This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[s390] Split out pre-prologue rewrite into separate pass
- From: Richard Sandiford <rsandifo at linux dot vnet dot ibm dot com>
- To: gcc-patches at gcc dot gnu dot org
- Cc: Andreas dot Krebbel at de dot ibm dot com, uweigand at de dot ibm dot com
- Date: Tue, 04 Feb 2014 11:14:48 +0000
- Subject: [s390] Split out pre-prologue rewrite into separate pass
- Authentication-results: sourceware.org; auth=none
s390_emit_prologue performs some optimisations on the function before
emitting the prologue. It also rewrites constant pool accesses to make
the base register explicit.
Doing this in the prologue pattern makes the interaction with direct
returns and simple_returns less obvious, so this patch splits the code
out into a new target-specific pre-prologue pass. I've called it
"early_mach" for want of a better name.
I also moved s390_option_override to the end of the file in order
to avoid some forward declarations that would have been needed otherwise.
The only change is at the very end of the function.
Tested on s390-linux-gnu and s390x-linux-gnu, using
unix{,-m31,-march=z10/-m31,-march=z10,-march=z196,-march=zEC12}
for the latter. OK to install?
Thanks,
Richard
gcc/
* config/s390/s390.c: Include tree-pass.h and context.h.
(s390_early_mach): New function, split out from...
(s390_emit_prologue): ...here.
(pass_data_s390_early_mach): New pass structure.
(pass_s390_early_mach): New class.
(s390_option_override): Create and register early_mach pass.
Move to end of file.
Index: gcc/config/s390/s390.c
===================================================================
--- gcc/config/s390/s390.c 2014-01-31 17:13:43.803357694 +0000
+++ gcc/config/s390/s390.c 2014-01-31 17:20:31.582196663 +0000
@@ -68,6 +68,8 @@ Software Foundation; either version 3, o
#include "params.h"
#include "cfgloop.h"
#include "opts.h"
+#include "tree-pass.h"
+#include "context.h"
/* Define the specific costs for a given cpu. */
@@ -1678,220 +1680,6 @@ s390_init_machine_status (void)
return ggc_alloc_cleared_machine_function ();
}
-static void
-s390_option_override (void)
-{
- unsigned int i;
- cl_deferred_option *opt;
- vec<cl_deferred_option> *v =
- (vec<cl_deferred_option> *) s390_deferred_options;
-
- if (v)
- FOR_EACH_VEC_ELT (*v, i, opt)
- {
- switch (opt->opt_index)
- {
- case OPT_mhotpatch:
- s390_hotpatch_trampoline_halfwords = (opt->value) ?
- s390_hotpatch_trampoline_halfwords_default : -1;
- break;
- case OPT_mhotpatch_:
- {
- int val;
-
- val = integral_argument (opt->arg);
- if (val == -1)
- {
- /* argument is not a plain number */
- error ("argument to %qs should be a non-negative integer",
- "-mhotpatch=");
- break;
- }
- else if (val > s390_hotpatch_trampoline_halfwords_max)
- {
- error ("argument to %qs is too large (max. %d)",
- "-mhotpatch=", s390_hotpatch_trampoline_halfwords_max);
- break;
- }
- s390_hotpatch_trampoline_halfwords = val;
- break;
- }
- default:
- gcc_unreachable ();
- }
- }
-
- /* Set up function hooks. */
- init_machine_status = s390_init_machine_status;
-
- /* Architecture mode defaults according to ABI. */
- if (!(target_flags_explicit & MASK_ZARCH))
- {
- if (TARGET_64BIT)
- target_flags |= MASK_ZARCH;
- else
- target_flags &= ~MASK_ZARCH;
- }
-
- /* Set the march default in case it hasn't been specified on
- cmdline. */
- if (s390_arch == PROCESSOR_max)
- {
- s390_arch_string = TARGET_ZARCH? "z900" : "g5";
- s390_arch = TARGET_ZARCH ? PROCESSOR_2064_Z900 : PROCESSOR_9672_G5;
