The IBM z14 POP adds an optional alignment operand to the vl, vst, vlm, and vstm instruction (vector loads and stores). Vectors residing on 8 or 16 byte boundaries might get loaded or stored faster on some models given the instruction uses the proper hint operand. A wrong hint will hurt performance though.
The attached testcase align-1 currently fails due to: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88085 Bootstrapped and regtested on s390x with z13 and z14 defaults. Committed to mainline gcc/ChangeLog: 2018-11-21 Andreas Krebbel <kreb...@linux.ibm.com> * configure.ac: Add check for Binutils to determine whether vector load/store alignments hints are being supported. * config.in: Regenerate. * configure: Regenerate. * config/s390/s390.c (print_operand): Support new output modifier A. * config/s390/s390.md ("movti"): Append alignment hint output using the new output modifier 'A'. * config/s390/vector.md ("mov<mode>", "*vec_tf_to_v1tf") ("*vec_ti_to_v1ti"): Likewise. gcc/testsuite/ChangeLog: 2018-11-21 Andreas Krebbel <kreb...@linux.ibm.com> * gcc.target/s390/vector/align-1.c: New test. * gcc.target/s390/vector/align-2.c: New test. --- gcc/config.in | 7 +++++ gcc/config/s390/s390.c | 13 ++++++++++ gcc/config/s390/s390.md | 4 +-- gcc/config/s390/vector.md | 12 ++++----- gcc/configure | 36 ++++++++++++++++++++++++++ gcc/configure.ac | 6 +++++ gcc/testsuite/gcc.target/s390/vector/align-1.c | 30 +++++++++++++++++++++ gcc/testsuite/gcc.target/s390/vector/align-2.c | 29 +++++++++++++++++++++ 8 files changed, 129 insertions(+), 8 deletions(-) create mode 100644 gcc/testsuite/gcc.target/s390/vector/align-1.c create mode 100644 gcc/testsuite/gcc.target/s390/vector/align-2.c diff --git a/gcc/config.in b/gcc/config.in index 67a1e6c..999ade4 100644 --- a/gcc/config.in +++ b/gcc/config.in @@ -657,6 +657,13 @@ #endif +/* Define if your assembler supports vl/vst/vlm/vstm with an optional + alignment hint argument. */ +#ifndef USED_FOR_TARGET +#undef HAVE_AS_VECTOR_LOADSTORE_ALIGNMENT_HINTS +#endif + + /* Define if your assembler supports VSX instructions. */ #ifndef USED_FOR_TARGET #undef HAVE_AS_VSX diff --git a/gcc/config/s390/s390.c b/gcc/config/s390/s390.c index ab06ada..277d555 100644 --- a/gcc/config/s390/s390.c +++ b/gcc/config/s390/s390.c @@ -7598,6 +7598,8 @@ print_operand_address (FILE *file, rtx addr) CODE specified the format flag. The following format flags are recognized: + 'A': On z14 or higher: If operand is a mem print the alignment + hint usable with vl/vst prefixed by a comma. 'C': print opcode suffix for branch condition. 'D': print opcode suffix for inverse branch condition. 