From 43558bcc9df59fc2a1d611030a936931e92cc6e5 Mon Sep 17 00:00:00 2001 From: Jan Hubicka Date: Mon, 23 Nov 2009 21:01:29 +0100 Subject: [PATCH] ipa-cp.c (ipcp_compute_node_scale): Work around completely wrong profile updates. * ipa-cp.c (ipcp_compute_node_scale): Work around completely wrong profile updates. * predict.c (counts_to_freqs): Be expected for ENTRY/EXIT block having largest frequency. * ira-live.c (ira_implicitly_set_insn_hard_regs): Silecne used uninitalized warning. * tree-optimize.c (execute_fixup_cfg): Rescale entry and exit block frequencies. From-SVN: r154462 --- gcc/ChangeLog | 11 +++++++++++ gcc/ipa-cp.c | 14 +++++++++++++- gcc/ira-lives.c | 2 +- gcc/predict.c | 2 +- gcc/tree-optimize.c | 4 ++++ 5 files changed, 30 insertions(+), 3 deletions(-) diff --git a/gcc/ChangeLog b/gcc/ChangeLog index bce59754feb..b8378a58b12 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,14 @@ +2009-11-23 Jan Hubicka + + * ipa-cp.c (ipcp_compute_node_scale): Work around completely + wrong profile updates. + * predict.c (counts_to_freqs): Be expected for ENTRY/EXIT block + having largest frequency. + * ira-live.c (ira_implicitly_set_insn_hard_regs): Silecne + used uninitalized warning. + * tree-optimize.c (execute_fixup_cfg): Rescale entry and exit block + frequencies. + 2009-11-23 Uros Bizjak * config/alpha/alpha.md (*cmp_sadd_sidi): Use gen_lowpart instead diff --git a/gcc/ipa-cp.c b/gcc/ipa-cp.c index 4b632c0d9fb..2af2c7f2d6f 100644 --- a/gcc/ipa-cp.c +++ b/gcc/ipa-cp.c @@ -578,7 +578,13 @@ build_const_val (struct ipcp_lattice *lat, tree tree_type) /* Compute the proper scale for NODE. It is the ratio between the number of direct calls (represented on the incoming cgraph_edges) and sum of all - invocations of NODE (represented as count in cgraph_node). */ + invocations of NODE (represented as count in cgraph_node). + + FIXME: This code is wrong. Since the callers can be also clones and + the clones are not scaled yet, the sums gets unrealistically high. + To properly compute the counts, we would need to do propagation across + callgraph (as external call to A might imply call to non-clonned B + if A's clone calls clonned B). */ static void ipcp_compute_node_scale (struct cgraph_node *node) { @@ -589,6 +595,12 @@ ipcp_compute_node_scale (struct cgraph_node *node) /* Compute sum of all counts of callers. */ for (cs = node->callers; cs != NULL; cs = cs->next_caller) sum += cs->count; + /* Work around the unrealistically high sum problem. We just don't want + the non-cloned body to have negative or very low frequency. Since + majority of execution time will be spent in clones anyway, this should + give good enough profile. */ + if (sum > node->count * 9 / 10) + sum = node->count * 9 / 10; if (node->count == 0) ipcp_set_node_scale (node, 0); else diff --git a/gcc/ira-lives.c b/gcc/ira-lives.c index c67e89cbd79..ea241f4732c 100644 --- a/gcc/ira-lives.c +++ b/gcc/ira-lives.c @@ -745,7 +745,7 @@ single_reg_operand_class (int op_num) void ira_implicitly_set_insn_hard_regs (HARD_REG_SET *set) { - int i, c, regno; + int i, c, regno = 0; bool ignore_p; enum reg_class cl; rtx op; diff --git a/gcc/predict.c b/gcc/predict.c index df859066b96..058901e5903 100644 --- a/gcc/predict.c +++ b/gcc/predict.c @@ -2020,7 +2020,7 @@ counts_to_freqs (void) gcov_type count_max, true_count_max = 0; basic_block bb; - FOR_EACH_BB (bb) + FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR, NULL, next_bb) true_count_max = MAX (bb->count, true_count_max); count_max = MAX (true_count_max, 1); diff --git a/gcc/tree-optimize.c b/gcc/tree-optimize.c index 61d687daa13..778658a70b4 100644 --- a/gcc/tree-optimize.c +++ b/gcc/tree-optimize.c @@ -255,6 +255,10 @@ execute_fixup_cfg (void) else count_scale = REG_BR_PROB_BASE; + ENTRY_BLOCK_PTR->count = cgraph_node (current_function_decl)->count; + EXIT_BLOCK_PTR->count = (EXIT_BLOCK_PTR->count * count_scale + + REG_BR_PROB_BASE / 2) / REG_BR_PROB_BASE; + FOR_EACH_BB (bb) { bb->count = (bb->count * count_scale -- 2.30.2