- s390_arch_flags = processor_flags_table[(int)s390_arch];
- }
-
- /* Determine processor to tune for. */
- if (s390_tune == PROCESSOR_max)
- {
- s390_tune = s390_arch;
- s390_tune_flags = s390_arch_flags;
- }
-
- /* Sanity checks. */
- if (TARGET_ZARCH && !TARGET_CPU_ZARCH)
- error ("z/Architecture mode not supported on %s", s390_arch_string);
- if (TARGET_64BIT && !TARGET_ZARCH)
- error ("64-bit ABI not supported in ESA/390 mode");
-
- /* Use hardware DFP if available and not explicitly disabled by
- user. E.g. with -m31 -march=z10 -mzarch */
- if (!(target_flags_explicit & MASK_HARD_DFP) && TARGET_DFP)
- target_flags |= MASK_HARD_DFP;
-
- /* Enable hardware transactions if available and not explicitly
- disabled by user. E.g. with -m31 -march=zEC12 -mzarch */
- if (!(target_flags_explicit & MASK_OPT_HTM) && TARGET_CPU_HTM && TARGET_ZARCH)
- target_flags |= MASK_OPT_HTM;
-
- if (TARGET_HARD_DFP && !TARGET_DFP)
- {
- if (target_flags_explicit & MASK_HARD_DFP)
- {
- if (!TARGET_CPU_DFP)
- error ("hardware decimal floating point instructions"
- " not available on %s", s390_arch_string);
- if (!TARGET_ZARCH)
- error ("hardware decimal floating point instructions"
- " not available in ESA/390 mode");
- }
- else
- target_flags &= ~MASK_HARD_DFP;
- }
-
- if ((target_flags_explicit & MASK_SOFT_FLOAT) && TARGET_SOFT_FLOAT)
- {
- if ((target_flags_explicit & MASK_HARD_DFP) && TARGET_HARD_DFP)
- error ("-mhard-dfp can%'t be used in conjunction with -msoft-float");
-
- target_flags &= ~MASK_HARD_DFP;
- }
-
- /* Set processor cost function. */
- switch (s390_tune)
- {
- case PROCESSOR_2084_Z990:
- s390_cost = &z990_cost;
- break;
- case PROCESSOR_2094_Z9_109:
- s390_cost = &z9_109_cost;
- break;
- case PROCESSOR_2097_Z10:
- s390_cost = &z10_cost;
- break;
- case PROCESSOR_2817_Z196:
- s390_cost = &z196_cost;
- break;
- case PROCESSOR_2827_ZEC12:
- s390_cost = &zEC12_cost;
- break;
- default:
- s390_cost = &z900_cost;
- }
-
- if (TARGET_BACKCHAIN && TARGET_PACKED_STACK && TARGET_HARD_FLOAT)
- error ("-mbackchain -mpacked-stack -mhard-float are not supported "
- "in combination");
-
- if (s390_stack_size)
- {
- if (s390_stack_guard >= s390_stack_size)
- error ("stack size must be greater than the stack guard value");
- else if (s390_stack_size > 1 << 16)
- error ("stack size must not be greater than 64k");
- }
- else if (s390_stack_guard)
- error ("-mstack-guard implies use of -mstack-size");
-
-#ifdef TARGET_DEFAULT_LONG_DOUBLE_128
- if (!(target_flags_explicit & MASK_LONG_DOUBLE_128))
- target_flags |= MASK_LONG_DOUBLE_128;
-#endif
-
- if (s390_tune == PROCESSOR_2097_Z10
- || s390_tune == PROCESSOR_2817_Z196
- || s390_tune == PROCESSOR_2827_ZEC12)
- {
- maybe_set_param_value (PARAM_MAX_UNROLLED_INSNS, 100,
- global_options.x_param_values,
- global_options_set.x_param_values);
- maybe_set_param_value (PARAM_MAX_UNROLL_TIMES, 32,
- global_options.x_param_values,
- global_options_set.x_param_values);
- maybe_set_param_value (PARAM_MAX_COMPLETELY_PEELED_INSNS, 2000,
- global_options.x_param_values,
- global_options_set.x_param_values);
- maybe_set_param_value (PARAM_MAX_COMPLETELY_PEEL_TIMES, 64,
- global_options.x_param_values,
- global_options_set.x_param_values);
- }
-
- maybe_set_param_value (PARAM_MAX_PENDING_LIST_LENGTH, 256,
- global_options.x_param_values,
- global_options_set.x_param_values);
- /* values for loop prefetching */
- maybe_set_param_value (PARAM_L1_CACHE_LINE_SIZE, 256,