'E': print opcode suffix for branch on index instruction. @@ -7635,6 +7637,17 @@ print_operand (FILE *file, rtx x, int code) switch (code) { + case 'A': +#ifdef HAVE_AS_VECTOR_LOADSTORE_ALIGNMENT_HINTS + if (TARGET_Z14 && MEM_P (x)) + { + if (MEM_ALIGN (x) >= 128) + fprintf (file, ",4"); + else if (MEM_ALIGN (x) == 64) + fprintf (file, ",3"); + } +#endif + return; case 'C': fprintf (file, s390_branch_condition_mnemonic (x, FALSE)); return; diff --git a/gcc/config/s390/s390.md b/gcc/config/s390/s390.md index 30d113f..c6c960f 100644 --- a/gcc/config/s390/s390.md +++ b/gcc/config/s390/s390.md @@ -1552,8 +1552,8 @@ vone\t%v0 vlvgp\t%v0,%1,%N1 # - vl\t%v0,%1 - vst\t%v1,%0 + vl\t%v0,%1%A1 + vst\t%v1,%0%A0 # # # diff --git a/gcc/config/s390/vector.md b/gcc/config/s390/vector.md index 102313f..f0e4049 100644 --- a/gcc/config/s390/vector.md +++ b/gcc/config/s390/vector.md @@ -198,8 +198,8 @@ "" "@ vlr\t%v0,%v1 - vl\t%v0,%1 - vst\t%v1,%0 + vl\t%v0,%1%A1 + vst\t%v1,%0%A0 vzero\t%v0 vone\t%v0 vgbm\t%v0,%t1 @@ -551,8 +551,8 @@ "TARGET_VX" "@ vmrhg\t%v0,%1,%N1 - vl\t%v0,%1 - vst\t%v1,%0 + vl\t%v0,%1%A1 + vst\t%v1,%0%A0 vzero\t%v0 vlvgp\t%v0,%1,%N1" [(set_attr "op_type" "VRR,VRX,VRX,VRI,VRR")]) @@ -563,8 +563,8 @@ "TARGET_VX" "@ vlr\t%v0,%v1 - vl\t%v0,%1 - vst\t%v1,%0 + vl\t%v0,%1%A1 + vst\t%v1,%0%A0 vzero\t%v0 vone\t%v0 vlvgp\t%v0,%1,%N1" diff --git a/gcc/configure b/gcc/configure index 8957362..3a20399 100755 --- a/gcc/configure +++ b/gcc/configure @@ -27520,6 +27520,42 @@ $as_echo "#define HAVE_AS_ARCHITECTURE_MODIFIERS 1" >>confdefs.h fi + { $as_echo "$as_me:${as_lineno-$LINENO}: checking assembler for vector load/store alignment hints" >&5 +$as_echo_n "checking assembler for vector load/store alignment hints... " >&6; } +if ${gcc_cv_as_s390_vector_loadstore_alignment_hints+:} false; then : + $as_echo_n "(cached) " >&6 +else + gcc_cv_as_s390_vector_loadstore_alignment_hints=no + if test $in_tree_gas = yes; then + if test $gcc_cv_gas_vers -ge `expr \( \( 2 \* 1000 \) + 31 \) \* 1000 + 0` + then gcc_cv_as_s390_vector_loadstore_alignment_hints=yes +fi + elif test x$gcc_cv_as != x; then + $as_echo ' vl %v24,0(%r15),3 ' > conftest.s + if { ac_try='$gcc_cv_as $gcc_cv_as_flags -o conftest.o conftest.s >&5' + { { eval echo "\"\$as_me\":${as_lineno-$LINENO}: \"$ac_try\""; } >&5 + (eval $ac_try) 2>&5 + ac_status=$? + $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5 + test $ac_status = 0; }; } + then + gcc_cv_as_s390_vector_loadstore_alignment_hints=yes + else + echo "configure: failed program was" >&5 + cat conftest.s >&5 + fi + rm -f conftest.o conftest.s + fi +fi +{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $gcc_cv_as_s390_vector_loadstore_alignment_hints" >&5 +$as_echo "$gcc_cv_as_s390_vector_loadstore_alignment_hints" >&6; } +if test $gcc_cv_as_s390_vector_loadstore_alignment_hints = yes; then + +$as_echo "#define HAVE_AS_VECTOR_LOADSTORE_ALIGNMENT_HINTS 1" >>confdefs.h + +fi + + ;; esac diff --git a/gcc/configure.ac b/gcc/configure.ac index 260d987..2a1f3bb 100644 --- a/gcc/configure.ac +++ b/gcc/configure.ac @@ -4846,6 +4846,12 @@ pointers into PC-relative form.]) [ .machine z13+vx ],, [AC_DEFINE(HAVE_AS_ARCHITECTURE_MODIFIERS, 1, [Define if your assembler supports architecture modifiers.])]) + gcc_GAS_CHECK_FEATURE([vector