diff mbox series

correct memcmp expansion of constant representations containing embedded nuls (PR 95189)

Message ID 546f1c1e-73f3-aa20-a3be-b7bae03fff55@gmail.com
State New
Headers show
Series correct memcmp expansion of constant representations containing embedded nuls (PR 95189) | expand

Commit Message

Martin Sebor July 1, 2020, 12:23 a.m. UTC
An enhancement to GCC 10 to improve the expansion of strncmp
calls with strings with embedded nuls introduced a regression
in similar calls to memcmp.  A review of the changes that led
up to the regression exposed a number of questionable choices
that likely conspired to cause the bug.

For example, the name of the function with both the strncmp
enhancement as well as the memcmp bug is
inline_expand_builtin_string_cmp().  It's easy to assume that
the function handles calls to strcmp and strncmp but not also
memcmp.

Another similar example is the name of the second c_getstr()
argument -- strlen -- that doesn't actually store the length
of the retrieved string but rather its size in bytes
(including any embedded nuls, but excluding those appended
implicitly to zero out the remainder of an array the string
is stored in, up to the array's size).

Yet another example of a dubious choice is string_constant()
returning the empty string (i.e., STRING_CST with size 1) for
zero initializers of constants of any type (as opposed to one
of the same size as the constant object).

Besides fixing the memcmp bug the attached patch (hopefully)
also rectifies some of the otherwise more or less benign
mistakes that precipitated it, mostly by clarifying comments
and changing misleading names of functions, their arguments,
or local variables.

A happy consequence of the fixes is that they improve codegen
for calls to memcpy with constants whose representation includes
embedded nuls.

Tested on x86_64-linux.

Martin

Comments

Martin Sebor July 7, 2020, 8:02 p.m. UTC | #1
Ping: https://gcc.gnu.org/pipermail/gcc-patches/2020-July/549225.html

On 6/30/20 6:23 PM, Martin Sebor wrote:
> An enhancement to GCC 10 to improve the expansion of strncmp
> calls with strings with embedded nuls introduced a regression
> in similar calls to memcmp.  A review of the changes that led
> up to the regression exposed a number of questionable choices
> that likely conspired to cause the bug.
> 
> For example, the name of the function with both the strncmp
> enhancement as well as the memcmp bug is
> inline_expand_builtin_string_cmp().  It's easy to assume that
> the function handles calls to strcmp and strncmp but not also
> memcmp.
> 
> Another similar example is the name of the second c_getstr()
> argument -- strlen -- that doesn't actually store the length
> of the retrieved string but rather its size in bytes
> (including any embedded nuls, but excluding those appended
> implicitly to zero out the remainder of an array the string
> is stored in, up to the array's size).
> 
> Yet another example of a dubious choice is string_constant()
> returning the empty string (i.e., STRING_CST with size 1) for
> zero initializers of constants of any type (as opposed to one
> of the same size as the constant object).
> 
> Besides fixing the memcmp bug the attached patch (hopefully)
> also rectifies some of the otherwise more or less benign
> mistakes that precipitated it, mostly by clarifying comments
> and changing misleading names of functions, their arguments,
> or local variables.
> 
> A happy consequence of the fixes is that they improve codegen
> for calls to memcpy with constants whose representation includes
> embedded nuls.
> 
> Tested on x86_64-linux.
> 
> Martin
>
Martin Sebor July 14, 2020, 12:06 a.m. UTC | #2
Ping: https://gcc.gnu.org/pipermail/gcc-patches/2020-July/549225.html

On 7/7/20 2:02 PM, Martin Sebor wrote:
> Ping: https://gcc.gnu.org/pipermail/gcc-patches/2020-July/549225.html
> 
> On 6/30/20 6:23 PM, Martin Sebor wrote:
>> An enhancement to GCC 10 to improve the expansion of strncmp
>> calls with strings with embedded nuls introduced a regression
>> in similar calls to memcmp.  A review of the changes that led
>> up to the regression exposed a number of questionable choices
>> that likely conspired to cause the bug.
>>
>> For example, the name of the function with both the strncmp
>> enhancement as well as the memcmp bug is
>> inline_expand_builtin_string_cmp().  It's easy to assume that
>> the function handles calls to strcmp and strncmp but not also
>> memcmp.
>>
>> Another similar example is the name of the second c_getstr()
>> argument -- strlen -- that doesn't actually store the length
>> of the retrieved string but rather its size in bytes
>> (including any embedded nuls, but excluding those appended
>> implicitly to zero out the remainder of an array the string
>> is stored in, up to the array's size).
>>
>> Yet another example of a dubious choice is string_constant()
>> returning the empty string (i.e., STRING_CST with size 1) for
>> zero initializers of constants of any type (as opposed to one
>> of the same size as the constant object).
>>
>> Besides fixing the memcmp bug the attached patch (hopefully)
>> also rectifies some of the otherwise more or less benign
>> mistakes that precipitated it, mostly by clarifying comments
>> and changing misleading names of functions, their arguments,
>> or local variables.
>>
>> A happy consequence of the fixes is that they improve codegen
>> for calls to memcpy with constants whose representation includes
>> embedded nuls.
>>
>> Tested on x86_64-linux.
>>
>> Martin
>>
>
Martin Sebor July 20, 2020, 6:43 p.m. UTC | #3
I have committed this change in r11-2231 after Jeff approved it
off list last Thursday.

