From: Joseph Myers Date: Tue, 15 Aug 2017 23:42:23 +0000 (+0100) Subject: Limit SH strncmp inline expansion (PR target/78460). X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=aaf46c35d38b85342076e6c07f9a13afd598a38d;p=gcc.git Limit SH strncmp inline expansion (PR target/78460). GCC mainline built for sh4-linux-gnu runs out of memory building a glibc test, which calls strncmp with very large constant size argument, resulting in the SH inline strncmp expansion trying to inline a fully unrolled expansion of strncmp for that size. This patch limits that fully unrolled expansion to the case of less than 32 bytes. This is explicitly *not* trying to be optimal in any way (very likely a lower threshold makes sense), just to limit enough to avoid the out-of-memory issue in the glibc testsuite. I have *not* run the GCC testsuite for SH. I have verified that this allows the glibc testsuite to build OK, with both GCC mainline and GCC 7 branch (and that the included test builds quickly with patched GCC, runs out of memory with unpatched GCC). PR target/78460 PR target/67712 gcc: * config/sh/sh-mem.cc (sh_expand_cmpnstr): Only unroll for constant count if that count is less than 32. gcc/testsuite: * gcc.c-torture/compile/string-large-1.c: New test. From-SVN: r251108 --- diff --git a/gcc/ChangeLog b/gcc/ChangeLog index c95dc2830a2..14fc3dbe3e0 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,10 @@ +2017-08-15 Joseph Myers + + PR target/78460 + PR target/67712 + * config/sh/sh-mem.cc (sh_expand_cmpnstr): Only unroll for + constant count if that count is less than 32. + 2017-08-15 Nathan Sidwell * gcc.c (execute): Emit friendlier message if inferior is killed diff --git a/gcc/config/sh/sh-mem.cc b/gcc/config/sh/sh-mem.cc index 3f7041c2ae2..8fce9799921 100644 --- a/gcc/config/sh/sh-mem.cc +++ b/gcc/config/sh/sh-mem.cc @@ -349,12 +349,13 @@ sh_expand_cmpnstr (rtx *operands) rtx len = copy_to_mode_reg (SImode, operands[3]); int constp = CONST_INT_P (operands[3]); + HOST_WIDE_INT bytes = constp ? INTVAL (operands[3]) : 0; const unsigned int addr1_alignment = MEM_ALIGN (operands[1]) / BITS_PER_UNIT; const unsigned int addr2_alignment = MEM_ALIGN (operands[2]) / BITS_PER_UNIT; /* Loop on a register count. */ - if (constp) + if (constp && bytes >= 0 && bytes < 32) { rtx tmp0 = gen_reg_rtx (SImode); rtx tmp3 = gen_reg_rtx (SImode); @@ -363,7 +364,6 @@ sh_expand_cmpnstr (rtx *operands) rtx_code_label *L_loop_long = gen_label_rtx (); rtx_code_label *L_end_loop_long = gen_label_rtx (); - int bytes = INTVAL (operands[3]); int witers = bytes / 4; if (witers > 1) diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index f13f9b05003..835a8f4af81 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,9 @@ +2017-08-15 Joseph Myers + + PR target/78460 + PR target/67712 + * gcc.c-torture/compile/string-large-1.c: New test. + 2017-08-15 Richard Biener PR tree-optimization/81790 diff --git a/gcc/testsuite/gcc.c-torture/compile/string-large-1.c b/gcc/testsuite/gcc.c-torture/compile/string-large-1.c new