On 09/01/16 07:01, Jakub Jelinek wrote: > On Fri, Jan 08, 2016 at 11:43:34AM -0800, Jim Wilson wrote: >> Here is a smaller simpler testcase. Only the first four args get >> passed in regs, so the fifth one has address equal to the virtual >> incoming args reg which triggers the failure. >> >> typedef __simd128_float32_t float32x4_t; >> float32x4_t >> sub (float32x4_t a, float32x4_t b, float32x4_t c, float32x4_t d, float32x4_t >> e) >> { >> return __builtin_neon_vld1v4sf((const float *)&e); >> } >> >> The original testcase failed with linaro gcc 4.9. I verified that >> this testcase fails with FSF gcc-5.3.0. > > Looks much better. I'll defer to ARM maintainers to ack this.
Thanks for the review. Please find the patch with the required changes. Regression tested with trunk and gcc-5 for arn-none-linux-gnu with no new regression. Is this OK ? Kugan 2016-01-10 Kugan Vivekanandarajah <kug...@linaro.org> Jim Wilson <jim.wil...@linaro.org> PR target/69194 * config/arm/arm-builtins.c (arm_expand_neon_args): Call copy_to_mode_reg instead of force_reg. gcc/testsuite/ChangeLog: 2016-01-10 Kugan Vivekanandarajah <kug...@linaro.org> Jim Wilson <jim.wil...@linaro.org> PR target/69194 * gcc.target/arm/pr69194.C: New test.
diff --git a/gcc/config/arm/arm-builtins.c b/gcc/config/arm/arm-builtins.c index 11cd17d..fbdf0a3 100644 --- a/gcc/config/arm/arm-builtins.c +++ b/gcc/config/arm/arm-builtins.c @@ -2146,7 +2146,8 @@ constant_arg: if (!(*insn_data[icode].operand[opno].predicate) (op[argc], mode[argc])) op[argc] = (replace_equiv_address - (op[argc], force_reg (Pmode, XEXP (op[argc], 0)))); + (op[argc], + copy_to_mode_reg (Pmode, XEXP (op[argc], 0)))); break; case NEON_ARG_STOP: diff --git a/gcc/testsuite/gcc.target/arm/pr69194.c b/gcc/testsuite/gcc.target/arm/pr69194.c index e69de29..10630c1 100644 --- a/gcc/testsuite/gcc.target/arm/pr69194.c +++ b/gcc/testsuite/gcc.target/arm/pr69194.c @@ -0,0 +1,12 @@ +/* PR target/69194 */ +/* { dg-do-compile } */ +/* { dg-require-effective-target arm_neon } */ +/* { dg-options "-O2 -mfpu=neon -mfloat-abi=hard" } */ + +typedef __simd128_float32_t float32x4_t; + +float32x4_t +sub (float32x4_t a, float32x4_t b, float32x4_t c, float32x4_t d, float32x4_t e) +{ + return __builtin_neon_vld1v4sf((const float *)&e); +}