On 6/30/20 6:23 PM, Martin Sebor wrote:
> An enhancement to GCC 10 to improve the expansion of strncmp
> calls with strings with embedded nuls introduced a regression
> in similar calls to memcmp.  A review of the changes that led
> up to the regression exposed a number of questionable choices
> that likely conspired to cause the bug.
> 
> For example, the name of the function with both the strncmp
> enhancement as well as the memcmp bug is
> inline_expand_builtin_string_cmp().  It's easy to assume that
> the function handles calls to strcmp and strncmp but not also
> memcmp.
> 
> Another similar example is the name of the second c_getstr()
> argument -- strlen -- that doesn't actually store the length
> of the retrieved string but rather its size in bytes
> (including any embedded nuls, but excluding those appended
> implicitly to zero out the remainder of an array the string
> is stored in, up to the array's size).
> 
> Yet another example of a dubious choice is string_constant()
> returning the empty string (i.e., STRING_CST with size 1) for
> zero initializers of constants of any type (as opposed to one
> of the same size as the constant object).
> 
> Besides fixing the memcmp bug the attached patch (hopefully)
> also rectifies some of the otherwise more or less benign
> mistakes that precipitated it, mostly by clarifying comments
> and changing misleading names of functions, their arguments,
> or local variables.
> 
> A happy consequence of the fixes is that they improve codegen
> for calls to memcpy with constants whose representation includes
> embedded nuls.
> 
> Tested on x86_64-linux.
> 
> Martin
>
Rainer Orth July 22, 2020, 8:23 a.m. UTC | #4
Hi Martin,

> I have committed this change in r11-2231 after Jeff approved it
> off list last Thursday.

the new gcc.target/i386/memcpy-pr95886.c test FAILs on 32-bit x86
(i386-pc-solaris2.11):

+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 1976943448883713" 1
+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 576467370915332609" 1
+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578431098682540545" 1
+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752110593" 1
+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752306689" 1
+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307200" 1
+FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307201" 2

	Rainer
Iain Sandoe July 22, 2020, 8:36 a.m. UTC | #5
Hi Martin,

Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE> wrote:
>>
>> I have committed this change in r11-2231 after Jeff approved it
>> off list last Thursday.
>
> the new gcc.target/i386/memcpy-pr95886.c test FAILs on 32-bit x86
> (i386-pc-solaris2.11):
>
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 1976943448883713" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 576467370915332609" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 578431098682540545" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 578437695685198337" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 578437695685198337" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 578437695752110593" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 578437695752306689" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 578437695752307200" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand  
> "const_int 578437695752307201” 2

likewise, the tests fail on m32 Darwin,
thanks

Iain
Martin Sebor July 23, 2020, 8:12 p.m. UTC | #6
On 7/22/20 2:23 AM, Rainer Orth wrote:
> Hi Martin,
> 
>> I have committed this change in r11-2231 after Jeff approved it
>> off list last Thursday.
> 
> the new gcc.target/i386/memcpy-pr95886.c test FAILs on 32-bit x86
> (i386-pc-solaris2.11):
> 
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 1976943448883713" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 576467370915332609" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578431098682540545" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752110593" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752306689" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307200" 1
> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307201" 2

Thanks for letting me know.  The test looks for patterns that are
apparently LP64-specific so I restricted it to just that data model.

Martin
H.J. Lu July 23, 2020, 8:18 p.m. UTC | #7
On Thu, Jul 23, 2020 at 1:14 PM Martin Sebor via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> On 7/22/20 2:23 AM, Rainer Orth wrote:
> > Hi Martin,
> >
> >> I have committed this change in r11-2231 after Jeff approved it
> >> off list last Thursday.
> >
> > the new gcc.target/i386/memcpy-pr95886.c test FAILs on 32-bit x86
> > (i386-pc-solaris2.11):
> >
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 1976943448883713" 1
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 576467370915332609" 1
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578431098682540545" 1
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752110593" 1
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752306689" 1
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307200" 1
> > +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307201" 2
>
> Thanks for letting me know.  The test looks for patterns that are
> apparently LP64-specific so I restricted it to just that data model.
>

Shouldn't it also work for x32?
Martin Sebor July 23, 2020, 11:50 p.m. UTC | #8
On 7/23/20 2:18 PM, H.J. Lu wrote:
> On Thu, Jul 23, 2020 at 1:14 PM Martin Sebor via Gcc-patches
> <gcc-patches@gcc.gnu.org> wrote:
>>
>> On 7/22/20 2:23 AM, Rainer Orth wrote:
>>> Hi Martin,
>>>
>>>> I have committed this change in r11-2231 after Jeff approved it
>>>> off list last Thursday.
>>>
>>> the new gcc.target/i386/memcpy-pr95886.c test FAILs on 32-bit x86
>>> (i386-pc-solaris2.11):
>>>
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 1976943448883713" 1
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 576467370915332609" 1
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578431098682540545" 1
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752110593" 1
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752306689" 1
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307200" 1
>>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307201" 2
>>
>> Thanks for letting me know.  The test looks for patterns that are
>> apparently LP64-specific so I restricted it to just that data model.
>>
> 
> Shouldn't it also work for x32?

I would expect the optimization to work on any target that does
the piecemeal copy, including x32.  But you probably meant if
the test should pass as is on x32.  Possibly yes, but I didn't
check.

A better test would exercise the solution at least on all three
i386 targets.  It could probably be done fairly simply by reducing
the sizes of the arrays, or by using the large arrays only on LP64
or whatever makes the difference and figuring out the right magic
target selector to use in the dg- directives.

