43, /* cost of FSQRT instruction. */
};
+static const
+struct processor_costs nocona_cost = {
+ 1, /* cost of an add instruction */
+ 1, /* cost of a lea instruction */
+ 1, /* variable shift costs */
+ 1, /* constant shift costs */
+ {10, 10, 10, 10, 10}, /* cost of starting a multiply */
+ 0, /* cost of multiply per each bit set */
+ {66, 66, 66, 66, 66}, /* cost of a divide/mod */
+ 1, /* cost of movsx */
+ 1, /* cost of movzx */
+ 16, /* "large" insn */
+ 9, /* MOVE_RATIO */
+ 4, /* cost for loading QImode using movzbl */
+ {4, 4, 4}, /* cost of loading integer registers
+ in QImode, HImode and SImode.
+ Relative to reg-reg move (2). */
+ {4, 4, 4}, /* cost of storing integer registers */
+ 3, /* cost of reg,reg fld/fst */
+ {12, 12, 12}, /* cost of loading fp registers
+ in SFmode, DFmode and XFmode */
+ {4, 4, 4}, /* cost of loading integer registers */
+ 6, /* cost of moving MMX register */
+ {12, 12}, /* cost of loading MMX registers
+ in SImode and DImode */
+ {12, 12}, /* cost of storing MMX registers
+ in SImode and DImode */
+ 6, /* cost of moving SSE register */
+ {12, 12, 12}, /* cost of loading SSE registers
+ in SImode, DImode and TImode */
+ {12, 12, 12}, /* cost of storing SSE registers
+ in SImode, DImode and TImode */
+ 8, /* MMX or SSE register to integer */
+ 128, /* size of prefetch block */
+ 8, /* number of parallel prefetches */
+ 1, /* Branch cost */
+ 6, /* cost of FADD and FSUB insns. */
+ 8, /* cost of FMUL instruction. */
+ 40, /* cost of FDIV instruction. */
+ 3, /* cost of FABS instruction. */
+ 3, /* cost of FCHS instruction. */
+ 44, /* cost of FSQRT instruction. */
+};
+
const struct processor_costs *ix86_cost = &pentium_cost;
/* Processor feature/optimization bitmasks. */
#define m_PENT4 (1<<PROCESSOR_PENTIUM4)
#define m_K8 (1<<PROCESSOR_K8)
#define m_ATHLON_K8 (m_K8 | m_ATHLON)
+#define m_NOCONA (1<<PROCESSOR_NOCONA)
const int x86_use_leave = m_386 | m_K6 | m_ATHLON_K8;
-const int x86_push_memory = m_386 | m_K6 | m_ATHLON_K8 | m_PENT4;
+const int x86_push_memory = m_386 | m_K6 | m_ATHLON_K8 | m_PENT4 | m_NOCONA;
const int x86_zero_extend_with_and = m_486 | m_PENT;
-const int x86_movx = m_ATHLON_K8 | m_PPRO | m_PENT4 /* m_386 | m_K6 */;
+const int x86_movx = m_ATHLON_K8 | m_PPRO | m_PENT4 | m_NOCONA /* m_386 | m_K6 */;
const int x86_double_with_add = ~m_386;
const int x86_use_bit_test = m_386;
const int x86_unroll_strlen = m_486 | m_PENT | m_PPRO | m_ATHLON_K8 | m_K6;
-const int x86_cmove = m_PPRO | m_ATHLON_K8 | m_PENT4;
+const int x86_cmove = m_PPRO | m_ATHLON_K8 | m_PENT4 | m_NOCONA;
const int x86_3dnow_a = m_ATHLON_K8;
-const int x86_deep_branch = m_PPRO | m_K6 | m_ATHLON_K8 | m_PENT4;
-const int x86_branch_hints = m_PENT4;
-const int x86_use_sahf = m_PPRO | m_K6 | m_PENT4;
+const int x86_deep_branch = m_PPRO | m_K6 | m_ATHLON_K8 | m_PENT4 | m_NOCONA;
+const int x86_branch_hints = m_PENT4 | m_NOCONA;
+const int x86_use_sahf = m_PPRO | m_K6 | m_PENT4 | m_NOCONA;
const int x86_partial_reg_stall = m_PPRO;
const int x86_use_loop = m_K6;
const int x86_use_fiop = ~(m_PPRO | m_ATHLON_K8 | m_PENT);
const int x86_split_long_moves = m_PPRO;
const int x86_promote_QImode = m_K6 | m_PENT | m_386 | m_486 | m_ATHLON_K8;
const int x86_fast_prefix = ~(m_PENT | m_486 | m_386);
-const int x86_single_stringop = m_386 | m_PENT4;
+const int x86_single_stringop = m_386 | m_PENT4 | m_NOCONA;
const int x86_qimode_math = ~(0);
const int x86_promote_qi_regs = 0;
const int x86_himode_math = ~(m_PPRO);
const int x86_promote_hi_regs = m_PPRO;
-const int x86_sub_esp_4 = m_ATHLON_K8 | m_PPRO | m_PENT4;
-const int x86_sub_esp_8 = m_ATHLON_K8 | m_PPRO | m_386 | m_486 | m_PENT4;
