sparc.c
@@ -24,6 +24,7 @@ Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.  */
 #include "libbfd.h"
 #include "elf-bfd.h"
 #include "elf/sparc.h"
+#include "opcode/sparc.h"
 
 static reloc_howto_type *elf32_sparc_reloc_type_lookup
   PARAMS ((bfd *, bfd_reloc_code_real_type));
@@ -36,6 +37,8 @@ static boolean elf32_sparc_adjust_dynamic_symbol
   PARAMS ((struct bfd_link_info *, struct elf_link_hash_entry *));
 static boolean elf32_sparc_size_dynamic_sections
   PARAMS ((bfd *, struct bfd_link_info *));
+static boolean elf32_sparc_relax_section
+  PARAMS ((bfd *, asection *, struct bfd_link_info *, boolean *));
 static boolean elf32_sparc_relocate_section
   PARAMS ((bfd *, struct bfd_link_info *, bfd *, asection *, bfd_byte *,
 	   Elf_Internal_Rela *, Elf_Internal_Sym *, asection **));
@@ -1059,6 +1062,23 @@ elf32_sparc_size_dynamic_sections (output_bfd, info)
   return true;
 }
 
+
+#define SET_SEC_DO_RELAX(section) do { elf_section_data(section)->tdata = (void *)1; } while (0)
+#define SEC_DO_RELAX(section) (elf_section_data(section)->tdata == (void *)1)
+
+/*ARGSUSED*/
+static boolean
+elf32_sparc_relax_section (abfd, section, link_info, again)
+     bfd *abfd ATTRIBUTE_UNUSED;
+     asection *section ATTRIBUTE_UNUSED;
+     struct bfd_link_info *link_info ATTRIBUTE_UNUSED;
+     boolean *again;
+{
+  *again = false;
+  SET_SEC_DO_RELAX (section);
+  return true;
+}
+
 /* Relocate a SPARC ELF section.  */
 
 static boolean
@@ -1518,6 +1538,7 @@ elf32_sparc_relocate_section (output_bfd, info, input_bfd, input_section,
 	  break;
 	}
 
+      r = bfd_reloc_continue;
       if (r_type == R_SPARC_WDISP16)
 	{
 	  bfd_vma x;
@@ -1549,7 +1570,97 @@ elf32_sparc_relocate_section (output_bfd, info, input_bfd, input_section,
 	  bfd_putl32 (/*input_bfd,*/ x, contents + rel->r_offset);
 	  r = bfd_reloc_ok;
 	}
-      else
+      else if ((r_type == R_SPARC_WDISP30 || r_type == R_SPARC_WPLT30)
+	       && SEC_DO_RELAX (input_section)
+	       && rel->r_offset + 4 < input_section->_raw_size)
+	{
+#define G0		0
+#define O7		15
+#define XCC		(2 << 20)
+#define COND(x)		(((x)&0xf)<<25)
+#define CONDA		COND(0x8)
+#define INSN_BPA	(F2(0,1) | CONDA | BPRED | XCC)
+#define INSN_BA		(F2(0,2) | CONDA)
+#define INSN_OR		F3(2, 0x2, 0)
+#define INSN_NOP	F2(0,4)
+
+	  bfd_vma x, y;
+
+	  /* If the instruction is a call with either:
+	     restore
+	     arithmetic instruction with rd == %o7
+	     where rs1 != %o7 and rs2 if it is register != %o7
+	     then we can optimize if the call destination is near
+	     by changing the call into a branch always.  */
+	  x = bfd_get_32 (input_bfd, contents + rel->r_offset);
+	  y = bfd_get_32 (input_bfd, contents + rel->r_offset + 4);
+	  if ((x & OP(~0)) == OP(1) && (y & OP(~0)) == OP(2))
+	    {
+	      if (((y & OP3(~0)) == OP3(0x3d) /* restore */
+		   || ((y & OP3(0x28)) == 0 /* arithmetic */
+		       && (y & RD(~0)) == RD(O7)))
+		  && (y & RS1(~0)) != RS1(O7)
+		  && ((y & F3I(~0))
+		      || (y & RS2(~0)) != RS2(O7)))
+		{
+		  bfd_vma reloc;
+
+		  reloc = relocation + rel->r_addend - rel->r_offset;
+		  reloc -= (input_section->output_section->vma
+			   + input_section->output_offset);
+
+		  /* Ensure the reloc fits into simm22.  */
+		  if ((reloc & 3) == 0
+		      && ((reloc & ~(bfd_vma)0x7fffff) == 0
+			  || ((reloc | 0x7fffff) == ~(bfd_vma)0)))
+		    {
+		      reloc >>= 2;
+		
+		      /* Check whether it fits into simm19 on v9.  */
+		      if (((reloc & 0x3c0000) == 0
+			   || (reloc & 0x3c0000) == 0x3c0000)
+			  && (elf_elfheader (output_bfd)->e_flags & EF_SPARC_32PLUS))
+			x = INSN_BPA | (reloc & 0x7ffff); /* ba,pt %xcc */
+		      else
+			x = INSN_BA | (reloc & 0x3fffff); /* ba */
+		      bfd_put_32 (input_bfd, x, contents + rel->r_offset);
+		      r = bfd_reloc_ok;
+		      if (rel->r_offset >= 4
+			  && (y & (0xffffffff ^ RS1(~0)))
+			      == (INSN_OR | RD(O7) | RS2(G0)))
+			{
+			  bfd_vma z;
+			  unsigned int reg;
+
+			  z = bfd_get_32 (input_bfd,
+					  contents + rel->r_offset - 4);
+			  if ((z & (0xffffffff ^ RD(~0)))
+			      != (INSN_OR | RS1(O7) | RS2(G0)))
+			    break;
+
+			  /* The sequence was
+			     or %o7, %g0, %rN
+			     call foo
+			     or %rN, %g0, %o7
+
+			     If call foo was replaced with ba, replace
+			     or %rN, %g0, %o7 with nop.  */
+
+			  reg = (y & RS1(~0)) >> 14;
+			  if (reg != ((z & RD(~0)) >> 25)
+			      || reg == G0 || reg == O7)
+			    break;
+
+			  bfd_put_32 (input_bfd, INSN_NOP,
+				      contents + rel->r_offset + 4);
+			}
+
+		    }
+		}
+	    }
+	}
+
+      if (r == bfd_reloc_continue)
 	r = _bfd_final_link_relocate (howto, input_bfd, input_section,
 				      contents, rel->r_offset,
 				      relocation, rel->r_addend);
@@ -1967,6 +2078,7 @@ elf32_sparc_final_write_processing (abfd, linker)
 #define ELF_MAXPAGESIZE 0x10000
 
