gcc/Simplify-X-C1-C2.patch

This backport contains 2 patch from gcc main stream tree.
The commit id of these patchs list as following in the order of time.

0001-Simplify-X-C1-C2-with-undefined-overflow.patch
ca2b8c082c4f16919071c9f8de8db0b33b54c405

0002-Simplify-X-C1-C2-with-wrapping-overflow.patch
287522613d661b4c5ba8403b051eb470c1674cba

diff -Nurp a/gcc/expr.c b/gcc/expr.c
--- a/gcc/expr.c	2021-03-17 16:34:24.700000000 +0800
+++ b/gcc/expr.c	2021-03-17 10:30:11.500000000 +0800
@@ -11706,38 +11706,6 @@ string_constant (tree arg, tree *ptr_off
   return init;
 }
 
-/* Compute the modular multiplicative inverse of A modulo M
-   using extended Euclid's algorithm.  Assumes A and M are coprime.  */
-static wide_int
-mod_inv (const wide_int &a, const wide_int &b)
-{
-  /* Verify the assumption.  */
-  gcc_checking_assert (wi::eq_p (wi::gcd (a, b), 1));
-
-  unsigned int p = a.get_precision () + 1;
-  gcc_checking_assert (b.get_precision () + 1 == p);
-  wide_int c = wide_int::from (a, p, UNSIGNED);
-  wide_int d = wide_int::from (b, p, UNSIGNED);
-  wide_int x0 = wide_int::from (0, p, UNSIGNED);
-  wide_int x1 = wide_int::from (1, p, UNSIGNED);
-
-  if (wi::eq_p (b, 1))
-    return wide_int::from (1, p, UNSIGNED);
-
-  while (wi::gt_p (c, 1, UNSIGNED))
-    {
-      wide_int t = d;
-      wide_int q = wi::divmod_trunc (c, d, UNSIGNED, &d);
-      c = t;
-      wide_int s = x0;
-      x0 = wi::sub (x1, wi::mul (q, x0));
-      x1 = s;
-    }
-  if (wi::lt_p (x1, 0, SIGNED))
-    x1 += d;
-  return x1;
-}
-
 /* Optimize x % C1 == C2 for signed modulo if C1 is a power of two and C2
    is non-zero and C3 ((1<<(prec-1)) | (C1 - 1)):
    for C2 > 0 to x & C3 == C2
@@ -11948,7 +11916,7 @@ maybe_optimize_mod_cmp (enum tree_code c
   w = wi::lrshift (w, shift);
   wide_int a = wide_int::from (w, prec + 1, UNSIGNED);
   wide_int b = wi::shifted_mask (prec, 1, false, prec + 1);
-  wide_int m = wide_int::from (mod_inv (a, b), prec, UNSIGNED);
+  wide_int m = wide_int::from (wi::mod_inv (a, b), prec, UNSIGNED);
   tree c3 = wide_int_to_tree (type, m);
   tree c5 = NULL_TREE;
   wide_int d, e;
diff -Nurp a/gcc/match.pd b/gcc/match.pd
--- a/gcc/match.pd	2021-03-17 16:34:19.320000000 +0800
+++ b/gcc/match.pd	2021-03-17 10:30:11.500000000 +0800
@@ -3290,6 +3290,35 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
      (scmp @0 @2)
      (cmp @0 @2))))))
 
+/* For integral types with undefined overflow fold
+   x * C1 == C2 into x == C2 / C1 or false.
+   If overflow wraps and C1 is odd, simplify to x == C2 / C1 in the ring
+   Z / 2^n Z.  */
+(for cmp (eq ne)
+ (simplify
+  (cmp (mult @0 INTEGER_CST@1) INTEGER_CST@2)
+  (if (INTEGRAL_TYPE_P (TREE_TYPE (@0))
+       && TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (@0))
+       && wi::to_wide (@1) != 0)
+   (with { widest_int quot; }
+    (if (wi::multiple_of_p (wi::to_widest (@2), wi::to_widest (@1),
+			    TYPE_SIGN (TREE_TYPE (@0)), &quot))
+     (cmp @0 { wide_int_to_tree (TREE_TYPE (@0), quot); })
+     { constant_boolean_node (cmp == NE_EXPR, type); }))
+   (if (INTEGRAL_TYPE_P (TREE_TYPE (@0))
+	&& TYPE_OVERFLOW_WRAPS (TREE_TYPE (@0))
+	&& (wi::bit_and (wi::to_wide (@1), 1) == 1))
+    (cmp @0
+     {
+       tree itype = TREE_TYPE (@0);
+       int p = TYPE_PRECISION (itype);
+       wide_int m = wi::one (p + 1) << p;
+       wide_int a = wide_int::from (wi::to_wide (@1), p + 1, UNSIGNED);
+       wide_int i = wide_int::from (wi::mod_inv (a, m),
+				    p, TYPE_SIGN (itype));
+       wide_int_to_tree (itype, wi::mul (i, wi::to_wide (@2)));
+     })))))
+
 /* Simplify comparison of something with itself.  For IEEE
    floating-point, we can only do some of these simplifications.  */
 (for cmp (eq ge le)
diff -Nurp a/gcc/testsuite/gcc.c-torture/execute/pr23135.c b/gcc/testsuite/gcc.c-torture/execute/pr23135.c
--- a/gcc/testsuite/gcc.c-torture/execute/pr23135.c	2021-03-17 16:34:24.016000000 +0800
+++ b/gcc/testsuite/gcc.c-torture/execute/pr23135.c	2021-03-17 10:30:13.572000000 +0800
@@ -1,7 +1,7 @@
 /* Based on execute/simd-1.c, modified by joern.rennecke@st.com to
    trigger a reload bug.  Verified for gcc mainline from 20050722 13:00 UTC
    for sh-elf -m4 -O2.  */
-/* { dg-options "-Wno-psabi" } */
+/* { dg-options "-Wno-psabi -fwrapv" } */
 /* { dg-add-options stack_size } */
 