file mode 100644 index 00000000000..ddac0d57e16 --- /dev/null +++ b/gcc/testsuite/gcc.c-torture/compile/string-large-1.c @@ -0,0 +1,119 @@ +/* Test built-in string functions with large sizes. PR 78460. */ + +typedef __SIZE_TYPE__ size_t; + +#define SIZE1 ((size_t) -1) +#define SIZE2 (SIZE1 >> 1) +#define SIZE3 ((unsigned int) -1) +#define SIZE4 (SIZE3 >> 1) + +volatile int v1, v2, v3, v4; +void *volatile vp1, *volatile vp2, *volatile vp3, *volatile vp4; + +void +test_memchr (const void *a, int b) +{ + vp1 = __builtin_memchr (a, b, SIZE1); + vp2 = __builtin_memchr (a, b, SIZE2); + vp3 = __builtin_memchr (a, b, SIZE3); + vp4 = __builtin_memchr (a, b, SIZE4); +} + +void +test_memcmp (const void *a, const void *b) +{ + v1 = __builtin_memcmp (a, b, SIZE1); + v2 = __builtin_memcmp (a, b, SIZE2); + v3 = __builtin_memcmp (a, b, SIZE3); + v4 = __builtin_memcmp (a, b, SIZE4); +} + +void +test_memcpy (void *a, const void *b) +{ + vp1 = __builtin_memcpy (a, b, SIZE1); + vp2 = __builtin_memcpy (a, b, SIZE2); + vp3 = __builtin_memcpy (a, b, SIZE3); + vp4 = __builtin_memcpy (a, b, SIZE4); +} + +void +test_memmove (void *a, const void *b) +{ + vp1 = __builtin_memmove (a, b, SIZE1); + vp2 = __builtin_memmove (a, b, SIZE2); + vp3 = __builtin_memmove (a, b, SIZE3); + vp4 = __builtin_memmove (a, b, SIZE4); +} + +void +test_mempcpy (void *a, const void *b) +{ + vp1 = __builtin_mempcpy (a, b, SIZE1); + vp2 = __builtin_mempcpy (a, b, SIZE2); + vp3 = __builtin_mempcpy (a, b, SIZE3); + vp4 = __builtin_mempcpy (a, b, SIZE4); +} + +void +test_memset (void *a, int b) +{ + vp1 = __builtin_memset (a, b, SIZE1); + vp2 = __builtin_memset (a, b, SIZE2); + vp3 = __builtin_memset (a, b, SIZE3); + vp4 = __builtin_memset (a, b, SIZE4); +} + +void +test_stpncpy (char *a, const char *b) +{ + vp1 = __builtin_stpncpy (a, b, SIZE1); + vp2 = __builtin_stpncpy (a, b, SIZE2); + vp3 = __builtin_stpncpy (a, b, SIZE3); + vp4 = __builtin_stpncpy (a, b, SIZE4); +} + +void +test_strndup (const char *a) +{ + vp1 = __builtin_strndup (a, SIZE1); + vp2 = __builtin_strndup (a, SIZE2); + vp3 = __builtin_strndup (a, SIZE3); + vp4 = __builtin_strndup (a, SIZE4); +} + +void +test_strncasecmp (const char *a, const char *b) +{ + v1 = __builtin_strncasecmp (a, b, SIZE1); + v2 = __builtin_strncasecmp (a, b, SIZE2); + v3 = __builtin_strncasecmp (a, b, SIZE3); + v4 = __builtin_strncasecmp (a, b, SIZE4); +} + +void +test_strncat (char *a, const char *b) +{ + vp1 = __builtin_strncat (a, b, SIZE1); + vp2 = __builtin_strncat (a, b, SIZE2); + vp3 = __builtin_strncat (a, b, SIZE3); + vp4 = __builtin_strncat (a, b, SIZE4); +} + +void +test_strncmp (const char *a, const char *b) +{ + v1 = __builtin_strncmp (a, b, SIZE1); + v2 = __builtin_strncmp (a, b, SIZE2); + v3 = __builtin_strncmp (a, b, SIZE3); + v4 = __builtin_strncmp (a, b, SIZE4); +} + +void +test_strncpy (char *a, const char *b) +{ + vp1 = __builtin_strncpy (a, b, SIZE1); + vp2 = __builtin_strncpy (a, b, SIZE2); + vp3 = __builtin_strncpy (a, b, SIZE3); + vp4 = __builtin_strncpy (a, b, SIZE4); +}