Martin
H.J. Lu July 24, 2020, 2:26 a.m. UTC | #9
On Thu, Jul 23, 2020 at 4:50 PM Martin Sebor <msebor@gmail.com> wrote:
>
> On 7/23/20 2:18 PM, H.J. Lu wrote:
> > On Thu, Jul 23, 2020 at 1:14 PM Martin Sebor via Gcc-patches
> > <gcc-patches@gcc.gnu.org> wrote:
> >>
> >> On 7/22/20 2:23 AM, Rainer Orth wrote:
> >>> Hi Martin,
> >>>
> >>>> I have committed this change in r11-2231 after Jeff approved it
> >>>> off list last Thursday.
> >>>
> >>> the new gcc.target/i386/memcpy-pr95886.c test FAILs on 32-bit x86
> >>> (i386-pc-solaris2.11):
> >>>
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 1976943448883713" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 576467370915332609" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578431098682540545" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695685198337" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752110593" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752306689" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307200" 1
> >>> +FAIL: gcc.target/i386/memcpy-pr95886.c scan-rtl-dump-times expand "const_int 578437695752307201" 2
> >>
> >> Thanks for letting me know.  The test looks for patterns that are
> >> apparently LP64-specific so I restricted it to just that data model.
> >>
> >
> > Shouldn't it also work for x32?
>
> I would expect the optimization to work on any target that does
> the piecemeal copy, including x32.  But you probably meant if
> the test should pass as is on x32.  Possibly yes, but I didn't
> check.
>
> A better test would exercise the solution at least on all three
> i386 targets.  It could probably be done fairly simply by reducing
> the sizes of the arrays, or by using the large arrays only on LP64
> or whatever makes the difference and figuring out the right magic
> target selector to use in the dg- directives.

Here is the patch I am checking in.
diff mbox series

Patch

Correct handling of constant representations containing embedded nuls.

Resolves:
PR middle-end/95189 - memcmp being wrongly stripped like strcm
PR middle-end/95886 - suboptimal memcpy with embedded zero bytes

gcc/ChangeLog:

	PR middle-end/95189
	PR middle-end/95886
	* builtins.c (inline_expand_builtin_string_cmp): Rename...
	(inline_expand_builtin_bytecmp): ...to this.
	(builtin_memcpy_read_str): Don't expect data to be nul-terminated.
	(expand_builtin_memory_copy_args): Handle object representations
	with embedded nul bytes.
	(expand_builtin_memcmp): Same.
	(expand_builtin_strcmp): Adjust call to naming change.
	(expand_builtin_strncmp): Same.
	* expr.c (string_constant): Create empty strings with nonzero size.
	* fold-const.c (c_getstr): Rename locals and update comments.
	* tree.c (build_string): Accept null pointer argument.
	(build_string_literal): Same.
	* tree.h (build_string): Provide a default.
	(build_string_literal): Same.

gcc/testsuite/ChangeLog:

	PR middle-end/95189
	PR middle-end/95886
	* gcc.dg/memcmp-pr95189.c: New test.
	* gcc/testsuite/gcc.dg/strncmp-3.c: New test.
	* gcc.target/i386/memcpy-pr95886.c: New test.

diff --git a/gcc/builtins.c b/gcc/builtins.c
index 4754602e0ec..2279b6bfdf7 100644
--- a/gcc/builtins.c
+++ b/gcc/builtins.c
@@ -119,7 +119,7 @@  static rtx expand_builtin_next_arg (void);
 static rtx expand_builtin_va_start (tree);
 static rtx expand_builtin_va_end (tree);
 static rtx expand_builtin_va_copy (tree);
-static rtx inline_expand_builtin_string_cmp (tree, rtx);
+static rtx inline_expand_builtin_bytecmp (tree, rtx);
 static rtx expand_builtin_strcmp (tree, rtx);
 static rtx expand_builtin_strncmp (tree, rtx, machine_mode);
 static rtx builtin_memcpy_read_str (void *, HOST_WIDE_INT, scalar_int_mode);
@@ -3227,20 +3227,18 @@  expand_builtin_strnlen (tree exp, rtx target, machine_mode target_mode)
 }
 
 /* Callback routine for store_by_pieces.  Read GET_MODE_BITSIZE (MODE)
-   bytes from constant string DATA + OFFSET and return it as target
-   constant.  */
+   bytes from bytes at DATA + OFFSET and return it reinterpreted as
+   a target constant.  */
 
 static rtx
 builtin_memcpy_read_str (void *data, HOST_WIDE_INT offset,
 			 scalar_int_mode mode)
 {
-  const char *str = (const char *) data;
+  /* The REPresentation pointed to by DATA need not be a nul-terminated
+     string but the caller guarantees it's large enough for MODE.  */
+  const char *rep = (const char *) data;
 
-  gcc_assert (offset >= 0
-	      && ((unsigned HOST_WIDE_INT) offset + GET_MODE_SIZE (mode)
-		  <= strlen (str) + 1));
-
-  return c_readstr (str + offset, mode);
+  return c_readstr (rep + offset, mode, /*nul_terminated=*/false);
 }
 
