1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/patches/gcc/4.3.0/170-x86-emit-cld.patch Mon Jul 28 21:32:33 2008 +0000
1.3 @@ -0,0 +1,156 @@
1.4 +Original patch from gentoo: gentoo/src/patchsets/gcc/4.3.0/gentoo/20_all_gcc-x86-emit-cld.patch
1.5 +bandaid for the cld issue. to be dropped when gcc-4.3 goes stable.
1.6 +
1.7 +http://gcc.gnu.org/ml/gcc-patches/2008-03/msg00417.html
1.8 +
1.9 +2008-03-06 Uros Bizjak <ubizjak@gmail.com>
1.10 +
1.11 + * config/i386/i386.h (TARGET_CLD): New define.
1.12 + (struct machine_function): Add needs_cld field.
1.13 + (ix86_current_function_needs_cld): New define.
1.14 + * config/i386/i386.md (UNSPEC_CLD): New unspec volatile constant.
1.15 + ("cld"): New isns pattern.
1.16 + ("strmov_singleop"): Set ix86_current_function_needs_cld flag.
1.17 + ("rep_mov"): Ditto.
1.18 + ("strset_singleop"): Ditto.
1.19 + ("rep_stos"): Ditto.
1.20 + ("cmpstrnqi_nz_1"): Ditto.
1.21 + ("cmpstrnqi_1"): Ditto.
1.22 + ("strlenqi_1"): Ditto.
1.23 + * config/i386/i386.c (ix86_expand_prologue): Emit cld insn for
1.24 + TARGET_CLD when ix86_current_function_needs_cld is set.
1.25 +
1.26 +diff -durN gcc-4.3.0.orig/gcc/config/i386/i386.c gcc-4.3.0/gcc/config/i386/i386.c
1.27 +--- gcc-4.3.0.orig/gcc/config/i386/i386.c 2008-02-21 13:30:00.000000000 +0100
1.28 ++++ gcc-4.3.0/gcc/config/i386/i386.c 2008-06-10 14:44:37.000000000 +0200
1.29 +@@ -6448,6 +6448,10 @@
1.30 + emit_insn (gen_prologue_use (pic_offset_table_rtx));
1.31 + emit_insn (gen_blockage ());
1.32 + }
1.33 ++
1.34 ++ /* Emit cld instruction if stringops are used in the function. */
1.35 ++ if (TARGET_CLD && ix86_current_function_needs_cld)
1.36 ++ emit_insn (gen_cld ());
1.37 + }
1.38 +
1.39 + /* Emit code to restore saved registers using MOV insns. First register
1.40 +diff -durN gcc-4.3.0.orig/gcc/config/i386/i386.h gcc-4.3.0/gcc/config/i386/i386.h
1.41 +--- gcc-4.3.0.orig/gcc/config/i386/i386.h 2008-02-15 09:12:02.000000000 +0100
1.42 ++++ gcc-4.3.0/gcc/config/i386/i386.h 2008-06-10 14:44:37.000000000 +0200
1.43 +@@ -388,6 +388,7 @@
1.44 +
1.45 + extern int x86_prefetch_sse;
1.46 +
1.47 ++#define TARGET_CLD 1
1.48 + #define TARGET_ABM x86_abm
1.49 + #define TARGET_CMPXCHG16B x86_cmpxchg16b
1.50 + #define TARGET_POPCNT x86_popcnt
1.51 +@@ -2443,8 +2444,9 @@
1.52 + int save_varrargs_registers;
1.53 + int accesses_prev_frame;
1.54 + int optimize_mode_switching[MAX_386_ENTITIES];
1.55 +- /* Set by ix86_compute_frame_layout and used by prologue/epilogue expander to
1.56 +- determine the style used. */
1.57 ++ int needs_cld;
1.58 ++ /* Set by ix86_compute_frame_layout and used by prologue/epilogue
1.59 ++ expander to determine the style used. */
1.60 + int use_fast_prologue_epilogue;
1.61 + /* Number of saved registers USE_FAST_PROLOGUE_EPILOGUE has been computed
1.62 + for. */
1.63 +@@ -2464,6 +2466,7 @@
1.64 + #define ix86_stack_locals (cfun->machine->stack_locals)
1.65 + #define ix86_save_varrargs_registers (cfun->machine->save_varrargs_registers)
1.66 + #define ix86_optimize_mode_switching (cfun->machine->optimize_mode_switching)
1.67 ++#define ix86_current_function_needs_cld (cfun->machine->needs_cld)
1.68 + #define ix86_tls_descriptor_calls_expanded_in_cfun \