-const int x86_add_esp_4 = m_ATHLON_K8 | m_K6 | m_PENT4;
-const int x86_add_esp_8 = m_ATHLON_K8 | m_PPRO | m_K6 | m_386 | m_486 | m_PENT4;
-const int x86_integer_DFmode_moves = ~(m_ATHLON_K8 | m_PENT4 | m_PPRO);
-const int x86_partial_reg_dependency = m_ATHLON_K8 | m_PENT4;
-const int x86_memory_mismatch_stall = m_ATHLON_K8 | m_PENT4;
-const int x86_accumulate_outgoing_args = m_ATHLON_K8 | m_PENT4 | m_PPRO;
+const int x86_sub_esp_4 = m_ATHLON_K8 | m_PPRO | m_PENT4 | m_NOCONA;
+const int x86_sub_esp_8 = m_ATHLON_K8 | m_PPRO | m_386 | m_486 | m_PENT4 | m_NOCONA;
+const int x86_add_esp_4 = m_ATHLON_K8 | m_K6 | m_PENT4 | m_NOCONA;
+const int x86_add_esp_8 = m_ATHLON_K8 | m_PPRO | m_K6 | m_386 | m_486 | m_PENT4 | m_NOCONA;
+const int x86_integer_DFmode_moves = ~(m_ATHLON_K8 | m_PENT4 | m_NOCONA | m_PPRO);
+const int x86_partial_reg_dependency = m_ATHLON_K8 | m_PENT4 | m_NOCONA;
+const int x86_memory_mismatch_stall = m_ATHLON_K8 | m_PENT4 | m_NOCONA;
+const int x86_accumulate_outgoing_args = m_ATHLON_K8 | m_PENT4 | m_NOCONA | m_PPRO;
const int x86_prologue_using_move = m_ATHLON_K8 | m_PPRO;
const int x86_epilogue_using_move = m_ATHLON_K8 | m_PPRO;
-const int x86_decompose_lea = m_PENT4;
+const int x86_decompose_lea = m_PENT4 | m_NOCONA;
const int x86_shift1 = ~m_486;
-const int x86_arch_always_fancy_math_387 = m_PENT | m_PPRO | m_ATHLON_K8 | m_PENT4;
-const int x86_sse_partial_reg_dependency = m_PENT4 | m_PPRO;
+const int x86_arch_always_fancy_math_387 = m_PENT | m_PPRO | m_ATHLON_K8 | m_PENT4 | m_NOCONA;
+const int x86_sse_partial_reg_dependency = m_PENT4 | m_NOCONA | m_PPRO;
/* Set for machines where the type and dependencies are resolved on SSE register
parts instead of whole registers, so we may maintain just lower part of
scalar values in proper format leaving the upper part undefined. */
need for extra instructions beforehand */
const int x86_sse_partial_regs_for_cvtsd2ss = 0;
const int x86_sse_typeless_stores = m_ATHLON_K8;
-const int x86_sse_load0_by_pxor = m_PPRO | m_PENT4;
+const int x86_sse_load0_by_pxor = m_PPRO | m_PENT4 | m_NOCONA;
const int x86_use_ffreep = m_ATHLON_K8;
const int x86_rep_movl_optimal = m_386 | m_PENT | m_PPRO | m_K6;
const int x86_inter_unit_moves = ~(m_ATHLON_K8);
-const int x86_ext_80387_constants = m_K6 | m_ATHLON | m_PENT4 | m_PPRO;
+const int x86_ext_80387_constants = m_K6 | m_ATHLON | m_PENT4 | m_NOCONA | m_PPRO;
/* Some CPU cores are not able to predict more than 4 branch instructions in
the 16 byte window. */
-const int x86_four_jump_limit = m_PPRO | m_ATHLON_K8 | m_PENT4;
+const int x86_four_jump_limit = m_PPRO | m_ATHLON_K8 | m_PENT4 | m_NOCONA;
/* In case the average insn count for single function invocation is
lower than this constant, emit fast (but longer) prologue and
static int memory_address_length (rtx addr);
static int ix86_flags_dependant (rtx, rtx, enum attr_type);
static int ix86_agi_dependant (rtx, rtx, enum attr_type);
-static enum attr_ppro_uops ix86_safe_ppro_uops (rtx);
-static void ix86_dump_ppro_packet (FILE *);
-static void ix86_reorder_insn (rtx *, rtx *);
static struct machine_function * ix86_init_machine_status (void);
static int ix86_split_to_parts (rtx, rtx *, enum machine_mode);
static int ix86_nsaved_regs (void);
static void ix86_emit_save_regs_using_mov (rtx, HOST_WIDE_INT);
static void ix86_emit_restore_regs_using_mov (rtx, HOST_WIDE_INT, int);
static void ix86_output_function_epilogue (FILE *, HOST_WIDE_INT);
-static void ix86_sched_reorder_ppro (rtx *, rtx *);
static HOST_WIDE_INT ix86_GOT_alias_set (void);
static void ix86_adjust_counter (rtx, HOST_WIDE_INT);
static rtx ix86_expand_aligntest (rtx, int);
static void ix86_expand_strlensi_unroll_1 (rtx, rtx, rtx);