- global_options.x_param_values,
- global_options_set.x_param_values);
- maybe_set_param_value (PARAM_L1_CACHE_SIZE, 128,
- global_options.x_param_values,
- global_options_set.x_param_values);
- /* s390 has more than 2 levels and the size is much larger. Since
- we are always running virtualized assume that we only get a small
- part of the caches above l1. */
- maybe_set_param_value (PARAM_L2_CACHE_SIZE, 1500,
- global_options.x_param_values,
- global_options_set.x_param_values);
- maybe_set_param_value (PARAM_PREFETCH_MIN_INSN_TO_MEM_RATIO, 2,
- global_options.x_param_values,
- global_options_set.x_param_values);
- maybe_set_param_value (PARAM_SIMULTANEOUS_PREFETCHES, 6,
- global_options.x_param_values,
- global_options_set.x_param_values);
-
- /* This cannot reside in s390_option_optimization_table since HAVE_prefetch
- requires the arch flags to be evaluated already. Since prefetching
- is beneficial on s390, we enable it if available. */
- if (flag_prefetch_loop_arrays < 0 && HAVE_prefetch && optimize >= 3)
- flag_prefetch_loop_arrays = 1;
-
- /* Use the alternative scheduling-pressure algorithm by default. */
- maybe_set_param_value (PARAM_SCHED_PRESSURE_ALGORITHM, 2,
- global_options.x_param_values,
- global_options_set.x_param_values);
-
- if (TARGET_TPF)
- {
- /* Don't emit DWARF3/4 unless specifically selected. The TPF
- debuggers do not yet support DWARF 3/4. */
- if (!global_options_set.x_dwarf_strict)
- dwarf_strict = 1;
- if (!global_options_set.x_dwarf_version)
- dwarf_version = 2;
- }
-}
-
/* Map for smallest class containing reg regno. */
const enum reg_class regclass_map[FIRST_PSEUDO_REGISTER] =
@@ -8825,16 +8613,14 @@ s390_restore_gprs_from_fprs (void)
}
}
-/* Expand the prologue into a bunch of separate insns. */
-void
-s390_emit_prologue (void)
+/* A pass run immediately before shrink-wrapping and prologue and epilogue
+ generation. */
+
+static unsigned int
+s390_early_mach (void)
{
- rtx insn, addr;
- rtx temp_reg;
- int i;
- int offset;
- int next_fpr = 0;
+ rtx insn;
/* Try to get rid of the FPR clobbers. */
s390_optimize_nonescaping_tx ();
@@ -8844,17 +8630,57 @@ s390_emit_prologue (void)
/* Annotate all constant pool references to let the scheduler know
they implicitly use the base register. */
-
- push_topmost_sequence ();
-
for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
if (INSN_P (insn))
{
annotate_constant_pool_refs (&PATTERN (insn));
df_insn_rescan (insn);
}
+ return 0;
+}
- pop_topmost_sequence ();
+namespace {
+
+const pass_data pass_data_s390_early_mach =
+{
+ RTL_PASS, /* type */
+ "early_mach", /* name */
+ OPTGROUP_NONE, /* optinfo_flags */
+ false, /* has_gate */
+ true, /* has_execute */
+ TV_MACH_DEP, /* tv_id */
+ 0, /* properties_required */
+ 0, /* properties_provided */
+ 0, /* properties_destroyed */
+ 0, /* todo_flags_start */
+ ( TODO_df_verify | TODO_df_finish
+ | TODO_verify_rtl_sharing ), /* todo_flags_finish */
+};
+
+class pass_s390_early_mach : public rtl_opt_pass
+{
+public:
+ pass_s390_early_mach (gcc::context *ctxt)
+ : rtl_opt_pass (pass_data_s390_early_mach, ctxt)
+ {}
+
+ /* opt_pass methods: */
+ unsigned int execute () { return s390_early_mach (); }
+
+}; // class pass_s390_early_mach
+
+} // anon namespace
+
+/* Expand the prologue into a bunch of separate insns. */
+
+void
+s390_emit_prologue (void)
+{
+ rtx insn, addr;
+ rtx temp_reg;
+ int i;
+ int offset;
+ int next_fpr = 0;
/* Choose best register to use for temp use within prologue.