1.69 + (cfun->machine->tls_descriptor_call_expanded_p)
1.70 + /* Since tls_descriptor_call_expanded is not cleared, even if all TLS
1.71 +diff -durN gcc-4.3.0.orig/gcc/config/i386/i386.md gcc-4.3.0/gcc/config/i386/i386.md
1.72 +--- gcc-4.3.0.orig/gcc/config/i386/i386.md 2008-02-06 12:34:00.000000000 +0100
1.73 ++++ gcc-4.3.0/gcc/config/i386/i386.md 2008-06-10 14:44:37.000000000 +0200
1.74 +@@ -205,6 +205,7 @@
1.75 + (UNSPECV_XCHG 12)
1.76 + (UNSPECV_LOCK 13)
1.77 + (UNSPECV_PROLOGUE_USE 14)
1.78 ++ (UNSPECV_CLD 15)
1.79 + ])
1.80 +
1.81 + ;; Constants to represent pcomtrue/pcomfalse variants
1.82 +@@ -18529,6 +18530,14 @@
1.83 +
1.84 + ;; Block operation instructions
1.85 +
1.86 ++(define_insn "cld"
1.87 ++ [(unspec_volatile [(const_int 0)] UNSPECV_CLD)]
1.88 ++ ""
1.89 ++ "cld"
1.90 ++ [(set_attr "length" "1")
1.91 ++ (set_attr "length_immediate" "0")
1.92 ++ (set_attr "modrm" "0")])
1.93 ++
1.94 + (define_expand "movmemsi"
1.95 + [(use (match_operand:BLK 0 "memory_operand" ""))
1.96 + (use (match_operand:BLK 1 "memory_operand" ""))
1.97 +@@ -18601,7 +18610,7 @@
1.98 + (set (match_operand 2 "register_operand" "")
1.99 + (match_operand 5 "" ""))])]
1.100 + "TARGET_SINGLE_STRINGOP || optimize_size"
1.101 +- "")
1.102 ++ "ix86_current_function_needs_cld = 1;")
1.103 +
1.104 + (define_insn "*strmovdi_rex_1"
1.105 + [(set (mem:DI (match_operand:DI 2 "register_operand" "0"))
1.106 +@@ -18718,7 +18727,7 @@
1.107 + (match_operand 3 "memory_operand" ""))
1.108 + (use (match_dup 4))])]
1.109 + ""
1.110 +- "")
1.111 ++ "ix86_current_function_needs_cld = 1;")
1.112 +
1.113 + (define_insn "*rep_movdi_rex64"
1.114 + [(set (match_operand:DI 2 "register_operand" "=c") (const_int 0))
1.115 +@@ -18878,7 +18887,7 @@
1.116 + (set (match_operand 0 "register_operand" "")
1.117 + (match_operand 3 "" ""))])]
1.118 + "TARGET_SINGLE_STRINGOP || optimize_size"
1.119 +- "")
1.120 ++ "ix86_current_function_needs_cld = 1;")
1.121 +
1.122 + (define_insn "*strsetdi_rex_1"
1.123 + [(set (mem:DI (match_operand:DI 1 "register_operand" "0"))
1.124 +@@ -18972,7 +18981,7 @@
1.125 + (use (match_operand 3 "register_operand" ""))
1.126 + (use (match_dup 1))])]
1.127 + ""
1.128 +- "")
1.129 ++ "ix86_current_function_needs_cld = 1;")
1.130 +
1.131 + (define_insn "*rep_stosdi_rex64"
1.132 + [(set (match_operand:DI 1 "register_operand" "=c") (const_int 0))
1.133 +@@ -19148,7 +19157,7 @@
1.134 + (clobber (match_operand 1 "register_operand" ""))
1.135 + (clobber (match_dup 2))])]
1.136 + ""
1.137 +- "")
1.138 ++ "ix86_current_function_needs_cld = 1;")
1.139 +
1.140 + (define_insn "*cmpstrnqi_nz_1"
1.141 + [(set (reg:CC FLAGS_REG)
1.142 +@@ -19195,7 +19204,7 @@
1.143 + (clobber (match_operand 1 "register_operand" ""))
1.144 + (clobber (match_dup 2))])]
1.145 + ""
1.146 +- "")
1.147 ++ "ix86_current_function_needs_cld = 1;")
1.148 +
1.149 + (define_insn "*cmpstrnqi_1"
1.150 + [(set (reg:CC FLAGS_REG)
1.151 +@@ -19264,7 +19273,7 @@
1.152 + (clobber (match_operand 1 "register_operand" ""))
1.153 + (clobber (reg:CC FLAGS_REG))])]
1.154 + ""
1.155 +- "")
1.156 ++ "ix86_current_function_needs_cld = 1;")
1.157 +
1.158 + (define_insn "*strlenqi_1"
1.159 + [(set (match_operand:SI 0 "register_operand" "=&c")