 /* LEN specify length of the block of memcpy/memset operation.
@@ -4411,7 +4409,6 @@  expand_builtin_memory_copy_args (tree dest, tree src, tree len,
 				 rtx target, tree exp, memop_ret retmode,
 				 bool might_overlap)
 {
-  const char *src_str;
   unsigned int src_align = get_pointer_alignment (src);
   unsigned int dest_align = get_pointer_alignment (dest);
   rtx dest_mem, src_mem, dest_addr, len_rtx;
@@ -4443,24 +4440,29 @@  expand_builtin_memory_copy_args (tree dest, tree src, tree len,
   len_rtx = expand_normal (len);
   determine_block_size (len, len_rtx, &min_size, &max_size,
 			&probable_max_size);
-  src_str = c_getstr (src);
-
-  /* If SRC is a string constant and block move would be done by
-     pieces, we can avoid loading the string from memory and only
-     stored the computed constants.  This works in the overlap
-     (memmove) case as well because store_by_pieces just generates a
-     series of stores of constants from the string constant returned
-     by c_getstr().  */
-  if (src_str
+
+  /* Try to get the byte representation of the constant SRC points to,
+     with its byte size in NBYTES.  */
+  unsigned HOST_WIDE_INT nbytes;
+  const char *rep = c_getstr (src, &nbytes);
+
+  /* If the function's constant bound LEN_RTX is less than or equal
+     to the byte size of the representation of the constant argument,
+     and if block move would be done by pieces, we can avoid loading
+     the bytes from memory and only store the computed constant.
+     This works in the overlap (memmove) case as well because
+     store_by_pieces just generates a series of stores of constants
+     from the representation returned by c_getstr().  */
+  if (rep
       && CONST_INT_P (len_rtx)
-      && (unsigned HOST_WIDE_INT) INTVAL (len_rtx) <= strlen (src_str) + 1
+      && (unsigned HOST_WIDE_INT) INTVAL (len_rtx) <= nbytes
       && can_store_by_pieces (INTVAL (len_rtx), builtin_memcpy_read_str,
-			      CONST_CAST (char *, src_str),
+			      CONST_CAST (char *, rep),
 			      dest_align, false))
     {
       dest_mem = store_by_pieces (dest_mem, INTVAL (len_rtx),
 				  builtin_memcpy_read_str,
-				  CONST_CAST (char *, src_str),
+				  CONST_CAST (char *, rep),
 				  dest_align, false, retmode);
       dest_mem = force_operand (XEXP (dest_mem, 0), target);
       dest_mem = convert_memory_address (ptr_mode, dest_mem);
@@ -4484,7 +4486,8 @@  expand_builtin_memory_copy_args (tree dest, tree src, tree len,
   dest_addr = emit_block_move_hints (dest_mem, src_mem, len_rtx, method,
 				     expected_align, expected_size,
 				     min_size, max_size, probable_max_size,
-				     use_mempcpy_call, &is_move_done, might_overlap);
+				     use_mempcpy_call, &is_move_done,
+				     might_overlap);
 
   /* Bail out when a mempcpy call would be expanded as libcall and when
      we have a target that provides a fast implementation
@@ -5319,7 +5322,7 @@  expand_builtin_memcmp (tree exp, rtx target, bool result_eq)
 
   if (!result_eq && fcode != BUILT_IN_BCMP)
     {
-      result = inline_expand_builtin_string_cmp (exp, target);
+      result = inline_expand_builtin_bytecmp (exp, target);
       if (result)
 	return result;
     }
@@ -5347,26 +5350,32 @@  expand_builtin_memcmp (tree exp, rtx target, bool result_eq)
 
   by_pieces_constfn constfn = NULL;
 
-  const char *src_str = c_getstr (arg2);
-  if (result_eq && src_str == NULL)
+  /* Try to get the byte representation of the constant ARG2 (or, only
+     when the function's result is used for equality to zero, ARG1)
+     points to, with its byte size in NBYTES.  */
+  unsigned HOST_WIDE_INT nbytes;
+  const char *rep = c_getstr (arg2, &nbytes);
+  if (result_eq && rep == NULL)
     {
-      src_str = c_getstr (arg1);
-      if (src_str != NULL)
+      /* For equality to zero the arguments are interchangeable.  */
+      rep = c_getstr (arg1, &nbytes);
+      if (rep != NULL)
 	std::swap (arg1_rtx, arg2_rtx);
     }
 
-  /* If SRC is a string constant and block move would be done
-     by pieces, we can avoid loading the string from memory
-     and only stored the computed constants.  */
-  if (src_str
+  /* If the function's constant bound LEN_RTX is less than or equal
+     to the byte size of the representation of the constant argument,
+     and if block move would be done by pieces, we can avoid loading
+     the bytes from memory and only store the computed constant result.  */
+  if (rep
       && CONST_INT_P (len_rtx)
-      && (unsigned HOST_WIDE_INT) INTVAL (len_rtx) <= strlen (src_str) + 1)
+      && (unsigned HOST_WIDE_INT) INTVAL (len_rtx) <= nbytes)
     constfn = builtin_memcpy_read_str;
 
   result = emit_block_cmp_hints (arg1_rtx, arg2_rtx, len_rtx,
 				 TREE_TYPE (len), target,
 				 result_eq, constfn,
-				 CONST_CAST (char *, src_str));
+				 CONST_CAST (char *, rep));
 
   if (result)
     {
@@ -5405,7 +5414,7 @@  expand_builtin_strcmp (tree exp, ATTRIBUTE_UNUSED rtx target)
 
   /* Due to the performance benefit, always inline the calls first.  */
   rtx result = NULL_RTX;
-  result = inline_expand_builtin_string_cmp (exp, target);
+  result = inline_expand_builtin_bytecmp (exp, target);
   if (result)
     return result;
 
@@ -5529,7 +5538,7 @@  expand_builtin_strncmp (tree exp, ATTRIBUTE_UNUSED rtx target,
 
   /* Due to the performance benefit, always inline the calls first.  */
   rtx result = NULL_RTX;
-  result = inline_expand_builtin_string_cmp (exp, target);
+  result = inline_expand_builtin_bytecmp (exp, target);
   if (result)
     return result;
 
@@ -7741,18 +7750,18 @@  inline_string_cmp (rtx target, tree var_str, const char *const_str,
   return result;
 }
 