 #ifndef STACK_SIZE
diff -Nurp a/gcc/testsuite/gcc.dg/tree-ssa/pr95433-2.c b/gcc/testsuite/gcc.dg/tree-ssa/pr95433-2.c
--- a/gcc/testsuite/gcc.dg/tree-ssa/pr95433-2.c	1970-01-01 08:00:00.000000000 +0800
+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr95433-2.c	2021-03-17 10:30:13.276000000 +0800
@@ -0,0 +1,15 @@
+/* { dg-do compile } */
+/* { dg-options "-O -fwrapv -fdump-tree-gimple" } */
+
+typedef __INT32_TYPE__ int32_t;
+typedef unsigned __INT32_TYPE__ uint32_t;
+
+int e(int32_t x){return 3*x==5;}
+int f(int32_t x){return 3*x==-5;}
+int g(int32_t x){return -3*x==5;}
+int h(int32_t x){return 7*x==3;}
+int i(uint32_t x){return 7*x==3;}
+
+/* { dg-final { scan-tree-dump-times "== 1431655767" 1 "gimple" } } */
+/* { dg-final { scan-tree-dump-times "== -1431655767" 2 "gimple" } } */
+/* { dg-final { scan-tree-dump-times "== 613566757" 2 "gimple" } } */
diff -Nurp a/gcc/testsuite/gcc.dg/tree-ssa/pr95433.c b/gcc/testsuite/gcc.dg/tree-ssa/pr95433.c
--- a/gcc/testsuite/gcc.dg/tree-ssa/pr95433.c	1970-01-01 08:00:00.000000000 +0800
+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr95433.c	2021-03-17 10:30:13.276000000 +0800
@@ -0,0 +1,8 @@
+/* { dg-do compile } */
+/* { dg-options "-O -fdump-tree-optimized" } */
+
+int f(int x){return x*7==17;}
+int g(int x){return x*3==15;}
+
+/* { dg-final { scan-tree-dump "return 0;" "optimized" } } */
+/* { dg-final { scan-tree-dump "== 5;" "optimized" } } */
diff -Nurp a/gcc/wide-int.cc b/gcc/wide-int.cc
--- a/gcc/wide-int.cc	2021-03-17 16:34:24.488000000 +0800
+++ b/gcc/wide-int.cc	2021-03-17 10:30:11.500000000 +0800
@@ -2223,6 +2223,39 @@ wi::round_up_for_mask (const wide_int &v
   return (val | tmp) & -tmp;
 }
 
+/* Compute the modular multiplicative inverse of A modulo B
+   using extended Euclid's algorithm.  Assumes A and B are coprime,
+   and that A and B have the same precision.  */
+wide_int
+wi::mod_inv (const wide_int &a, const wide_int &b)
+{
+  /* Verify the assumption.  */
+  gcc_checking_assert (wi::eq_p (wi::gcd (a, b), 1));
+
+  unsigned int p = a.get_precision () + 1;
+  gcc_checking_assert (b.get_precision () + 1 == p);
+  wide_int c = wide_int::from (a, p, UNSIGNED);
+  wide_int d = wide_int::from (b, p, UNSIGNED);
+  wide_int x0 = wide_int::from (0, p, UNSIGNED);
+  wide_int x1 = wide_int::from (1, p, UNSIGNED);
+
+  if (wi::eq_p (b, 1))
+    return wide_int::from (1, p, UNSIGNED);
+
+  while (wi::gt_p (c, 1, UNSIGNED))
+    {
+      wide_int t = d;
+      wide_int q = wi::divmod_trunc (c, d, UNSIGNED, &d);
+      c = t;
+      wide_int s = x0;
+      x0 = wi::sub (x1, wi::mul (q, x0));
+      x1 = s;
+    }
+  if (wi::lt_p (x1, 0, SIGNED))
+    x1 += d;
+  return x1;
+}
+
 /*
  * Private utilities.
  */
diff -Nurp a/gcc/wide-int.h b/gcc/wide-int.h
--- a/gcc/wide-int.h	2021-03-17 16:34:14.792000000 +0800
+++ b/gcc/wide-int.h	2021-03-17 10:30:11.500000000 +0800
@@ -3368,6 +3368,8 @@ namespace wi
   wide_int round_down_for_mask (const wide_int &, const wide_int &);
   wide_int round_up_for_mask (const wide_int &, const wide_int &);
 
+  wide_int mod_inv (const wide_int &a, const wide_int &b);
+
   template <typename T>
   T mask (unsigned int, bool);