From 4ef79c960aa0967cf0298dc496a30a40d86ebd3c Mon Sep 17 00:00:00 2001 From: Richard Sandiford Date: Tue, 3 Jul 2018 10:03:17 +0000 Subject: [PATCH] [15/n] PR85694: Try to split existing casts in widened patterns The main over-widening patch can introduce quite a few extra casts, and in many cases those casts simply "tap into" an intermediate point in an existing extension. E.g. if we have: unsigned char a; int ax = (int) a; and a later operation using ax is shortened to "unsigned short", we would need: unsigned short ax' = (unsigned short) a; The a->ax extension requires one set of unpacks to get to unsigned short and another set of unpacks to get to int. The first set are then duplicated for ax'. If both ax and ax' are needed, the a->ax' extension would end up counting twice during cost calculations. This patch rewrites the original: int ax = (int) a; into a pattern: unsigned short ax' = (unsigned short) a; int ax = (int) ax'; so that each extension only counts once. 2018-07-03 Richard Sandiford gcc/ * tree-vect-patterns.c (vect_split_statement): New function. (vect_convert_input): Use it to try to split an existing cast. gcc/testsuite/ * gcc.dg/vect/vect-over-widen-5.c: Test that the extensions get split into two for use by the over-widening pattern. * gcc.dg/vect/vect-over-widen-6.c: Likewise. * gcc.dg/vect/vect-over-widen-7.c: Likewise. * gcc.dg/vect/vect-over-widen-8.c: Likewise. * gcc.dg/vect/vect-over-widen-9.c: Likewise. * gcc.dg/vect/vect-over-widen-10.c: Likewise. * gcc.dg/vect/vect-over-widen-11.c: Likewise. * gcc.dg/vect/vect-over-widen-12.c: Likewise. * gcc.dg/vect/vect-over-widen-13.c: Likewise. * gcc.dg/vect/vect-over-widen-14.c: Likewise. * gcc.dg/vect/vect-over-widen-15.c: Likewise. * gcc.dg/vect/vect-over-widen-16.c: Likewise. * gcc.dg/vect/vect-over-widen-22.c: New test. From-SVN: r262334 --- gcc/ChangeLog | 5 + gcc/testsuite/ChangeLog | 17 +++ .../gcc.dg/vect/vect-over-widen-10.c | 1 + .../gcc.dg/vect/vect-over-widen-11.c | 1 + .../gcc.dg/vect/vect-over-widen-12.c | 1 + .../gcc.dg/vect/vect-over-widen-13.c | 1 + .../gcc.dg/vect/vect-over-widen-14.c | 1 + .../gcc.dg/vect/vect-over-widen-15.c | 1 + .../gcc.dg/vect/vect-over-widen-16.c | 1 + .../gcc.dg/vect/vect-over-widen-22.c | 53 +++++++++ gcc/testsuite/gcc.dg/vect/vect-over-widen-5.c | 1 + gcc/testsuite/gcc.dg/vect/vect-over-widen-6.c | 1 + gcc/testsuite/gcc.dg/vect/vect-over-widen-7.c | 1 + gcc/testsuite/gcc.dg/vect/vect-over-widen-8.c | 1 + gcc/testsuite/gcc.dg/vect/vect-over-widen-9.c | 1 + gcc/tree-vect-patterns.c | 103 ++++++++++++++++++ 16 files changed, 190 insertions(+) create mode 100644 gcc/testsuite/gcc.dg/vect/vect-over-widen-22.c diff --git a/gcc/ChangeLog b/gcc/ChangeLog index ab97a84fb0b..fd928b0a1fd 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,8 @@ +2018-07-03 Richard Sandiford + + * tree-vect-patterns.c (vect_split_statement): New function. + (vect_convert_input): Use it to try to split an existing cast. + 2018-07-03 Richard Sandiford * poly-int.h (print_hex): New function. diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index 0028d4f88ac..90aa4d7e22a 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,20 @@ +2018-07-03 Richard Sandiford + + * gcc.dg/vect/vect-over-widen-5.c: Test that the extensions + get split into two for use by the over-widening pattern. + * gcc.dg/vect/vect-over-widen-6.c: Likewise. + * gcc.dg/vect/vect-over-widen-7.c: Likewise. + * gcc.dg/vect/vect-over-widen-8.c: Likewise. + * gcc.dg/vect/vect-over-widen-9.c: Likewise. + * gcc.dg/vect/vect-over-widen-10.c: Likewise. + * gcc.dg/vect/vect-over-widen-11.c: Likewise. + * gcc.dg/vect/vect-over-widen-12.c: Likewise. + * gcc.dg/vect/vect-over-widen-13.c: Likewise. + * gcc.dg/vect/vect-over-widen-14.c: Likewise. + * gcc.dg/vect/vect-over-widen-15.c: Likewise. + * gcc.dg/vect/vect-over-widen-16.c: Likewise. + * gcc.dg/vect/vect-over-widen-22.c: New test. + 2018-07-03 Richard Sandiford * gcc.dg/vect/vect-widen-mult-u8-u32.c: Check specifically for a diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-10.