-/* Inline expansion a call to str(n)cmp, with result going to
-   TARGET if that's convenient.
+/* Inline expansion of a call to str(n)cmp and memcmp, with result going
+   to TARGET if that's convenient.
    If the call is not been inlined, return NULL_RTX.  */
+
 static rtx
-inline_expand_builtin_string_cmp (tree exp, rtx target)
+inline_expand_builtin_bytecmp (tree exp, rtx target)
 {
   tree fndecl = get_callee_fndecl (exp);
   enum built_in_function fcode = DECL_FUNCTION_CODE (fndecl);
-  unsigned HOST_WIDE_INT length = 0;
   bool is_ncmp = (fcode == BUILT_IN_STRNCMP || fcode == BUILT_IN_MEMCMP);
 
-  /* Do NOT apply this inlining expansion when optimizing for size or 
+  /* Do NOT apply this inlining expansion when optimizing for size or
      optimization level below 2.  */
   if (optimize < 2 || optimize_insn_for_size_p ())
     return NULL_RTX;
@@ -7775,29 +7784,47 @@  inline_expand_builtin_string_cmp (tree exp, rtx target)
   unsigned HOST_WIDE_INT len2 = 0;
   unsigned HOST_WIDE_INT len3 = 0;
 
-  const char *src_str1 = c_getstr (arg1, &len1);
-  const char *src_str2 = c_getstr (arg2, &len2);
+  /* Get the object representation of the initializers of ARG1 and ARG2
+     as strings, provided they refer to constant objects, with their byte
+     sizes in LEN1 and LEN2, respectively.  */
+  const char *bytes1 = c_getstr (arg1, &len1);
+  const char *bytes2 = c_getstr (arg2, &len2);
 
-  /* If neither strings is constant string, the call is not qualify.  */
-  if (!src_str1 && !src_str2)
+  /* Fail if neither argument refers to an initialized constant.  */
+  if (!bytes1 && !bytes2)
     return NULL_RTX;
 
-  /* For strncmp, if the length is not a const, not qualify.  */
   if (is_ncmp)
     {
+      /* Fail if the memcmp/strncmp bound is not a constant.  */
       if (!tree_fits_uhwi_p (len3_tree))
 	return NULL_RTX;
-      else
-	len3 = tree_to_uhwi (len3_tree);
-    }
 
-  if (src_str1 != NULL)
-    len1 = strnlen (src_str1, len1) + 1;
+      len3 = tree_to_uhwi (len3_tree);
 
-  if (src_str2 != NULL)
-    len2 = strnlen (src_str2, len2) + 1;
+      if (fcode == BUILT_IN_MEMCMP)
+	{
+	  /* Fail if the memcmp bound is greater than the size of either
+	     of the two constant objects.  */
+	  if ((bytes1 && len1 < len3)
+	      || (bytes2 && len2 < len3))
+	    return NULL_RTX;
+	}
+    }
 
-  int const_str_n = 0;
+  if (fcode != BUILT_IN_MEMCMP)
+    {
+      /* For string functions (i.e., strcmp and strncmp) reduce LEN1
+	 and LEN2 to the length of the nul-terminated string stored
+	 in each.  */
+      if (bytes1 != NULL)
+	len1 = strnlen (bytes1, len1) + 1;
+      if (bytes2 != NULL)
+	len2 = strnlen (bytes2, len2) + 1;
+    }
+
+  /* See inline_string_cmp.  */
+  int const_str_n;
   if (!len1)
     const_str_n = 2;
   else if (!len2)
@@ -7807,23 +7834,23 @@  inline_expand_builtin_string_cmp (tree exp, rtx target)
   else
     const_str_n = 2;
 
-  gcc_checking_assert (const_str_n > 0);
-  length = (const_str_n == 1) ? len1 : len2;
-
-  if (is_ncmp && len3 < length)
-    length = len3;
+  /* For strncmp only, compute the new bound as the smallest of
+     the lengths of the two strings (plus 1) and the bound provided
+     to the function.  */
+  unsigned HOST_WIDE_INT bound = (const_str_n == 1) ? len1 : len2;
+  if (is_ncmp && len3 < bound)
+    bound = len3;
 
-  /* If the length of the comparision is larger than the threshold,
+  /* If the bound of the comparison is larger than the threshold,
      do nothing.  */
-  if (length > (unsigned HOST_WIDE_INT)
-	       param_builtin_string_cmp_inline_length)
+  if (bound > (unsigned HOST_WIDE_INT) param_builtin_string_cmp_inline_length)
     return NULL_RTX;
 
   machine_mode mode = TYPE_MODE (TREE_TYPE (exp));
 
   /* Now, start inline expansion the call.  */
   return inline_string_cmp (target, (const_str_n == 1) ? arg2 : arg1,
-			    (const_str_n == 1) ? src_str1 : src_str2, length,
+			    (const_str_n == 1) ? bytes1 : bytes2, bound,
 			    const_str_n, mode);
 }
 
diff --git a/gcc/expr.c b/gcc/expr.c
index 3c68b0d754c..9b8be1c87bd 100644
--- a/gcc/expr.c
+++ b/gcc/expr.c
@@ -11822,12 +11822,12 @@  string_constant (tree arg, tree *ptr_offset, tree *mem_size, tree *decl)
       while (TREE_CODE (chartype) == ARRAY_TYPE)
 	chartype = TREE_TYPE (chartype);
       /* Convert a char array to an empty STRING_CST having an array
-	 of the expected type.  */
+	 of the expected type and size.  */
       if (!initsize)
 	  initsize = integer_zero_node;
 
       unsigned HOST_WIDE_INT size = tree_to_uhwi (initsize);
-      init = build_string_literal (size ? 1 : 0, "", chartype, size);
+      init = build_string_literal (size, NULL, chartype, size);
       init = TREE_OPERAND (init, 0);
       init = TREE_OPERAND (init, 0);
 
diff --git a/gcc/fold-const.c b/gcc/fold-const.c
index 67a379f4f2b..553b055a8a3 100644
--- a/gcc/fold-const.c
+++ b/gcc/fold-const.c
@@ -15379,24 +15379,29 @@  fold_build_pointer_plus_hwi_loc (location_t loc, tree ptr, HOST_WIDE_INT off)
 			  ptr, size_int (off));
 }
 