static int ix86_issue_rate (void);
static int ix86_adjust_cost (rtx, rtx, rtx, int);
-static void ix86_sched_init (FILE *, int, int);
-static int ix86_sched_reorder (FILE *, int, rtx *, int *, int);
-static int ix86_variable_issue (FILE *, int, rtx, int);
static int ia32_use_dfa_pipeline_interface (void);
static int ia32_multipass_dfa_lookahead (void);
static void ix86_init_mmx_sse_builtins (void);
static int extended_reg_mentioned_1 (rtx *, void *);
static bool ix86_rtx_costs (rtx, int, int, int *);
static int min_insn_size (rtx);
+static tree ix86_md_asm_clobbers (tree clobbers);
#if defined (DO_GLOBAL_CTORS_BODY) && defined (HAS_INIT_SECTION)
static void ix86_svr3_asm_out_constructor (rtx, int);
#define TARGET_SCHED_ADJUST_COST ix86_adjust_cost
#undef TARGET_SCHED_ISSUE_RATE
#define TARGET_SCHED_ISSUE_RATE ix86_issue_rate
-#undef TARGET_SCHED_VARIABLE_ISSUE
-#define TARGET_SCHED_VARIABLE_ISSUE ix86_variable_issue
-#undef TARGET_SCHED_INIT
-#define TARGET_SCHED_INIT ix86_sched_init
-#undef TARGET_SCHED_REORDER
-#define TARGET_SCHED_REORDER ix86_sched_reorder
#undef TARGET_SCHED_USE_DFA_PIPELINE_INTERFACE
#define TARGET_SCHED_USE_DFA_PIPELINE_INTERFACE \
ia32_use_dfa_pipeline_interface
#undef TARGET_BUILD_BUILTIN_VA_LIST
#define TARGET_BUILD_BUILTIN_VA_LIST ix86_build_builtin_va_list
+#undef TARGET_MD_ASM_CLOBBERS
+#define TARGET_MD_ASM_CLOBBERS ix86_md_asm_clobbers
+
#undef TARGET_PROMOTE_PROTOTYPES
#define TARGET_PROMOTE_PROTOTYPES hook_bool_tree_true
#define TARGET_SETUP_INCOMING_VARARGS ix86_setup_incoming_varargs
struct gcc_target targetm = TARGET_INITIALIZER;
+
\f
/* The svr4 ABI for the i386 says that records and unions are returned
in memory. */
{&k6_cost, 0, 0, 32, 7, 32, 7, 32},
{&athlon_cost, 0, 0, 16, 7, 16, 7, 16},
{&pentium4_cost, 0, 0, 0, 0, 0, 0, 0},
- {&k8_cost, 0, 0, 16, 7, 16, 7, 16}
+ {&k8_cost, 0, 0, 16, 7, 16, 7, 16},
+ {&nocona_cost, 0, 0, 0, 0, 0, 0, 0}
};
static const char * const cpu_names[] = TARGET_CPU_DEFAULT_NAMES;
{
PTA_SSE = 1,
PTA_SSE2 = 2,
- PTA_MMX = 4,
- PTA_PREFETCH_SSE = 8,
- PTA_3DNOW = 16,
+ PTA_SSE3 = 4,
+ PTA_MMX = 8,
+ PTA_PREFETCH_SSE = 16,
+ PTA_3DNOW = 32,
PTA_3DNOW_A = 64,
PTA_64BIT = 128
} flags;
{"pentiumpro", PROCESSOR_PENTIUMPRO, 0},
{"pentium2", PROCESSOR_PENTIUMPRO, PTA_MMX},
{"pentium3", PROCESSOR_PENTIUMPRO, PTA_MMX | PTA_SSE | PTA_PREFETCH_SSE},
- {"pentium4", PROCESSOR_PENTIUM4, PTA_SSE | PTA_SSE2 |
- PTA_MMX | PTA_PREFETCH_SSE},
+ {"pentium3m", PROCESSOR_PENTIUMPRO, PTA_MMX | PTA_SSE | PTA_PREFETCH_SSE},
+ {"pentium-m", PROCESSOR_PENTIUMPRO, PTA_MMX | PTA_SSE | PTA_PREFETCH_SSE | PTA_SSE2},
+ {"pentium4", PROCESSOR_PENTIUM4, PTA_SSE | PTA_SSE2
+ | PTA_MMX | PTA_PREFETCH_SSE},
+ {"pentium4m", PROCESSOR_PENTIUM4, PTA_SSE | PTA_SSE2
+ | PTA_MMX | PTA_PREFETCH_SSE},
+ {"prescott", PROCESSOR_NOCONA, PTA_SSE | PTA_SSE2 | PTA_SSE3
+ | PTA_MMX | PTA_PREFETCH_SSE},
+ {"nocona", PROCESSOR_NOCONA, PTA_SSE | PTA_SSE2 | PTA_SSE3 | PTA_64BIT
+ | PTA_MMX | PTA_PREFETCH_SSE},
{"k6", PROCESSOR_K6, PTA_MMX},
{"k6-2", PROCESSOR_K6, PTA_MMX | PTA_3DNOW},
{"k6-3", PROCESSOR_K6, PTA_MMX | PTA_3DNOW},
if (processor_alias_table[i].flags & PTA_SSE2
&& !(target_flags_explicit & MASK_SSE2))
target_flags |= MASK_SSE2;
+ if (processor_alias_table[i].flags & PTA_SSE3
+ && !(target_flags_explicit & MASK_SSE3))
+ target_flags |= MASK_SSE3;
if (processor_alias_table[i].flags & PTA_PREFETCH_SSE)
x86_prefetch_sse = true;
if (TARGET_64BIT && !(processor_alias_table[i].flags & PTA_64BIT))
if (x86_arch_always_fancy_math_387 & (1 << ix86_arch))
target_flags &= ~MASK_NO_FANCY_MATH_387;
- /* Turn on SSE2 builtins for -mpni. */
- if (TARGET_PNI)
+ /* Turn on SSE2 builtins for -msse3. */
