From 781634daea8cb788efb33994f4a19df76598542e Mon Sep 17 00:00:00 2001 From: Jakub Jelinek Date: Fri, 9 Oct 2020 10:19:16 +0200 Subject: [PATCH] vrp: Fix up gcc.target/aarch64/pr90838.c [PR97312, PR94801] > Perhaps another way out of this would be document and enforce that > __builtin_c[lt]z{,l,ll} etc calls are undefined at zero, but C[TL]Z ifn > calls are defined there based on *_DEFINED_VALUE_AT_ZERO (*) == 2 The following patch implements that, i.e. __builtin_c?z* now take full advantage of them being UB at zero, while the ifns are well defined at zero if *_DEFINED_VALUE_AT_ZERO (*) == 2. That is what fixes PR94801. Furthermore, to fix PR97312, if it is well defined at zero and the value at zero is prec, we don't lower the maximum unless the argument is known to be non-zero. For gimple-range.cc I guess we could improve it if needed e.g. by returning a [0,7][32,32] range for .CTZ of e.g. [0,137], but for now it (roughly) matches what vr-values.c does. 2020-10-09 Jakub Jelinek PR tree-optimization/94801 PR target/97312 * vr-values.c (vr_values::extract_range_basic) : When stmt is not an internal-fn call or C?Z_DEFINED_VALUE_AT_ZERO is not 2, assume argument is not zero and thus use [0, prec-1] range unless it can be further improved. For CTZ, don't update maxi from upper bound if it was previously prec. * gimple-range.cc (gimple_ranger::range_of_builtin_call) : Likewise. * gcc.dg/tree-ssa/pr94801.c: New test. --- gcc/gimple-range.cc | 94 +++++++++++++++++-------- gcc/testsuite/gcc.dg/tree-ssa/pr94801.c | 16 +++++ gcc/vr-values.c | 92 +++++++++++++++--------- 3 files changed, 136 insertions(+), 66 deletions(-) create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr94801.c diff --git a/gcc/gimple-range.cc b/gcc/gimple-range.cc index 2461bb78394..2ca86ed0e4c 100644 --- a/gcc/gimple-range.cc +++ b/gcc/gimple-range.cc @@ -636,28 +636,38 @@ gimple_ranger::range_of_builtin_call (irange &r, gcall *call) // __builtin_c[lt]z* return [0, prec-1], except when the // argument is 0, but that is undefined behavior. // - // On many targets where the CLZ RTL or optab value is defined - // for 0, the value is prec, so include that in the range by - // default. + // For __builtin_c[lt]z* consider argument of 0 always undefined + // behavior, for internal fns depending on C?Z_DEFINED_VALUE_AT_ZERO. arg = gimple_call_arg (call, 0); prec = TYPE_PRECISION (TREE_TYPE (arg)); mini = 0; - maxi = prec; + maxi = prec - 1; mode = SCALAR_INT_TYPE_MODE (TREE_TYPE (arg)); - if (optab_handler (clz_optab, mode) != CODE_FOR_nothing - && CLZ_DEFINED_VALUE_AT_ZERO (mode, zerov) - // Only handle the single common value. - && zerov != prec) - // Magic value to give up, unless we can prove arg is non-zero. - mini = -2; + if (gimple_call_internal_p (call)) + { + if (optab_handler (clz_optab, mode) != CODE_FOR_nothing + && CLZ_DEFINED_VALUE_AT_ZERO (mode, zerov) == 2) + { + // Only handle the single common value. + if (zerov == prec) + maxi = prec; + else + // Magic value to give up, unless we can prove arg is non-zero. + mini = -2; + } + } gcc_assert (range_of_expr (r, arg, call)); // From clz of minimum we can compute result maximum. if (r.constant_p ()) { - maxi = prec - 1 - wi::floor_log2 (r.lower_bound ()); - if (maxi != prec) - mini = 0; + int newmaxi = prec - 1 - wi::floor_log2 (r.lower_bound ()); + // Argument is unsigned, so do nothing if it is [0, ...] range. + if (newmaxi != prec) + { + mini = 0; + maxi = newmaxi; + } } else if (!range_includes_zero_p (&r)) { @@ -669,9 +679,17 @@ gimple_ranger::range_of_builtin_call (irange &r, gcall *call) // From clz of maximum we can compute result minimum. if (r.constant_p ()) { - mini = prec - 1 - wi::floor_log2 (r.upper_bound ()); - if (mini == prec) - break; + int newmini = prec - 1 - wi::floor_log2 (r.upper_bound ()); + if (newmini == prec) + { + // Argument range is [0, 0]. If CLZ_DEFINED_VALUE_AT_ZERO + // is 2 with VALUE of prec, return [prec, prec], otherwise + // ignore the range. + if (maxi == prec) + mini = prec; + } + else + mini = newmini; } if (mini == -2) break; @@ -682,25 +700,27 @@ gimple_ranger::range_of_builtin_call (irange &r, gcall *call) // __builtin_ctz* return [0, prec-1], except for when the // argument is 0, but that is undefined behavior. // - // If there is a ctz optab for this mode and - // CTZ_DEFINED_VALUE_AT_ZERO, include that in the range, - // otherwise just assume 0 won't be seen. + // For __builtin_ctz* consider argument of 0 always undefined + // behavior, for internal fns depending on CTZ_DEFINED_VALUE_AT_ZERO. arg = gimple_call_arg (call, 0); prec = TYPE_PRECISION (TREE_TYPE (arg)); mini = 0; maxi = prec - 1; mode = SCALAR_INT_TYPE_MODE (TREE_TYPE (arg)); - if (optab_handler (ctz_optab, mode) != CODE_FOR_nothing - && CTZ_DEFINED_VALUE_AT_ZERO (mode, zerov)) + if (gimple_call_internal_p (call)) { - // Handle only the two common values. - if (zerov == -1) - mini = -1; - else if (zerov == prec) - maxi = prec; - else - // Magic value to give up, unless we can prove arg is non-zero. - mini = -2; + if (optab_handler (ctz_optab, mode) != CODE_FOR_nothing + && CTZ_DEFINED_VALUE_AT_ZERO (mode, zerov) == 2) + { + // Handle only the two common values. + if (zerov == -1) + mini = -1; + else if (zerov == prec) + maxi = prec; + else + // Magic value to give up, unless we can prove arg is non-zero. + mini = -2; + } } gcc_assert (range_of_expr (r, arg, call)); if (!r.undefined_p ()) @@ -714,8 +734,20 @@ gimple_ranger::range_of_builtin_call (irange &r, gcall *call) // the maximum. wide_int max = r.upper_bound (); if (max == 0) - break; - maxi = wi::floor_log2 (max); + { + // Argument is [0, 0]. If CTZ_DEFINED_VALUE_AT_ZERO + // is 2 with value -1 or prec, return [-1, -1] or [prec, prec]. + // Otherwise ignore the range. + if (mini == -1) + maxi = -1; + else if (maxi == prec) + mini = prec; + } + // If value at zero is prec and 0 is in the range, we can't lower + // the upper bound. We could create two separate ranges though, + // [0,floor_log2(max)][prec,prec] though. + else if (maxi != prec) + maxi = wi::floor_log2 (max); } if (mini == -2) break; diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr94801.c b/gcc/testsuite/gcc.dg/tree-ssa/pr94801.c new file mode 100644 index 00000000000..5382e5ee960 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr94801.c @@ -0,0 +1,16 @@ +/* PR tree-optimization/94801 */ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-optimized" } */ +/* { dg-final { scan-tree-dump-times "return 0;" 2 "optimized" } } */ + +int +foo (int a) +{ + return __builtin_clz (a) >> 5; +} + +int +bar (int a) +{ + return __builtin_ctz (a) >> 5; +} diff --git a/gcc/vr-values.c b/gcc/vr-values.c index 88aa672466c..da0b249278b 100644 --- a/gcc/vr-values.c +++ b/gcc/vr-values.c @@ -1208,34 +1208,42 @@ vr_values::extract_range_basic (value_range_equiv *vr, gimple *stmt) mini = 0; maxi = 1; goto bitop_builtin; - /* __builtin_c[lt]z* return [0, prec-1], except for + /* __builtin_clz* return [0, prec-1], except for when the argument is 0, but that is undefined behavior. - On many targets where the CLZ RTL or optab value is defined - for 0 the value is prec, so include that in the range - by default. */ + Always handle __builtin_clz* which can be only written + by user as UB on 0 and so [0, prec-1] range, and the internal-fn + calls depending on how CLZ_DEFINED_VALUE_AT_ZERO is defined. */ CASE_CFN_CLZ: arg = gimple_call_arg (stmt, 0); prec = TYPE_PRECISION (TREE_TYPE (arg)); mini = 0; - maxi = prec; + maxi = prec - 1; mode = SCALAR_INT_TYPE_MODE (TREE_TYPE (arg)); - if (optab_handler (clz_optab, mode) != CODE_FOR_nothing - && CLZ_DEFINED_VALUE_AT_ZERO (mode, zerov) - /* Handle only the single common value. */ - && zerov != prec) - /* Magic value to give up, unless vr0 proves - arg is non-zero. */ - mini = -2; + if (gimple_call_internal_p (stmt)) + { + if (optab_handler (clz_optab, mode) != CODE_FOR_nothing + && CLZ_DEFINED_VALUE_AT_ZERO (mode, zerov) == 2) + { + /* Handle only the single common value. */ + if (zerov == prec) + maxi = prec; + /* Magic value to give up, unless vr0 proves + arg is non-zero. */ + else + mini = -2; + } + } if (TREE_CODE (arg) == SSA_NAME) { const value_range_equiv *vr0 = get_value_range (arg); /* From clz of VR_RANGE minimum we can compute result maximum. */ if (vr0->kind () == VR_RANGE - && TREE_CODE (vr0->min ()) == INTEGER_CST) + && TREE_CODE (vr0->min ()) == INTEGER_CST + && integer_nonzerop (vr0->min ())) { maxi = prec - 1 - tree_floor_log2 (vr0->min ()); - if (maxi != prec) + if (mini == -2) mini = 0; } else if (vr0->kind () == VR_ANTI_RANGE @@ -1251,9 +1259,14 @@ vr_values::extract_range_basic (value_range_equiv *vr, gimple *stmt) if (vr0->kind () == VR_RANGE && TREE_CODE (vr0->max ()) == INTEGER_CST) { - mini = prec - 1 - tree_floor_log2 (vr0->max ()); - if (mini == prec) - break; + int newmini = prec - 1 - tree_floor_log2 (vr0->max ()); + if (newmini == prec) + { + if (maxi == prec) + mini = prec; + } + else + mini = newmini; } } if (mini == -2) @@ -1261,27 +1274,30 @@ vr_values::extract_range_basic (value_range_equiv *vr, gimple *stmt) goto bitop_builtin; /* __builtin_ctz* return [0, prec-1], except for when the argument is 0, but that is undefined behavior. - If there is a ctz optab for this mode and - CTZ_DEFINED_VALUE_AT_ZERO, include that in the range, - otherwise just assume 0 won't be seen. */ + Always handle __builtin_ctz* which can be only written + by user as UB on 0 and so [0, prec-1] range, and the internal-fn + calls depending on how CTZ_DEFINED_VALUE_AT_ZERO is defined. */ CASE_CFN_CTZ: arg = gimple_call_arg (stmt, 0); prec = TYPE_PRECISION (TREE_TYPE (arg)); mini = 0; maxi = prec - 1; mode = SCALAR_INT_TYPE_MODE (TREE_TYPE (arg)); - if (optab_handler (ctz_optab, mode) != CODE_FOR_nothing - && CTZ_DEFINED_VALUE_AT_ZERO (mode, zerov)) + if (gimple_call_internal_p (stmt)) { - /* Handle only the two common values. */ - if (zerov == -1) - mini = -1; - else if (zerov == prec) - maxi = prec; - else - /* Magic value to give up, unless vr0 proves - arg is non-zero. */ - mini = -2; + if (optab_handler (ctz_optab, mode) != CODE_FOR_nothing + && CTZ_DEFINED_VALUE_AT_ZERO (mode, zerov) == 2) + { + /* Handle only the two common values. */ + if (zerov == -1) + mini = -1; + else if (zerov == prec) + maxi = prec; + else + /* Magic value to give up, unless vr0 proves + arg is non-zero. */ + mini = -2; + } } if (TREE_CODE (arg) == SSA_NAME) { @@ -1300,10 +1316,16 @@ vr_values::extract_range_basic (value_range_equiv *vr, gimple *stmt) if (vr0->kind () == VR_RANGE && TREE_CODE (vr0->max ()) == INTEGER_CST) { - maxi = tree_floor_log2 (vr0->max ()); - /* For vr0 [0, 0] give up. */ - if (maxi == -1) - break; + int newmaxi = tree_floor_log2 (vr0->max ()); + if (newmaxi == -1) + { + if (mini == -1) + maxi = -1; + else if (maxi == prec) + mini = prec; + } + else if (maxi != prec) + maxi = newmaxi; } } if (mini == -2) -- 2.30.2