-/* Return a pointer P to a NUL-terminated string representing the sequence
-   of constant characters referred to by SRC (or a subsequence of such
-   characters within it if SRC is a reference to a string plus some
-   constant offset).  If STRLEN is non-null, store the number of bytes
-   in the string constant including the terminating NUL char.  *STRLEN is
-   typically strlen(P) + 1 in the absence of embedded NUL characters.  */
+/* Return a pointer P to a NUL-terminated string containing the sequence
+   of bytes corresponding to the representation of the object referred to
+   by SRC (or a subsequence of such bytes within it if SRC is a reference
+   to an initialized constant array plus some constant offset).
+   If STRSIZE is non-null, store the number of bytes in the constant
+   sequence including the terminating NUL byte.  *STRSIZE is equal to
+   sizeof(A) - OFFSET where A is the array that stores the constant
+   sequence that SRC points to and OFFSET is the byte offset of SRC from
+   the beginning of A.  SRC need not point to a string or even an array
+   of characters but may point to an object of any type.  */
 
 const char *
-c_getstr (tree src, unsigned HOST_WIDE_INT *strlen /* = NULL */)
+c_getstr (tree src, unsigned HOST_WIDE_INT *strsize /* = NULL */)
 {
+  /* The offset into the array A storing the string, and A's byte size.  */
   tree offset_node;
   tree mem_size;
 
-  if (strlen)
-    *strlen = 0;
+  if (strsize)
+    *strsize = 0;
 
   src = string_constant (src, &offset_node, &mem_size, NULL);
-  if (src == 0)
+  if (!src)
     return NULL;
 
   unsigned HOST_WIDE_INT offset = 0;
@@ -15411,34 +15416,44 @@  c_getstr (tree src, unsigned HOST_WIDE_INT *strlen /* = NULL */)
   if (!tree_fits_uhwi_p (mem_size))
     return NULL;
 
-  /* STRING_LENGTH is the size of the string literal, including any
-     embedded NULs.  STRING_SIZE is the size of the array the string
-     literal is stored in.  */
-  unsigned HOST_WIDE_INT string_length = TREE_STRING_LENGTH (src);
-  unsigned HOST_WIDE_INT string_size = tree_to_uhwi (mem_size);
+  /* ARRAY_SIZE is the byte size of the array the constant sequence
+     is stored in and equal to sizeof A.  INIT_BYTES is the number
+     of bytes in the constant sequence used to initialize the array,
+     including any embedded NULs as well as the terminating NUL (for
+     strings), but not including any trailing zeros/NULs past
+     the terminating one appended implicitly to a string literal to
+     zero out the remainder of the array it's stored in.  For example,
+     given:
+       const char a[7] = "abc\0d";
+       n = strlen (a + 1);
+     ARRAY_SIZE is 7, INIT_BYTES is 6, and OFFSET is 1.  For a valid
+     (i.e., nul-terminated) string with no embedded nuls, INIT_BYTES
+     is equal to strlen (A) + 1.  */
+  const unsigned HOST_WIDE_INT array_size = tree_to_uhwi (mem_size);
+  unsigned HOST_WIDE_INT init_bytes = TREE_STRING_LENGTH (src);
 
   /* Ideally this would turn into a gcc_checking_assert over time.  */
-  if (string_length > string_size)
-    string_length = string_size;
+  if (init_bytes > array_size)
+    init_bytes = array_size;
 
   const char *string = TREE_STRING_POINTER (src);
 
   /* Ideally this would turn into a gcc_checking_assert over time.  */
-  if (string_length > string_size)
-    string_length = string_size;
+  if (init_bytes > array_size)
+    init_bytes = array_size;
 
-  if (string_length == 0
-      || offset >= string_size)
+  if (init_bytes == 0 || offset >= array_size)
     return NULL;
 
-  if (strlen)
+  if (strsize)
     {
-      /* Compute and store the length of the substring at OFFSET.
-	 All offsets past the initial length refer to null strings.  */
-      if (offset < string_length)
-	*strlen = string_length - offset;
+      /* Compute and store the number of characters from the beginning
+	 of the substring at OFFSET to the end, including the terminating
+	 nul.  Offsets past the initial length refer to null strings.  */
+      if (offset < init_bytes)
+	*strsize = init_bytes - offset;
       else
-	*strlen = 1;
+	*strsize = 1;
     }
   else
     {
@@ -15446,11 +15461,11 @@  c_getstr (tree src, unsigned HOST_WIDE_INT *strlen /* = NULL */)
       /* Support only properly NUL-terminated single byte strings.  */
       if (tree_to_uhwi (TYPE_SIZE_UNIT (eltype)) != 1)
 	return NULL;
-      if (string[string_length - 1] != '\0')
+      if (string[init_bytes - 1] != '\0')
 	return NULL;
     }
 
-  return offset < string_length ? string + offset : "";
+  return offset < init_bytes ? string + offset : "";
 }
 