+ if (TARGET_SSE3)
target_flags |= MASK_SSE2;
/* Turn on SSE builtins for -msse2. */
if (!lookup_attribute (rtdstr, TYPE_ATTRIBUTES (type1))
!= !lookup_attribute (rtdstr, TYPE_ATTRIBUTES (type2)))
return 0;
+ if (ix86_function_regparm (type1, NULL)
+ != ix86_function_regparm (type2, NULL))
+ return 0;
return 1;
}
\f
{
/* On Pentium4, the inc and dec operations causes extra dependency on flag
registers, since carry flag is not set. */
- if (TARGET_PENTIUM4 && !optimize_size)
+ if ((TARGET_PENTIUM4 || TARGET_NOCONA) && !optimize_size)
return 0;
return op == const1_rtx || op == constm1_rtx;
}
enum rtx_code code = GET_CODE (op);
if (mode != VOIDmode && GET_MODE (op) != mode)
return 0;
- if (GET_RTX_CLASS (code) != '<')
+ if (!COMPARISON_P (op))
return 0;
inmode = GET_MODE (XEXP (op, 0));
if (mode != VOIDmode && GET_MODE (op) != mode)
return 0;
- if (GET_RTX_CLASS (code) != '<')
+ if (!COMPARISON_P (op))
return 0;
inmode = GET_MODE (XEXP (op, 0));
if (GET_CODE (XEXP (op, 0)) != REG
if (mode != VOIDmode && GET_MODE (op) != mode)
return 0;
- if (GET_RTX_CLASS (code) != '<')
+ if (!COMPARISON_P (op))
return 0;
inmode = GET_MODE (XEXP (op, 0));
if (inmode == CCFPmode || inmode == CCFPUmode)
arith_or_logical_operator (rtx op, enum machine_mode mode)
{
return ((mode == VOIDmode || GET_MODE (op) == mode)
- && (GET_RTX_CLASS (GET_CODE (op)) == 'c'
- || GET_RTX_CLASS (GET_CODE (op)) == '2'));
+ && ARITHMETIC_P (op));
}
/* Returns 1 if OP is memory operand with a displacement. */
case 'c':
/* Check to see if argument to %c is really a constant
and not a condition code which needs to be reversed. */
- if (GET_RTX_CLASS (GET_CODE (x)) != '<')
+ if (!COMPARISON_P (x))
{
output_operand_lossage ("operand is neither a constant nor a condition code, invalid operand code 'c'");
return;
src2 = operands[2];
/* Recognize <var1> = <value> <op> <var1> for commutative operators */
- if (GET_RTX_CLASS (code) == 'c'
+ if (GET_RTX_CLASS (code) == RTX_COMM_ARITH
&& (rtx_equal_p (dst, src2)
|| immediate_operand (src1, mode)))
{
{
if (rtx_equal_p (dst, src1))
matching_memory = 1;
- else if (GET_RTX_CLASS (code) == 'c'
+ else if (GET_RTX_CLASS (code) == RTX_COMM_ARITH
&& rtx_equal_p (dst, src2))
matching_memory = 2;
else
or non-matching memory. */
if ((CONSTANT_P (src1)
|| (!matching_memory && GET_CODE (src1) == MEM))
- && GET_RTX_CLASS (code) != 'c')
+ && GET_RTX_CLASS (code) != RTX_COMM_ARITH)
src1 = force_reg (mode, src1);
/* If optimizing, copy to regs to improve CSE */
if (GET_CODE (operands[1]) == MEM && GET_CODE (operands[2]) == MEM)
return 0;
/* If the operation is not commutable, source 1 cannot be a constant. */
- if (CONSTANT_P (operands[1]) && GET_RTX_CLASS (code) != 'c')
+ if (CONSTANT_P (operands[1]) && GET_RTX_CLASS (code) != RTX_COMM_ARITH)
return 0;
/* If the destination is memory, we must have a matching source operand. */
if (GET_CODE (operands[0]) == MEM
&& ! (rtx_equal_p (operands[0], operands[1])
- || (GET_RTX_CLASS (code) == 'c'
+ || (GET_RTX_CLASS (code) == RTX_COMM_ARITH
&& rtx_equal_p (operands[0], operands[2]))))
return 0;
/* If the operation is not commutable and the source 1 is memory, we must
have a matching destination. */
if (GET_CODE (operands[1]) == MEM
- && GET_RTX_CLASS (code) != 'c'
+ && GET_RTX_CLASS (code) != RTX_COMM_ARITH
&& ! rtx_equal_p (operands[0], operands[1]))
return 0;
return 1;
long l[3];
REAL_VALUE_FROM_CONST_DOUBLE (r, operand);
- REAL_VALUE_TO_TARGET_LONG_DOUBLE (r, l);
+ real_to_target (l, &r, mode);
/* Do not use shift by 32 to avoid warning on 32bit systems. */
if (HOST_BITS_PER_WIDE_INT >= 64)
parts[0]
split_di (operands, 2, low, high);
count = INTVAL (operands[2]) & 63;
- if (count >= 32)