load/store alignment hints], + gcc_cv_as_s390_vector_loadstore_alignment_hints, [2,31,0],, + [ vl %v24,0(%r15),3 ],, + [AC_DEFINE(HAVE_AS_VECTOR_LOADSTORE_ALIGNMENT_HINTS, 1, + [Define if your assembler supports vl/vst/vlm/vstm with an optional alignment hint argument.])]) + ;; esac diff --git a/gcc/testsuite/gcc.target/s390/vector/align-1.c b/gcc/testsuite/gcc.target/s390/vector/align-1.c new file mode 100644 index 0000000..cc7777a --- /dev/null +++ b/gcc/testsuite/gcc.target/s390/vector/align-1.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-O3 -mzarch -march=z14" } */ + +/* The user alignment ends up in DECL_ALIGN of the VAR_DECL and is + currently ignored if it is smaller than the alignment of the type. + In this testcase an alignment hint ",3" is emitted also for + accessing a4 which is wrong. + Hence this testcase currently fails: + https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88085 */ + +typedef int __attribute__((vector_size(16))) v4si; + +v4si a4 __attribute__((aligned(4))); +v4si a8 __attribute__((aligned(8))); +v4si a16 __attribute__((aligned(16))); +v4si a32 __attribute__((aligned(32))); + +void +foo (v4si a) +{ + a4 += a; /* vl ... vst ... */ + a8 += a; /* vl ...,3 vst ...,3 */ + a16 += a; /* vl ...,4 vst ...,4 */ + a32 += a; /* vl ...,4 vst ...,4 */ +} + +/* { dg-final { scan-assembler-times "vl\t%v\[0-9\]*,\[0-9\]*\\(%r\[0-9\]*\\),3\n" 1 } } */ +/* { dg-final { scan-assembler-times "vl\t%v\[0-9\]*,\[0-9\]*\\(%r\[0-9\]*\\),4\n" 2 } } */ +/* { dg-final { scan-assembler-times "vst\t%v\[0-9\]*,\[0-9\]*\\(%r\[0-9\]*\\),3\n" 1 } } */ +/* { dg-final { scan-assembler-times "vst\t%v\[0-9\]*,\[0-9\]*\\(%r\[0-9\]*\\),4" 2 } } */ diff --git a/gcc/testsuite/gcc.target/s390/vector/align-2.c b/gcc/testsuite/gcc.target/s390/vector/align-2.c new file mode 100644 index 0000000..e4e2fba --- /dev/null +++ b/gcc/testsuite/gcc.target/s390/vector/align-2.c @@ -0,0 +1,29 @@ +/* { dg-do compile } */ +/* { dg-options "-O3 -mzarch -march=z14" } */ + +/* The user alignment ends up in TYPE_ALIGN of the type of the + VAR_DECL. */ + +typedef int __attribute__((vector_size(16),aligned(4))) v4si_4; +typedef int __attribute__((vector_size(16),aligned(8))) v4si_8; +typedef int __attribute__((vector_size(16),aligned(16))) v4si_16; +typedef int __attribute__((vector_size(16),aligned(32))) v4si_32; + +v4si_4 a4; +v4si_8 a8; +v4si_16 a16; +v4si_32 a32; + +void +foo (v4si_8 a) +{ + a4 += a; /* vl ... vst ... */ + a8 += a; /* vl ...,3 vst ...,3 */ + a16 += a; /* vl ...,4 vst ...,4 */ + a32 += a; /* vl ...,4 vst ...,4 */ +} + +/* { dg-final { scan-assembler-times "vl\t%v\[0-9\]*,\[0-9\]*\\(%r\[0-9\]*\\),3\n" 1 } } */ +/* { dg-final { scan-assembler-times "vl\t%v\[0-9\]*,\[0-9\]*\\(%r\[0-9\]*\\),4\n" 2 } } */ +/* { dg-final { scan-assembler-times "vst\t%v\[0-9\]*,\[0-9\]*\\(%r\[0-9\]*\\),3\n" 1 } } */ +/* { dg-final { scan-assembler-times "vst\t%v\[0-9\]*,\[0-9\]*\\(%r\[0-9\]*\\),4" 2 } } */ -- 2.7.4