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-10.c index 394a5a19e0d..f0140e4ef6d 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-10.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-10.c @@ -11,6 +11,7 @@ #include "vect-over-widen-9.c" +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+ } "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 1} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 2} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-11.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-11.c index 97ab57f12c4..222d854b2d6 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-11.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-11.c @@ -55,6 +55,7 @@ main (void) return 0; } +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+ } "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 1} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 2} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-12.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-12.c index 0d5473ea166..ddb3bd8c0d3 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-12.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-12.c @@ -11,6 +11,7 @@ #include "vect-over-widen-11.c" +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+ } "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 1} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 2} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-13.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-13.c index b89ed8bf4a7..b25db881afb 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-13.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-13.c @@ -43,6 +43,7 @@ main (void) return 0; } +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* / 2} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_cast_forwprop_pattern: detected:[^\n]* = \(signed char\)} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-14.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-14.c index 7b5ba2311a0..dfa09f5d2ca 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-14.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-14.c @@ -11,6 +11,7 @@ #include "vect-over-widen-13.c" +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 1} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_cast_forwprop_pattern: detected:[^\n]* = \(unsigned char\)} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-15.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-15.c index e898e87fd0b..d31050ee926 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-15.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-15.c @@ -45,6 +45,7 @@ main (void) return 0; } +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* / 2} "vect" } } */ /* { dg-final { scan-tree-dump-not {vect_recog_cast_forwprop_pattern: detected} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-16.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-16.c index 04293454d32..4584c586da1 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-16.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-16.c @@ -11,6 +11,7 @@ #include "vect-over-widen-15.c" +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 1} "vect" } } */ /* { dg-final { scan-tree-dump-not {vect_recog_cast_forwprop_pattern: detected} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-22.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-22.c new file mode 100644 index 00000000000..187bdf159fe --- /dev/null +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-22.c @@ -0,0 +1,53 @@ +/* { dg-require-effective-target vect_int } */ +/* { dg-require-effective-target vect_shift } */ +/* { dg-require-effective-target vect_pack_trunc } */ +/* { dg-require-effective-target vect_unpack } */ + +#include "tree-vect.h" + +#define N 111 + +/* The addition should be narrowed to short. */ + +void __attribute__ ((noipa)) +f (unsigned int *restrict a, unsigned int *restrict b, + unsigned short *restrict c, unsigned char *restrict d, unsigned int e) +{ + e &= 0xff; + for (__INTPTR_TYPE__ i = 0; i < N; ++i) + { + unsigned int xor = d[i] ^ e; + a[i] = c[i] | xor; + b[i] = xor; + } +} + +int +main (void) +{ + check_vect (); + + unsigned int a[N], b[N]; + unsigned short c[N]; + unsigned char d[N]; + for (int i = 0; i < N; ++i) + { + c[i] = i * 11; + d[i] = i * 2 + 3; + asm volatile ("" ::: "memory"); + } + f (a, b, c, d, 0x73); + for (int i = 0; i < N; ++i) + if (b[i] != ((i * 2 + 3) ^ 0x73) + || a[i] != ((i * 11) | b[i])) + __builtin_abort (); + + return 0; +} + +/* { dg-final { scan-tree-dump {Splitting pattern statement} "vect" } } */ +/* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \^} "vect" } } */ +/* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \|} "vect" } } */ +/* { dg-final { scan-tree-dump {demoting [^\n]* to [^\n]*char} "vect" } } */ +/* { dg-final { scan-tree-dump {demoting [^\n]* to [^\n]*short} "vect" } } */ +/* { dg-final { scan-tree-dump-times "vectorized 1 loop" 1 "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-5.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-5.c index 56d2396e394..269df5387d2 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-5.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-5.c @@ -44,6 +44,7 @@ main (void) return 0; } +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+ } "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 1} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_cast_forwprop_pattern: detected:[^\n]* \(signed char\)} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-6.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-6.c index 9fe0e056cc4..bda92c965e0 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-6.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-6.c @@ -9,6 +9,7 @@ #include "vect-over-widen-5.c" +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+ } "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 1} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_cast_forwprop_pattern: detected:[^\n]* \(unsigned char\)} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-7.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-7.c index a8166b38505..314a6828c53 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-7.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-7.c @@ -46,6 +46,7 @@ main (void) return 0; } +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+ } "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 2} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_cast_forwprop_pattern: detected:[^\n]* \(signed char\)} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-8.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-8.c index 238f577adea..553c0712a79 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-8.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-8.c @@ -12,6 +12,7 @@ #include "vect-over-widen-7.c" +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+ } "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 2} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_cast_forwprop_pattern: detected:[^\n]* \(unsigned char\)} "vect" } } */ diff --git a/gcc/testsuite/gcc.dg/vect/vect-over-widen-9.c b/gcc/testsuite/gcc.dg/vect/vect-over-widen-9.c index a50f8198bfd..5baba09a575 100644 --- a/gcc/testsuite/gcc.dg/vect/vect-over-widen-9.c +++ b/gcc/testsuite/gcc.dg/vect/vect-over-widen-9.c @@ -50,6 +50,7 @@ main (void) return 0; } +/* { dg-final { scan-tree-dump {Splitting statement} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* \+ } "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 1} "vect" } } */ /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 2} "vect" } } */ diff --git a/gcc/tree-vect-patterns.c b/gcc/tree-vect-patterns.c index 91076f41964..a1649d8b0fe 100644 --- a/gcc/tree-vect-patterns.c +++ b/gcc/tree-vect-patterns.c @@ -634,6 +634,97 @@ vect_recog_temp_ssa_var (tree type, gimple *stmt) return make_temp_ssa_name (type, stmt, "patt"); } +/* STMT2_INFO describes a type conversion that could be split into STMT1 + followed by a version