+ if (count == 63)
+ {
+ emit_move_insn (high[0], high[1]);
+ emit_insn (gen_ashrsi3 (high[0], high[0], GEN_INT (31)));
+ emit_move_insn (low[0], high[0]);
+
+ }
+ else if (count >= 32)
{
emit_move_insn (low[0], high[1]);
case PROCESSOR_PENTIUM4:
case PROCESSOR_ATHLON:
case PROCESSOR_K8:
+ case PROCESSOR_NOCONA:
return 3;
default:
return cost;
}
-static union
-{
- struct ppro_sched_data
- {
- rtx decode[3];
- int issued_this_cycle;
- } ppro;
-} ix86_sched_data;
-
-static enum attr_ppro_uops
-ix86_safe_ppro_uops (rtx insn)
-{
- if (recog_memoized (insn) >= 0)
- return get_attr_ppro_uops (insn);
- else
- return PPRO_UOPS_MANY;
-}
-
-static void
-ix86_dump_ppro_packet (FILE *dump)
-{
- if (ix86_sched_data.ppro.decode[0])
- {
- fprintf (dump, "PPRO packet: %d",
- INSN_UID (ix86_sched_data.ppro.decode[0]));
- if (ix86_sched_data.ppro.decode[1])
- fprintf (dump, " %d", INSN_UID (ix86_sched_data.ppro.decode[1]));
- if (ix86_sched_data.ppro.decode[2])
- fprintf (dump, " %d", INSN_UID (ix86_sched_data.ppro.decode[2]));
- fputc ('\n', dump);
- }
-}
-
-/* We're beginning a new block. Initialize data structures as necessary. */
-
-static void
-ix86_sched_init (FILE *dump ATTRIBUTE_UNUSED,
- int sched_verbose ATTRIBUTE_UNUSED,
- int veclen ATTRIBUTE_UNUSED)
-{
- memset (&ix86_sched_data, 0, sizeof (ix86_sched_data));
-}
-
-/* Shift INSN to SLOT, and shift everything else down. */
-
-static void
-ix86_reorder_insn (rtx *insnp, rtx *slot)
-{
- if (insnp != slot)
- {
- rtx insn = *insnp;
- do
- insnp[0] = insnp[1];
- while (++insnp != slot);
- *insnp = insn;
- }
-}
-
-static void
-ix86_sched_reorder_ppro (rtx *ready, rtx *e_ready)
-{
- rtx decode[3];
- enum attr_ppro_uops cur_uops;
- int issued_this_cycle;
- rtx *insnp;
- int i;
-
- /* At this point .ppro.decode contains the state of the three
- decoders from last "cycle". That is, those insns that were
- actually independent. But here we're scheduling for the
- decoder, and we may find things that are decodable in the
- same cycle. */
-
- memcpy (decode, ix86_sched_data.ppro.decode, sizeof (decode));
- issued_this_cycle = 0;
-
- insnp = e_ready;
- cur_uops = ix86_safe_ppro_uops (*insnp);
-
- /* If the decoders are empty, and we've a complex insn at the
- head of the priority queue, let it issue without complaint. */
- if (decode[0] == NULL)
- {
- if (cur_uops == PPRO_UOPS_MANY)
- {
- decode[0] = *insnp;
- goto ppro_done;
- }
-
- /* Otherwise, search for a 2-4 uop unsn to issue. */
- while (cur_uops != PPRO_UOPS_FEW)
- {
- if (insnp == ready)
- break;
- cur_uops = ix86_safe_ppro_uops (*--insnp);
- }
-
- /* If so, move it to the head of the line. */
- if (cur_uops == PPRO_UOPS_FEW)
- ix86_reorder_insn (insnp, e_ready);
-
- /* Issue the head of the queue. */
- issued_this_cycle = 1;
- decode[0] = *e_ready--;
- }
-
- /* Look for simple insns to fill in the other two slots. */
- for (i = 1; i < 3; ++i)
- if (decode[i] == NULL)
- {
- if (ready > e_ready)
- goto ppro_done;
-
- insnp = e_ready;
- cur_uops = ix86_safe_ppro_uops (*insnp);
- while (cur_uops != PPRO_UOPS_ONE)
- {
- if (insnp == ready)
- break;
- cur_uops = ix86_safe_ppro_uops (*--insnp);
- }
-
- /* Found one. Move it to the head of the queue and issue it. */
- if (cur_uops == PPRO_UOPS_ONE)
- {
- ix86_reorder_insn (insnp, e_ready);
- decode[i] = *e_ready--;
- issued_this_cycle++;
- continue;
- }
-
- /* ??? Didn't find one. Ideally, here we would do a lazy split
- of 2-uop insns, issue one and queue the other. */
- }
-
- ppro_done:
- if (issued_this_cycle == 0)
- issued_this_cycle = 1;
- ix86_sched_data.ppro.issued_this_cycle = issued_this_cycle;
-}
-
-/* We are about to being issuing insns for this clock cycle.