 /* Given a tree T, compute which bits in T may be nonzero.  */
diff --git a/gcc/testsuite/gcc.dg/memcmp-pr95189.c b/gcc/testsuite/gcc.dg/memcmp-pr95189.c
new file mode 100644
index 00000000000..d8250ecfc98
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/memcmp-pr95189.c
@@ -0,0 +1,28 @@ 
+/* PR middle-end/95189 - memcmp being wrongly stripped like strcmp
+   { dg-do run }
+   { dg-options "-O2 -Wall" } */
+
+char a4[] = "\0abc";
+char a8[] = "\0abcdefg";
+char a16[] = "\0abcdefghijklmno";
+
+int cmp4 (void)
+{
+  return __builtin_memcmp (a4, "\0\0\0\0", 4);
+}
+
+int cmp8 (void)
+{
+  return __builtin_memcmp (a8, "\0\0\0\0\0\0\0\0", 8);
+}
+
+int cmp16 (void)
+{
+  return __builtin_memcmp (a16, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0", 16);
+}
+
+int main (void)
+{
+  if (cmp4 () < 1 || cmp8 () < 1 || cmp16 () < 1)
+    __builtin_abort ();
+}
diff --git a/gcc/testsuite/gcc.dg/strncmp-3.c b/gcc/testsuite/gcc.dg/strncmp-3.c
new file mode 100644
index 00000000000..0e8101cd61b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/strncmp-3.c
@@ -0,0 +1,57 @@ 
+/* PR middle-end/95189 - memcmp being wrongly stripped like strcmp
+   { dg-do run }
+   { dg-options "-O2 -Wall" } */
+
+#define AB_D "ab\0d"
+#define ABCDEF_H "abcdef\0h"
+#define ABCDEFGHIJKLMN_P "abcdefghijklmn\0p"
+
+char ab_d[] = AB_D;
+char abcdef_h[] = ABCDEF_H;
+
+extern int strncmp (const char*, const char*, __SIZE_TYPE__);
+
+__attribute__((noipa)) void sink (const void *p, ...) { (void)&p; }
+
+#define strncmp(a, b, n) (sink (a, b), strncmp (a, b, n))
+
+int main (void)
+{
+  int zero = 0;
+
+  zero += strncmp (ab_d, AB_D, 1);
+  zero += strncmp (ab_d, AB_D, 2);
+  zero += strncmp (ab_d, AB_D, 3);
+  zero += strncmp (ab_d, AB_D, 4);
+  zero += strncmp (ab_d, AB_D, 5);
+
+  zero += strncmp (ab_d, ABCDEF_H, 1);
+  zero += strncmp (ab_d, ABCDEF_H, 2);
+
+  zero += strncmp (abcdef_h, AB_D, 2);
+
+  zero += strncmp (abcdef_h, ABCDEF_H, 2);
+  zero += strncmp (abcdef_h, ABCDEF_H, 3);
+  zero += strncmp (abcdef_h, ABCDEF_H, 4);
+  zero += strncmp (abcdef_h, ABCDEF_H, 5);
+  zero += strncmp (abcdef_h, ABCDEF_H, 6);
+  zero += strncmp (abcdef_h, ABCDEF_H, 7);
+  zero += strncmp (abcdef_h, ABCDEF_H, 8);
+  zero += strncmp (abcdef_h, ABCDEF_H, 9);
+
+  if (zero != 0)
+    __builtin_abort ();
+
+  int neg = 0;
+
+  neg -= strncmp (ab_d, ABCDEF_H, 3) < 0;
+  neg -= strncmp (ab_d, ABCDEF_H, 4) < 0;
+  neg -= strncmp (ab_d, ABCDEF_H, 5) < 0;
+  neg -= strncmp (ab_d, ABCDEF_H, 6) < 0;
+  neg -= strncmp (ab_d, ABCDEF_H, 7) < 0;
+  neg -= strncmp (ab_d, ABCDEF_H, 8) < 0;
+  neg -= strncmp (ab_d, ABCDEF_H, 9) < 0;
+
+  if (neg != -7)
+    __builtin_abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/memcpy-pr95886.c b/gcc/testsuite/gcc.target/i386/memcpy-pr95886.c
new file mode 100644
index 00000000000..c0a04d27fac
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/memcpy-pr95886.c
@@ -0,0 +1,107 @@ 
+/* PR middle-end/95886 - suboptimal memcpy with embedded zero bytes
+   { dg-do compile }
+   { dg-options "-O2 -Wall -fdump-rtl-expand" } */
+
+const char a1234567890[10] = { 1, 2, 3, 4, 5, 6, 7, 8, 9 };
+
+void cpy_123456789 (void *d)
+{
+  /* Expands into:
+       movabsq  $578437695752307201, %rax
+       movb     $9, 8(%rdi)
+       movq     %rax, (%rdi)  */
+  __builtin_memcpy (d, a1234567890, 9);
+}
+
+const char a1234567800[10] = { 1, 2, 3, 4, 5, 6, 7, 8, 0 };
+
+void cpy_1234567800 (void *d)
+{
+  /* Expands into:
+       movabsq  $578437695752307201, %rax
+       movb     $0, 8(%rdi)
+       movq     %rax, (%rdi)  */
+  __builtin_memcpy (d, a1234567800, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 578437695752307201" 2 "expand"} } */
+
+
+const char a0234567890[10] = { 0, 2, 3, 4, 5, 6, 7, 8, 9 };
+
+void cpy_023456789 (void *d)
+{
+  __builtin_memcpy (d, a0234567890, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 578437695752307200" 1 "expand"} } */
+
+
+const char a1034567890[10] = { 1, 0, 3, 4, 5, 6, 7, 8, 9 };
+
+void cpy_103456789 (void *d)
+{
+  __builtin_memcpy (d, a1034567890, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 578437695752306689" 1 "expand"} } */
+
+
+const char a1204567890[10] = { 1, 2, 0, 4, 5, 6, 7, 8, 9 };
+
+void cpy_120456789 (void *d)
+{
+  __builtin_memcpy (d, a1204567890, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 578437695752110593" 1 "expand"} } */
+
+
+const char a1230567890[10] = { 1, 2, 3, 0, 5, 6, 7, 8, 9 };
+
+void cpy_123056789 (void *d)
+{
+  __builtin_memcpy (d, a1230567890, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 578437695685198337" 1 "expand"} } */
+
+
+const char a1234067890[10] = { 1, 2, 3, 4, 0, 6, 7, 8, 9 };
+
+void cpy_123406789 (void *d)
+{
+  __builtin_memcpy (d, a1234067890, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 578437695685198337" 1 "expand"} } */
+
+
+const char a1234507890[10] = { 1, 2, 3, 4, 5, 0, 7, 8, 9 };
+
+void cpy_123450789 (void *d)
+{
+  __builtin_memcpy (d, a1234507890, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 578431098682540545" 1 "expand"} } */
+
+
+const char a1234560890[10] = { 1, 2, 3, 4, 5, 6, 0, 8, 9 };
+
+void cpy_123456089 (void *d)
+{
+  __builtin_memcpy (d, a1234560890, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 576467370915332609" 1 "expand"} } */
+
+
+const char a1234567090[10] = { 1, 2, 3, 4, 5, 6, 7, 0, 9 };
+
+void cpy_123456709 (void *d)
+{
+  __builtin_memcpy (d, a1234567090, 9);
+}
+
+/* { dg-final { scan-rtl-dump-times "const_int 1976943448883713" 1 "expand"} } */
diff --git a/gcc/tree.c b/gcc/tree.c
index 342da55bba7..6d4575acd69 100644
--- a/gcc/tree.c
+++ b/gcc/tree.c
@@ -2204,29 +2204,29 @@  build_real_from_int_cst (tree type, const_tree i)
   return v;
 }
 
