diff options
Diffstat (limited to 'gcc/config/i386')
| -rw-r--r-- | gcc/config/i386/i386.c | 32 |
1 files changed, 32 insertions, 0 deletions
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c index 59101d8028a..c50f4dac199 100644 --- a/gcc/config/i386/i386.c +++ b/gcc/config/i386/i386.c @@ -22711,6 +22711,35 @@ ix86_push_rounding (poly_int64 bytes) return ROUND_UP (bytes, UNITS_PER_WORD); } +/* Return TRUE if offloaded OpenACC target-code regions should have their + parameters passed as separate function arguments, rather than in an array. + This can be a performance win on some (NVidia) GPUs. */ + +bool +ix86_goacc_explode_args (void) +{ +#ifdef OFFLOAD_TARGETS + const char *offload_targets = OFFLOAD_TARGETS; + if (strstr (offload_targets, "nvptx")) + { + if (strchr (offload_targets, ',')) + { + static bool warned_ptx_args = false; + if (!warned_ptx_args) + { + warning (0, "NVidia PTX parameter-passing optimization disabled " + "with multiple offload targets"); + warned_ptx_args = true; + } + return false; + } + return true; + } + +#endif + return false; +} + /* Target-specific selftests. */ #if CHECKING_P @@ -23496,6 +23525,9 @@ static bool ix86_libc_has_fast_function (int fcode ATTRIBUTE_UNUSED) #undef TARGET_LIBC_HAS_FAST_FUNCTION #define TARGET_LIBC_HAS_FAST_FUNCTION ix86_libc_has_fast_function +#undef TARGET_GOACC_EXPLODE_ARGS +#define TARGET_GOACC_EXPLODE_ARGS ix86_goacc_explode_args + #if CHECKING_P #undef TARGET_RUN_TARGET_SELFTESTS #define TARGET_RUN_TARGET_SELFTESTS selftest::ix86_run_selftests |