- Override the default sort algorithm to better slot instructions. */
-static int
-ix86_sched_reorder (FILE *dump ATTRIBUTE_UNUSED,
- int sched_verbose ATTRIBUTE_UNUSED, rtx *ready,
- int *n_readyp, int clock_var ATTRIBUTE_UNUSED)
-{
- int n_ready = *n_readyp;
- rtx *e_ready = ready + n_ready - 1;
-
- /* Make sure to go ahead and initialize key items in
- ix86_sched_data if we are not going to bother trying to
- reorder the ready queue. */
- if (n_ready < 2)
- {
- ix86_sched_data.ppro.issued_this_cycle = 1;
- goto out;
- }
-
- switch (ix86_tune)
- {
- default:
- break;
-
- case PROCESSOR_PENTIUMPRO:
- ix86_sched_reorder_ppro (ready, e_ready);
- break;
- }
-
-out:
- return ix86_issue_rate ();
-}
-
-/* We are about to issue INSN. Return the number of insns left on the
- ready queue that can be issued this cycle. */
-
-static int
-ix86_variable_issue (FILE *dump, int sched_verbose, rtx insn,
- int can_issue_more)
-{
- int i;
- switch (ix86_tune)
- {
- default:
- return can_issue_more - 1;
-
- case PROCESSOR_PENTIUMPRO:
- {
- enum attr_ppro_uops uops = ix86_safe_ppro_uops (insn);
-
- if (uops == PPRO_UOPS_MANY)
- {
- if (sched_verbose)
- ix86_dump_ppro_packet (dump);
- ix86_sched_data.ppro.decode[0] = insn;
- ix86_sched_data.ppro.decode[1] = NULL;
- ix86_sched_data.ppro.decode[2] = NULL;
- if (sched_verbose)
- ix86_dump_ppro_packet (dump);
- ix86_sched_data.ppro.decode[0] = NULL;
- }
- else if (uops == PPRO_UOPS_FEW)
- {
- if (sched_verbose)
- ix86_dump_ppro_packet (dump);
- ix86_sched_data.ppro.decode[0] = insn;
- ix86_sched_data.ppro.decode[1] = NULL;
- ix86_sched_data.ppro.decode[2] = NULL;
- }
- else
- {
- for (i = 0; i < 3; ++i)
- if (ix86_sched_data.ppro.decode[i] == NULL)
- {
- ix86_sched_data.ppro.decode[i] = insn;
- break;
- }
- if (i == 3)
- abort ();
- if (i == 2)
- {
- if (sched_verbose)
- ix86_dump_ppro_packet (dump);
- ix86_sched_data.ppro.decode[0] = NULL;
- ix86_sched_data.ppro.decode[1] = NULL;
- ix86_sched_data.ppro.decode[2] = NULL;
- }
- }
- }
- return --ix86_sched_data.ppro.issued_this_cycle;
- }
-}
-
static int
ia32_use_dfa_pipeline_interface (void)
{
- if (TARGET_PENTIUM || TARGET_ATHLON_K8)
+ if (TARGET_PENTIUM
+ || TARGET_PENTIUMPRO
+ || TARGET_ATHLON_K8)
return 1;
return 0;
}
{
if (ix86_tune == PROCESSOR_PENTIUM)
return 2;
+
+ if (ix86_tune == PROCESSOR_PENTIUMPRO)
+ return 1;
+
else
- return 0;
+ return 0;
}
\f
{ MASK_SSE2, CODE_FOR_cvtsd2ss, 0, IX86_BUILTIN_CVTSD2SS, 0, 0 },
{ MASK_SSE2, CODE_FOR_cvtss2sd, 0, IX86_BUILTIN_CVTSS2SD, 0, 0 },
- /* PNI MMX */
- { MASK_PNI, CODE_FOR_addsubv4sf3, "__builtin_ia32_addsubps", IX86_BUILTIN_ADDSUBPS, 0, 0 },
- { MASK_PNI, CODE_FOR_addsubv2df3, "__builtin_ia32_addsubpd", IX86_BUILTIN_ADDSUBPD, 0, 0 },
- { MASK_PNI, CODE_FOR_haddv4sf3, "__builtin_ia32_haddps", IX86_BUILTIN_HADDPS, 0, 0 },
- { MASK_PNI, CODE_FOR_haddv2df3, "__builtin_ia32_haddpd", IX86_BUILTIN_HADDPD, 0, 0 },
- { MASK_PNI, CODE_FOR_hsubv4sf3, "__builtin_ia32_hsubps", IX86_BUILTIN_HSUBPS, 0, 0 },
- { MASK_PNI, CODE_FOR_hsubv2df3, "__builtin_ia32_hsubpd", IX86_BUILTIN_HSUBPD, 0, 0 }
+ /* SSE3 MMX */
+ { MASK_SSE3, CODE_FOR_addsubv4sf3, "__builtin_ia32_addsubps", IX86_BUILTIN_ADDSUBPS, 0, 0 },
+ { MASK_SSE3, CODE_FOR_addsubv2df3, "__builtin_ia32_addsubpd", IX86_BUILTIN_ADDSUBPD, 0, 0 },
+ { MASK_SSE3, CODE_FOR_haddv4sf3, "__builtin_ia32_haddps", IX86_BUILTIN_HADDPS, 0, 0 },
+ { MASK_SSE3, CODE_FOR_haddv2df3, "__builtin_ia32_haddpd", IX86_BUILTIN_HADDPD, 0, 0 },