-/* Return a newly constructed STRING_CST node whose value is
-   the LEN characters at STR.
+/* Return a newly constructed STRING_CST node whose value is the LEN
+   characters at STR when STR is nonnull, or all zeros otherwise.
    Note that for a C string literal, LEN should include the trailing NUL.
    The TREE_TYPE is not initialized.  */
 
 tree
-build_string (int len, const char *str)
+build_string (unsigned len, const char *str /*= NULL */)
 {
-  tree s;
-  size_t length;
-
   /* Do not waste bytes provided by padding of struct tree_string.  */
-  length = len + offsetof (struct tree_string, str) + 1;
+  unsigned size = len + offsetof (struct tree_string, str) + 1;
 
-  record_node_allocation_statistics (STRING_CST, length);
+  record_node_allocation_statistics (STRING_CST, size);
 
-  s = (tree) ggc_internal_alloc (length);
+  tree s = (tree) ggc_internal_alloc (size);
 
   memset (s, 0, sizeof (struct tree_typed));
   TREE_SET_CODE (s, STRING_CST);
   TREE_CONSTANT (s) = 1;
   TREE_STRING_LENGTH (s) = len;
-  memcpy (s->string.str, str, len);
+  if (str)
+    memcpy (s->string.str, str, len);
+  else
+    memset (s->string.str, 0, len);
   s->string.str[len] = '\0';
 
   return s;
@@ -11570,12 +11570,12 @@  build_alloca_call_expr (tree size, unsigned int align, HOST_WIDE_INT max_size)
 
 /* Create a new constant string literal of type ELTYPE[SIZE] (or LEN
    if SIZE == -1) and return a tree node representing char* pointer to
-   it as an ADDR_EXPR (ARRAY_REF (ELTYPE, ...)).  The STRING_CST value
-   is the LEN bytes at STR (the representation of the string, which may
-   be wide).  */
+   it as an ADDR_EXPR (ARRAY_REF (ELTYPE, ...)).  When STR is nonnull
+   the STRING_CST value is the LEN bytes at STR (the representation
+   of the string, which may be wide).  Otherwise it's all zeros.  */
 
 tree
-build_string_literal (int len, const char *str,
+build_string_literal (unsigned len, const char *str /* = NULL */,
 		      tree eltype /* = char_type_node */,
 		      unsigned HOST_WIDE_INT size /* = -1 */)
 {
diff --git a/gcc/tree.h b/gcc/tree.h
index a74872f5f3e..7f35706d284 100644
--- a/gcc/tree.h
+++ b/gcc/tree.h
@@ -4423,7 +4423,7 @@  extern tree build_one_cst (tree);
 extern tree build_minus_one_cst (tree);
 extern tree build_all_ones_cst (tree);
 extern tree build_zero_cst (tree);
-extern tree build_string (int, const char *);
+extern tree build_string (unsigned, const char * = NULL);
 extern tree build_poly_int_cst (tree, const poly_wide_int_ref &);
 extern tree build_tree_list (tree, tree CXX_MEM_STAT_INFO);
 extern tree build_tree_list_vec (const vec<tree, va_gc> * CXX_MEM_STAT_INFO);
@@ -4454,7 +4454,8 @@  extern tree build_call_expr_internal_loc_array (location_t, enum internal_fn,
 extern tree maybe_build_call_expr_loc (location_t, combined_fn, tree,
 				       int, ...);
 extern tree build_alloca_call_expr (tree, unsigned int, HOST_WIDE_INT);
-extern tree build_string_literal (int, const char *, tree = char_type_node,
+extern tree build_string_literal (unsigned, const char * = NULL,
+				  tree = char_type_node,
 				  unsigned HOST_WIDE_INT = HOST_WIDE_INT_M1U);
 
 /* Construct various nodes representing data types.  */