 #define bfd_elf32_bfd_reloc_type_lookup	elf32_sparc_reloc_type_lookup
+#define bfd_elf32_bfd_relax_section	elf32_sparc_relax_section
 #define elf_info_to_howto		elf32_sparc_info_to_howto
 #define elf_backend_create_dynamic_sections \
 					_bfd_elf_create_dynamic_sections
diff --git a/bfd/elf64-sparc.c b/bfd/elf64-sparc.c
index c9ac36c..ca2e7c1 100644
--- a/bfd/elf64-sparc.c
+++ b/bfd/elf64-sparc.c
@@ -22,6 +22,7 @@ Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.  */
 #include "sysdep.h"
 #include "libbfd.h"
 #include "elf-bfd.h"
+#include "opcode/sparc.h"
 
 /* This is defined if one wants to build upward compatible binaries
    with the original sparc64-elf toolchain.  The support is kept in for
@@ -65,6 +66,8 @@ static void sparc64_elf_symbol_processing
 static boolean sparc64_elf_merge_private_bfd_data
   PARAMS ((bfd *, bfd *));
 
+static boolean sparc64_elf_relax_section
+  PARAMS ((bfd *, asection *, struct bfd_link_info *, boolean *));
 static boolean sparc64_elf_relocate_section
   PARAMS ((bfd *, struct bfd_link_info *, bfd *, asection *, bfd_byte *,
 	   Elf_Internal_Rela *, Elf_Internal_Sym *, asection **));
@@ -1853,6 +1856,22 @@ sparc64_elf_size_dynamic_sections (output_bfd, info)
   return true;
 }
 
+#define SET_SEC_DO_RELAX(section) do { elf_section_data(section)->tdata = (void *)1; } while (0)
+#define SEC_DO_RELAX(section) (elf_section_data(section)->tdata == (void *)1)
+
+/*ARGSUSED*/
+static boolean
+sparc64_elf_relax_section (abfd, section, link_info, again)
+     bfd *abfd ATTRIBUTE_UNUSED;
+     asection *section ATTRIBUTE_UNUSED;
+     struct bfd_link_info *link_info ATTRIBUTE_UNUSED;
+     boolean *again;
+{
+  *again = false;
+  SET_SEC_DO_RELAX (section);
+  return true;
+}
+
 /* Relocate a SPARC64 ELF section.  */
 
 static boolean
@@ -2394,6 +2413,8 @@ sparc64_elf_relocate_section (output_bfd, info, input_bfd, input_section,
 	  relocation = (splt->output_section->vma
 			+ splt->output_offset
 			+ sparc64_elf_plt_entry_offset (h->plt.offset));
+	  if (r_type == R_SPARC_WPLT30)
+	    goto do_wplt30;
 	  goto do_default;
 
 	case R_SPARC_OLO10:
@@ -2469,6 +2490,97 @@ sparc64_elf_relocate_section (output_bfd, info, input_bfd, input_section,
 	  }
 	  break;
 