+ { MASK_SSE3, CODE_FOR_hsubv4sf3, "__builtin_ia32_hsubps", IX86_BUILTIN_HSUBPS, 0, 0 },
+ { MASK_SSE3, CODE_FOR_hsubv2df3, "__builtin_ia32_hsubpd", IX86_BUILTIN_HSUBPD, 0, 0 }
};
static const struct builtin_description bdesc_1arg[] =
{ MASK_SSE2, CODE_FOR_sse2_movq, 0, IX86_BUILTIN_MOVQ, 0, 0 },
- /* PNI */
- { MASK_PNI, CODE_FOR_movshdup, 0, IX86_BUILTIN_MOVSHDUP, 0, 0 },
- { MASK_PNI, CODE_FOR_movsldup, 0, IX86_BUILTIN_MOVSLDUP, 0, 0 },
- { MASK_PNI, CODE_FOR_movddup, 0, IX86_BUILTIN_MOVDDUP, 0, 0 }
+ /* SSE3 */
+ { MASK_SSE3, CODE_FOR_movshdup, 0, IX86_BUILTIN_MOVSHDUP, 0, 0 },
+ { MASK_SSE3, CODE_FOR_movsldup, 0, IX86_BUILTIN_MOVSLDUP, 0, 0 },
+ { MASK_SSE3, CODE_FOR_movddup, 0, IX86_BUILTIN_MOVDDUP, 0, 0 }
};
void
const struct builtin_description * d;
size_t i;
+ tree V16QI_type_node = build_vector_type_for_mode (intQI_type_node, V16QImode);
+ tree V2SI_type_node = build_vector_type_for_mode (intSI_type_node, V2SImode);
+ tree V2SF_type_node = build_vector_type_for_mode (float_type_node, V2SFmode);
+ tree V2DI_type_node = build_vector_type_for_mode (intDI_type_node, V2DImode);
+ tree V2DF_type_node = build_vector_type_for_mode (double_type_node, V2DFmode);
+ tree V4SF_type_node = build_vector_type_for_mode (float_type_node, V4SFmode);
+ tree V4SI_type_node = build_vector_type_for_mode (intSI_type_node, V4SImode);
+ tree V4HI_type_node = build_vector_type_for_mode (intHI_type_node, V4HImode);
+ tree V8QI_type_node = build_vector_type_for_mode (intQI_type_node, V8QImode);
+ tree V8HI_type_node = build_vector_type_for_mode (intHI_type_node, V8HImode);
+
tree pchar_type_node = build_pointer_type (char_type_node);
tree pcchar_type_node = build_pointer_type (
build_type_variant (char_type_node, 1, 0));
def_builtin (MASK_SSE2, "__builtin_ia32_pmaddwd128", v4si_ftype_v8hi_v8hi, IX86_BUILTIN_PMADDWD128);
/* Prescott New Instructions. */
- def_builtin (MASK_PNI, "__builtin_ia32_monitor",
+ def_builtin (MASK_SSE3, "__builtin_ia32_monitor",
void_ftype_pcvoid_unsigned_unsigned,
IX86_BUILTIN_MONITOR);
- def_builtin (MASK_PNI, "__builtin_ia32_mwait",
+ def_builtin (MASK_SSE3, "__builtin_ia32_mwait",
void_ftype_unsigned_unsigned,
IX86_BUILTIN_MWAIT);
- def_builtin (MASK_PNI, "__builtin_ia32_movshdup",
+ def_builtin (MASK_SSE3, "__builtin_ia32_movshdup",
v4sf_ftype_v4sf,
IX86_BUILTIN_MOVSHDUP);
- def_builtin (MASK_PNI, "__builtin_ia32_movsldup",
+ def_builtin (MASK_SSE3, "__builtin_ia32_movsldup",
v4sf_ftype_v4sf,
IX86_BUILTIN_MOVSLDUP);
- def_builtin (MASK_PNI, "__builtin_ia32_lddqu",
+ def_builtin (MASK_SSE3, "__builtin_ia32_lddqu",
v16qi_ftype_pcchar, IX86_BUILTIN_LDDQU);
- def_builtin (MASK_PNI, "__builtin_ia32_loadddup",
+ def_builtin (MASK_SSE3, "__builtin_ia32_loadddup",
v2df_ftype_pcdouble, IX86_BUILTIN_LOADDDUP);
- def_builtin (MASK_PNI, "__builtin_ia32_movddup",
+ def_builtin (MASK_SSE3, "__builtin_ia32_movddup",
v2df_ftype_v2df, IX86_BUILTIN_MOVDDUP);
}
case MULT:
if (FLOAT_MODE_P (mode))
- *total = COSTS_N_INSNS (ix86_cost->fmul);
- else if (GET_CODE (XEXP (x, 1)) == CONST_INT)
{
- unsigned HOST_WIDE_INT value = INTVAL (XEXP (x, 1));
- int nbits;
-
- for (nbits = 0; value != 0; value >>= 1)
- nbits++;
-
- *total = COSTS_N_INSNS (ix86_cost->mult_init[MODE_INDEX (mode)]
- + nbits * ix86_cost->mult_bit);
+ *total = COSTS_N_INSNS (ix86_cost->fmul);
+ return false;
}
else
{
- /* This is arbitrary */
- *total = COSTS_N_INSNS (ix86_cost->mult_init[MODE_INDEX (mode)]
- + 7 * ix86_cost->mult_bit);
+ rtx op0 = XEXP (x, 0);
+ rtx op1 = XEXP (x, 1);
