fold-const: Ignore padding bits in native_interpret_expr REAL_CST reverse verificatio...

In the following testcase we try to std::bit_cast a (pair of) integral
value(s) which has some non-zero bits in the place of x86 long double
(for 64-bit 16 byte type with 10 bytes actually loaded/stored by hw,
for 32-bit 12 byte) and starting with my PR104522 change we reject that
as native_interpret_expr fails on it.  The PR104522 change extends what
has been done before for MODE_COMPOSITE_P (but those don't have any padding
bits) to all floating point types, because e.g. the exact x86 long double
has various bit combinations we don't support, like
pseudo-(denormals,infinities,NaNs) or unnormals.  The HW handles some of
those as exceptional cases and others similarly to the non-pseudo ones.
But for the padding bits it actually doesn't load/store those bits at all,
it loads/stores 10 bytes.  So, I think we should exempt the padding bits
from the reverse comparison (the native_encode_expr bits for the padding
will be all zeros), which the following patch does.  For bit_cast it is
similar to e.g. ignoring padding bits if the destination is a structure
which has padding bits in there.

The change changed auto-init-4.c to how it has been behaving before the
PR105259 change, where some more VCEs can be now done.

2023-03-02  Jakub Jelinek  <jakub@redhat.com>

	PR c++/108934
	* fold-const.cc (native_interpret_expr) <case REAL_CST>: Before memcmp
	comparison copy the bytes from ptr to a temporary buffer and clearing
	padding bits in there.

	* gcc.target/i386/auto-init-4.c: Revert PR105259 change.
	* g++.target/i386/pr108934.C: New test.

(cherry picked from commit cc88366a80e35b3e53141f49d3071010ff3c2ef8)

diff --git a/gcc/fold-const.cc b/gcc/fold-const.cc
index eec4c1820ed875ecc2330b59f22d19f816360abe..d7923974c8530c3be7fa1742df84815a5c363719 100644 (file)
--- a/gcc/fold-const.cc
+++ b/gcc/fold-const.cc
@@ -8839,11 +8839,13 @@ native_interpret_expr (tree type, const unsigned char *ptr, int len)
             valid values that GCC can't really represent accurately.
             See PR95450.  Even for other modes, e.g. x86 XFmode can have some
             bit combinationations which GCC doesn't preserve.  */
-         unsigned char buf[24];
+         unsigned char buf[24 * 2];
          scalar_float_mode mode = SCALAR_FLOAT_TYPE_MODE (type);
          int total_bytes = GET_MODE_SIZE (mode);
+         memcpy (buf + 24, ptr, total_bytes);
+         clear_type_padding_in_mask (type, buf + 24);
          if (native_encode_expr (ret, buf, total_bytes, 0) != total_bytes
-             || memcmp (ptr, buf, total_bytes) != 0)
+             || memcmp (buf + 24, buf, total_bytes) != 0)
            return NULL_TREE;
          return ret;
        }

diff --git a/gcc/testsuite/g++.target/i386/pr108934.C b/gcc/testsuite/g++.target/i386/pr108934.C
new file mode 100644 (file)
index 0000000..bd8f0ff
--- /dev/null
+++ b/gcc/testsuite/g++.target/i386/pr108934.C
@@ -0,0 +1,28 @@
+// PR c++/108934
+// { dg-do compile { target c++11 } }
+
+struct S { unsigned long long a[2]; };
+struct T { unsigned long long b[6]; };
+struct U { unsigned long long c[2]; long double d; unsigned long long e[2]; };
+
+#if __SIZEOF_LONG_DOUBLE__ == 16 && __LDBL_MANT_DIG__ == 64 && __SIZEOF_LONG_LONG__ == 8
+constexpr long double
+foo (S x)
+{
+  return __builtin_bit_cast (long double, x);
+}
+
+constexpr S a = { 0ULL, 0xffffffffffff0000ULL };
+constexpr long double b = foo (a);
+static_assert (b == 0.0L, "");
+
+constexpr U
+bar (T x)
+{
+  return __builtin_bit_cast (U, x);
+}
+
+constexpr T c = { 0ULL, 0ULL, 0ULL, 0xffffffffffff0000ULL, 0ULL, 0ULL };
+constexpr U d = bar (c);
+static_assert (d.d == 0.0L, "");
+#endif

diff --git a/gcc/testsuite/gcc.target/i386/auto-init-4.c b/gcc/testsuite/gcc.target/i386/auto-init-4.c
index 5b4fd870c36c1b254f71c98efa49cd90df45ae95..d9af8f30bb05a5adf8b8f2be26060136bc170504 100644 (file)
--- a/gcc/testsuite/gcc.target/i386/auto-init-4.c
+++ b/gcc/testsuite/gcc.target/i386/auto-init-4.c
@@ -15,6 +15,5 @@ long double foo()
 }
 
 
-/* The long double init isn't expanded optimally, see PR105259.  For ia32
-   it uses zero-initialization.  */
-/* { dg-final { scan-assembler-times "long\t-16843010" 3 } } */
+/* { dg-final { scan-assembler-times "long\t-16843010" 5  { target { ! ia32 } } } } */
+/* { dg-final { scan-assembler-times "long\t-16843010" 3  { target { ia32 } } } } */