See below for why TPF must use the register 1. */
@@ -11921,6 +11747,235 @@ s390_loop_unroll_adjust (unsigned nunrol
}
}
+static void
+s390_option_override (void)
+{
+ unsigned int i;
+ cl_deferred_option *opt;
+ vec<cl_deferred_option> *v =
+ (vec<cl_deferred_option> *) s390_deferred_options;
+
+ if (v)
+ FOR_EACH_VEC_ELT (*v, i, opt)
+ {
+ switch (opt->opt_index)
+ {
+ case OPT_mhotpatch:
+ s390_hotpatch_trampoline_halfwords = (opt->value) ?
+ s390_hotpatch_trampoline_halfwords_default : -1;
+ break;
+ case OPT_mhotpatch_:
+ {
+ int val;
+
+ val = integral_argument (opt->arg);
+ if (val == -1)
+ {
+ /* argument is not a plain number */
+ error ("argument to %qs should be a non-negative integer",
+ "-mhotpatch=");
+ break;
+ }
+ else if (val > s390_hotpatch_trampoline_halfwords_max)
+ {
+ error ("argument to %qs is too large (max. %d)",
+ "-mhotpatch=", s390_hotpatch_trampoline_halfwords_max);
+ break;
+ }
+ s390_hotpatch_trampoline_halfwords = val;
+ break;
+ }
+ default:
+ gcc_unreachable ();
+ }
+ }
+
+ /* Set up function hooks. */
+ init_machine_status = s390_init_machine_status;
+
+ /* Architecture mode defaults according to ABI. */
+ if (!(target_flags_explicit & MASK_ZARCH))
+ {
+ if (TARGET_64BIT)
+ target_flags |= MASK_ZARCH;
+ else
+ target_flags &= ~MASK_ZARCH;
+ }
+
+ /* Set the march default in case it hasn't been specified on
+ cmdline. */
+ if (s390_arch == PROCESSOR_max)
+ {
+ s390_arch_string = TARGET_ZARCH? "z900" : "g5";
+ s390_arch = TARGET_ZARCH ? PROCESSOR_2064_Z900 : PROCESSOR_9672_G5;
+ s390_arch_flags = processor_flags_table[(int)s390_arch];
+ }
+
+ /* Determine processor to tune for. */
+ if (s390_tune == PROCESSOR_max)
+ {
+ s390_tune = s390_arch;
+ s390_tune_flags = s390_arch_flags;
+ }
+
+ /* Sanity checks. */
+ if (TARGET_ZARCH && !TARGET_CPU_ZARCH)
+ error ("z/Architecture mode not supported on %s", s390_arch_string);
+ if (TARGET_64BIT && !TARGET_ZARCH)
+ error ("64-bit ABI not supported in ESA/390 mode");
+
+ /* Use hardware DFP if available and not explicitly disabled by
+ user. E.g. with -m31 -march=z10 -mzarch */
+ if (!(target_flags_explicit & MASK_HARD_DFP) && TARGET_DFP)
+ target_flags |= MASK_HARD_DFP;
+
+ /* Enable hardware transactions if available and not explicitly
+ disabled by user. E.g. with -m31 -march=zEC12 -mzarch */
+ if (!(target_flags_explicit & MASK_OPT_HTM) && TARGET_CPU_HTM && TARGET_ZARCH)
+ target_flags |= MASK_OPT_HTM;
+
+ if (TARGET_HARD_DFP && !TARGET_DFP)
+ {
+ if (target_flags_explicit & MASK_HARD_DFP)
+ {
+ if (!TARGET_CPU_DFP)
+ error ("hardware decimal floating point instructions"
+ " not available on %s", s390_arch_string);
+ if (!TARGET_ZARCH)
+ error ("hardware decimal floating point instructions"
+ " not available in ESA/390 mode");
+ }
+ else
+ target_flags &= ~MASK_HARD_DFP;
+ }
+
+ if ((target_flags_explicit & MASK_SOFT_FLOAT) && TARGET_SOFT_FLOAT)
+ {
+ if ((target_flags_explicit & MASK_HARD_DFP) && TARGET_HARD_DFP)
+ error ("-mhard-dfp can%'t be used in conjunction with -msoft-float");
+
+ target_flags &= ~MASK_HARD_DFP;
+ }
+
+ /* Set processor cost function. */
+ switch (s390_tune)
+ {
+ case PROCESSOR_2084_Z990:
+ s390_cost = &z990_cost;
+ break;
+ case PROCESSOR_2094_Z9_109:
+ s390_cost = &z9_109_cost;
+ break;
+ case PROCESSOR_2097_Z10:
+ s390_cost = &z10_cost;
+ break;
+ case PROCESSOR_2817_Z196:
+ s390_cost = &z196_cost;
+ break;
+ case PROCESSOR_2827_ZEC12:
+ s390_cost = &zEC12_cost;
+ break;
+ default:
+ s390_cost = &z900_cost;
+ }
+
+ if (TARGET_BACKCHAIN && TARGET_PACKED_STACK && TARGET_HARD_FLOAT)
+ error ("-mbackchain -mpacked-stack -mhard-float are not supported "
+ "in combination");
+
+ if (s390_stack_size)
+ {
+ if (s390_stack_guard >= s390_stack_size)
+ error ("stack size must be greater than the stack guard value");
+ else if (s390_stack_size > 1 << 16)
+ error ("stack size must not be greater than 64k");
+ }
+ else if (s390_stack_guard)
+ error ("-mstack-guard implies use of -mstack-size");
+
+#ifdef TARGET_DEFAULT_LONG_DOUBLE_128
+ if (!(target_flags_explicit & MASK_LONG_DOUBLE_128))
+ target_flags |= MASK_LONG_DOUBLE_128;
+#endif
+
+ if (s390_tune == PROCESSOR_2097_Z10
+ || s390_tune == PROCESSOR_2817_Z196
+ || s390_tune == PROCESSOR_2827_ZEC12)
+ {
+ maybe_set_param_value (PARAM_MAX_UNROLLED_INSNS, 100,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ maybe_set_param_value (PARAM_MAX_UNROLL_TIMES, 32,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ maybe_set_param_value (PARAM_MAX_COMPLETELY_PEELED_INSNS, 2000,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ maybe_set_param_value (PARAM_MAX_COMPLETELY_PEEL_TIMES, 64,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ }
+
+ maybe_set_param_value (PARAM_MAX_PENDING_LIST_LENGTH, 256,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ /* values for loop prefetching */
+ maybe_set_param_value (PARAM_L1_CACHE_LINE_SIZE, 256,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ maybe_set_param_value (PARAM_L1_CACHE_SIZE, 128,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ /* s390 has more than 2 levels and the size is much larger. Since
+ we are always running virtualized assume that we only get a small
+ part of the caches above l1. */
+ maybe_set_param_value (PARAM_L2_CACHE_SIZE, 1500,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ maybe_set_param_value (PARAM_PREFETCH_MIN_INSN_TO_MEM_RATIO, 2,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+ maybe_set_param_value (PARAM_SIMULTANEOUS_PREFETCHES, 6,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+
+ /* This cannot reside in s390_option_optimization_table since HAVE_prefetch
+ requires the arch flags to be evaluated already. Since prefetching
+ is beneficial on s390, we enable it if available. */
+ if (flag_prefetch_loop_arrays < 0 && HAVE_prefetch && optimize >= 3)
+ flag_prefetch_loop_arrays = 1;
+
+ /* Use the alternative scheduling-pressure algorithm by default. */
+ maybe_set_param_value (PARAM_SCHED_PRESSURE_ALGORITHM, 2,
+ global_options.x_param_values,
+ global_options_set.x_param_values);
+
+ if (TARGET_TPF)
+ {
+ /* Don't emit DWARF3/4 unless specifically selected. The TPF
+ debuggers do not yet support DWARF 3/4. */
+ if (!global_options_set.x_dwarf_strict)
+ dwarf_strict = 1;
+ if (!global_options_set.x_dwarf_version)
+ dwarf_version = 2;
+ }
+
+ /* Register a target-specific optimization-and-lowering pass
+ to run immediately before prologue and epilogue generation.
+
+ Registering the pass must be done at start up. It's
+ convenient to do it here. */
+ opt_pass *new_pass = new pass_s390_early_mach (g);
+ struct register_pass_info insert_pass_s390_early_mach =
+ {
+ new_pass, /* pass */
+ "pro_and_epilogue", /* reference_pass_name */
+ 1, /* ref_pass_instance_number */
+ PASS_POS_INSERT_BEFORE /* po_op */
+ };
+ register_pass (&insert_pass_s390_early_mach);
+}
+
/* Initialize GCC target structure. */
#undef TARGET_ASM_ALIGNED_HI_OP