+ int nbits;
+ if (GET_CODE (XEXP (x, 1)) == CONST_INT)
+ {
+ unsigned HOST_WIDE_INT value = INTVAL (XEXP (x, 1));
+ for (nbits = 0; value != 0; value &= value - 1)
+ nbits++;
+ }
+ else
+ /* This is arbitrary. */
+ nbits = 7;
+
+ /* Compute costs correctly for widening multiplication. */
+ if ((GET_CODE (op0) == SIGN_EXTEND || GET_CODE (op1) == ZERO_EXTEND)
+ && GET_MODE_SIZE (GET_MODE (XEXP (op0, 0))) * 2
+ == GET_MODE_SIZE (mode))
+ {
+ int is_mulwiden = 0;
+ enum machine_mode inner_mode = GET_MODE (op0);
+
+ if (GET_CODE (op0) == GET_CODE (op1))
+ is_mulwiden = 1, op1 = XEXP (op1, 0);
+ else if (GET_CODE (op1) == CONST_INT)
+ {
+ if (GET_CODE (op0) == SIGN_EXTEND)
+ is_mulwiden = trunc_int_for_mode (INTVAL (op1), inner_mode)
+ == INTVAL (op1);
+ else
+ is_mulwiden = !(INTVAL (op1) & ~GET_MODE_MASK (inner_mode));
+ }
+
+ if (is_mulwiden)
+ op0 = XEXP (op0, 0), mode = GET_MODE (op0);
+ }
+
+ *total = COSTS_N_INSNS (ix86_cost->mult_init[MODE_INDEX (mode)]
+ + nbits * ix86_cost->mult_bit)
+ + rtx_cost (op0, outer_code) + rtx_cost (op1, outer_code);
+
+ return true;
}
- return false;
case DIV:
case UDIV:
{
nbytes += min_insn_size (insn);
- if (rtl_dump_file)
- fprintf(rtl_dump_file, "Insn %i estimated to %i bytes\n",
+ if (dump_file)
+ fprintf(dump_file, "Insn %i estimated to %i bytes\n",
INSN_UID (insn), min_insn_size (insn));
if ((GET_CODE (insn) == JUMP_INSN
&& GET_CODE (PATTERN (insn)) != ADDR_VEC
}
if (njumps < 0)
abort ();
- if (rtl_dump_file)
- fprintf(rtl_dump_file, "Interval %i to %i has %i bytes\n",
+ if (dump_file)
+ fprintf (dump_file, "Interval %i to %i has %i bytes\n",
INSN_UID (start), INSN_UID (insn), nbytes);
if (njumps == 3 && isjump && nbytes < 16)
{
int padsize = 15 - nbytes + min_insn_size (insn);
- if (rtl_dump_file)
- fprintf (rtl_dump_file, "Padding insn %i by %i bytes!\n", INSN_UID (insn), padsize);
+ if (dump_file)
+ fprintf (dump_file, "Padding insn %i by %i bytes!\n",
+ INSN_UID (insn), padsize);
emit_insn_before (gen_align (GEN_INT (padsize)), insn);
}
}
}
/* ... values where only first field is non-constant are best loaded
- from the pool and overwriten via move later. */
+ from the pool and overwritten via move later. */
if (!i)
{
rtx op = simplify_gen_subreg (mode, XVECEXP (vals, 0, 0),
}
}
+/* Worker function for TARGET_MD_ASM_CLOBBERS.
+
+ We do this in the new i386 backend to maintain source compatibility
+ with the old cc0-based compiler. */
+
+static tree
+ix86_md_asm_clobbers (tree clobbers)
+{
+ clobbers = tree_cons (NULL_TREE, build_string (5, "flags"),
+ clobbers);
+ clobbers = tree_cons (NULL_TREE, build_string (4, "fpsr"),
+ clobbers);
+ clobbers = tree_cons (NULL_TREE, build_string (7, "dirflag"),
+ clobbers);
+ return clobbers;
+}
+
+/* Worker function for REVERSE_CONDITION. */
+
+enum rtx_code
+ix86_reverse_condition (enum rtx_code code, enum machine_mode mode)
+{
+ return (mode != CCFPmode && mode != CCFPUmode
+ ? reverse_condition (code)
+ : reverse_condition_maybe_unordered (code));
+}
+
+/* Output code to perform an x87 FP register move, from OPERANDS[1]
+ to OPERANDS[0]. */
+
+const char *
+output_387_reg_move (rtx insn, rtx *operands)
+{
+ if (REG_P (operands[1])
+ && find_regno_note (insn, REG_DEAD, REGNO (operands[1])))
+ {
+ if (REGNO (operands[0]) == FIRST_STACK_REG
+ && TARGET_USE_FFREEP)
+ return "ffreep\t%y0";
+ return "fstp\t%y0";
+ }
+ if (STACK_TOP_P (operands[0]))
+ return "fld%z1\t%y1";
+ return "fst\t%y0";
+}
+
#include "gt-i386.h"
OSDN Git Service