+	case R_SPARC_WDISP30:
+	do_wplt30:
+	  if (SEC_DO_RELAX (input_section)
+	      && rel->r_offset + 4 < input_section->_raw_size)
+	    {
+#define G0		0
+#define O7		15
+#define XCC		(2 << 20)
+#define COND(x)		(((x)&0xf)<<25)
+#define CONDA		COND(0x8)
+#define INSN_BPA	(F2(0,1) | CONDA | BPRED | XCC)
+#define INSN_BA		(F2(0,2) | CONDA)
+#define INSN_OR		F3(2, 0x2, 0)
+#define INSN_NOP	F2(0,4)
+
+	      bfd_vma x, y;
+
+	      /* If the instruction is a call with either:
+		 restore
+		 arithmetic instruction with rd == %o7
+		 where rs1 != %o7 and rs2 if it is register != %o7
+		 then we can optimize if the call destination is near
+		 by changing the call into a branch always.  */
+	      x = bfd_get_32 (input_bfd, contents + rel->r_offset);
+	      y = bfd_get_32 (input_bfd, contents + rel->r_offset + 4);
+	      if ((x & OP(~0)) == OP(1) && (y & OP(~0)) == OP(2))
+		{
+		  if (((y & OP3(~0)) == OP3(0x3d) /* restore */
+		       || ((y & OP3(0x28)) == 0 /* arithmetic */
+			   && (y & RD(~0)) == RD(O7)))
+		      && (y & RS1(~0)) != RS1(O7)
+		      && ((y & F3I(~0))
+			  || (y & RS2(~0)) != RS2(O7)))
+		    {
+		      bfd_vma reloc;
+
+		      reloc = relocation + rel->r_addend - rel->r_offset;
+		      reloc -= (input_section->output_section->vma
+				+ input_section->output_offset);
+		      if (reloc & 3)
+			goto do_default;
+
+		      /* Ensure the branch fits into simm22.  */
+		      if ((reloc & ~(bfd_vma)0x7fffff)
+			   && ((reloc | 0x7fffff) != MINUS_ONE))
+			goto do_default;
+		      reloc >>= 2;
+
+		      /* Check whether it fits into simm19.  */
+		      if ((reloc & 0x3c0000) == 0
+			  || (reloc & 0x3c0000) == 0x3c0000)
+			x = INSN_BPA | (reloc & 0x7ffff); /* ba,pt %xcc */
+		      else
+			x = INSN_BA | (reloc & 0x3fffff); /* ba */
+		      bfd_put_32 (input_bfd, x, contents + rel->r_offset);
+		      r = bfd_reloc_ok;
+		      if (rel->r_offset >= 4
+			  && (y & (0xffffffff ^ RS1(~0)))
+			     == (INSN_OR | RD(O7) | RS2(G0)))
+			{
+			  bfd_vma z;
+			  unsigned int reg;
+
+			  z = bfd_get_32 (input_bfd,
+					  contents + rel->r_offset - 4);
+			  if ((z & (0xffffffff ^ RD(~0)))
+			      != (INSN_OR | RS1(O7) | RS2(G0)))
+			    break;
+
+			  /* The sequence was
+			     or %o7, %g0, %rN
+			     call foo
+			     or %rN, %g0, %o7
+
+			     If call foo was replaced with ba, replace
+			     or %rN, %g0, %o7 with nop.  */
+
+			  reg = (y & RS1(~0)) >> 14;
+			  if (reg != ((z & RD(~0)) >> 25)
+			      || reg == G0 || reg == O7)
+			    break;
+
+			  bfd_put_32 (input_bfd, INSN_NOP,
+				      contents + rel->r_offset + 4);
+			}
+		      break;
+		    }
+		}
+	    }
+	  /* FALLTHROUGH */
+
 	default:
 	do_default:
 	  r = _bfd_final_link_relocate (howto, input_bfd, input_section,
@@ -2961,6 +3073,8 @@ const struct elf_size_info sparc64_elf_size_info =
   sparc64_elf_canonicalize_dynamic_reloc
 #define bfd_elf64_bfd_reloc_type_lookup \
   sparc64_elf_reloc_type_lookup
+#define bfd_elf64_bfd_relax_section \
+  sparc64_elf_relax_section
 
 #define elf_backend_create_dynamic_sections \
   _bfd_elf_create_dynamic_sections