of STMT2_INFO that takes NEW_RHS as its first + input. Try to do this using pattern statements, returning true on + success. */ + +static bool +vect_split_statement (stmt_vec_info stmt2_info, tree new_rhs, + gimple *stmt1, tree vectype) +{ + if (is_pattern_stmt_p (stmt2_info)) + { + /* STMT2_INFO is part of a pattern. Get the statement to which + the pattern is attached. */ + stmt_vec_info orig_stmt2_info + = vinfo_for_stmt (STMT_VINFO_RELATED_STMT (stmt2_info)); + vect_init_pattern_stmt (stmt1, orig_stmt2_info, vectype); + + if (dump_enabled_p ()) + { + dump_printf_loc (MSG_NOTE, vect_location, + "Splitting pattern statement: "); + dump_gimple_stmt (MSG_NOTE, TDF_SLIM, stmt2_info->stmt, 0); + } + + /* Since STMT2_INFO is a pattern statement, we can change it + in-situ without worrying about changing the code for the + containing block. */ + gimple_assign_set_rhs1 (stmt2_info->stmt, new_rhs); + + if (dump_enabled_p ()) + { + dump_printf_loc (MSG_NOTE, vect_location, "into: "); + dump_gimple_stmt (MSG_NOTE, TDF_SLIM, stmt1, 0); + dump_printf_loc (MSG_NOTE, vect_location, "and: "); + dump_gimple_stmt (MSG_NOTE, TDF_SLIM, stmt2_info->stmt, 0); + } + + gimple_seq *def_seq = &STMT_VINFO_PATTERN_DEF_SEQ (orig_stmt2_info); + if (STMT_VINFO_RELATED_STMT (orig_stmt2_info) == stmt2_info->stmt) + /* STMT2_INFO is the actual pattern statement. Add STMT1 + to the end of the definition sequence. */ + gimple_seq_add_stmt_without_update (def_seq, stmt1); + else + { + /* STMT2_INFO belongs to the definition sequence. Insert STMT1 + before it. */ + gimple_stmt_iterator gsi = gsi_for_stmt (stmt2_info->stmt, def_seq); + gsi_insert_before_without_update (&gsi, stmt1, GSI_SAME_STMT); + } + return true; + } + else + { + /* STMT2_INFO doesn't yet have a pattern. Try to create a + two-statement pattern now. */ + gcc_assert (!STMT_VINFO_RELATED_STMT (stmt2_info)); + tree lhs_type = TREE_TYPE (gimple_get_lhs (stmt2_info->stmt)); + tree lhs_vectype = get_vectype_for_scalar_type (lhs_type); + if (!lhs_vectype) + return false; + + if (dump_enabled_p ()) + { + dump_printf_loc (MSG_NOTE, vect_location, + "Splitting statement: "); + dump_gimple_stmt (MSG_NOTE, TDF_SLIM, stmt2_info->stmt, 0); + } + + /* Add STMT1 as a singleton pattern definition sequence. */ + gimple_seq *def_seq = &STMT_VINFO_PATTERN_DEF_SEQ (stmt2_info); + vect_init_pattern_stmt (stmt1, stmt2_info, vectype); + gimple_seq_add_stmt_without_update (def_seq, stmt1); + + /* Build the second of the two pattern statements. */ + tree new_lhs = vect_recog_temp_ssa_var (lhs_type, NULL); + gassign *new_stmt2 = gimple_build_assign (new_lhs, NOP_EXPR, new_rhs); + vect_set_pattern_stmt (new_stmt2, stmt2_info, lhs_vectype); + + if (dump_enabled_p ()) + { + dump_printf_loc (MSG_NOTE, vect_location, + "into pattern statements: "); + dump_gimple_stmt (MSG_NOTE, TDF_SLIM, stmt1, 0); + dump_printf_loc (MSG_NOTE, vect_location, "and: "); + dump_gimple_stmt (MSG_NOTE, TDF_SLIM, new_stmt2, 0); + } + + return true; + } +} + /* Convert UNPROM to TYPE and return the result, adding new statements to STMT_INFO's pattern definition statements if no better way is available. VECTYPE is the vector form of TYPE. */ @@ -662,6 +753,18 @@ vect_convert_input (stmt_vec_info stmt_info, tree type, tree new_op = vect_recog_temp_ssa_var (type, NULL); gassign *new_stmt = gimple_build_assign (new_op, NOP_EXPR, unprom->op); + /* If the operation is the input to a vectorizable cast, try splitting + that cast into two, taking the required result as a mid-way point. */ + if (unprom->caster) + { + tree lhs = gimple_get_lhs (unprom->caster->stmt); + if (TYPE_PRECISION (TREE_TYPE (lhs)) > TYPE_PRECISION (type) + && TYPE_PRECISION (type) > TYPE_PRECISION (unprom->type) + && (TYPE_UNSIGNED (unprom->type) || !TYPE_UNSIGNED (type)) + && vect_split_statement (unprom->caster, new_op, new_stmt, vectype)) + return new_op; + } + /* If OP is an external value, see if we can insert the new statement on an incoming edge. */ if (unprom->dt == vect_external_def) -- 2.30.2