[V2,3/3] split complicate constant to memory
Checks
Commit Message
Hi,
Sometimes, a complicated constant is built via 3(or more)
instructions to build. Generally speaking, it would not be
as faster as loading it from the constant pool (as a few
discussions in PR63281):
* "ld" is one instruction. If consider "address/toc"
adjust, we may count it as 2 instructions (the high part
of address computation could be optimized as nop by linker
further). And "pld" may need less cycles.
* As testing(SPEC2017), it could get better/stable runtime
if set the threshold as "> 2" (compare with "> 3").
As tested on spec2017, for visible performance changes, we
can find the runtime improvement on 500.perlbench_r about
~1.8% (-O2, P10) with the patch. And for performance
downgrade on other benchmarks, as investigation, the recessions
are not caused by this patch.
Compare with previous version:
https://gcc.gnu.org/pipermail/gcc-patches/2023-October/634197.html
This verion updates commit message.
Boostrap & regtest pass on ppc64{,le}.
Is this ok for trunk?
BR,
Jeff (Jiufu Guo)
PR target/63281
gcc/ChangeLog:
* config/rs6000/rs6000.cc (rs6000_emit_set_const): Update to split
complicate constant to memory.
gcc/testsuite/ChangeLog:
* gcc.target/powerpc/const_anchors.c: Update to test final-rtl.
* gcc.target/powerpc/parall_5insn_const.c: Update to keep original test
point.
* gcc.target/powerpc/pr106550.c: Likewise..
* gcc.target/powerpc/pr106550_1.c: Likewise.
* gcc.target/powerpc/pr87870.c: Update according to latest behavior.
* gcc.target/powerpc/pr93012.c: Likewise.
---
gcc/config/rs6000/rs6000.cc | 16 ++++++++++++++++
.../gcc.target/powerpc/const_anchors.c | 5 ++---
.../gcc.target/powerpc/parall_5insn_const.c | 14 ++++++++++++--
gcc/testsuite/gcc.target/powerpc/pr106550.c | 17 +++++++++++++++--
gcc/testsuite/gcc.target/powerpc/pr106550_1.c | 15 +++++++++++++--
gcc/testsuite/gcc.target/powerpc/pr87870.c | 5 ++++-
gcc/testsuite/gcc.target/powerpc/pr93012.c | 4 +++-
7 files changed, 65 insertions(+), 11 deletions(-)
@@ -10271,6 +10271,22 @@ rs6000_emit_set_const (rtx dest, rtx source)
c = sext_hwi (c, 32);
emit_move_insn (lo, GEN_INT (c));
}
+
+ /* If it can be stored to the constant pool and profitable. */
+ else if (base_reg_operand (dest, mode)
+ && num_insns_constant (source, mode) > 2)
+ {
+ rtx sym = force_const_mem (mode, source);
+ if (TARGET_TOC && SYMBOL_REF_P (XEXP (sym, 0))
+ && use_toc_relative_ref (XEXP (sym, 0), mode))
+ {
+ rtx toc = create_TOC_reference (XEXP (sym, 0), copy_rtx (dest));
+ sym = gen_const_mem (mode, toc);
+ set_mem_alias_set (sym, get_TOC_alias_set ());
+ }
+
+ emit_insn (gen_rtx_SET (dest, sym));
+ }
else
rs6000_emit_set_long_const (dest, c, NULL);
break;
@@ -1,5 +1,5 @@
/* { dg-do compile { target has_arch_ppc64 } } */
-/* { dg-options "-O2" } */
+/* { dg-options "-O2 -fdump-rtl-final" } */
#define C1 0x2351847027482577ULL
#define C2 0x2351847027482578ULL
@@ -16,5 +16,4 @@ void __attribute__ ((noinline)) foo1 (long long *a, long long b)
if (b)
*a++ = C2;
}
-
-/* { dg-final { scan-assembler-times {\maddi\M} 2 } } */
+/* { dg-final { scan-rtl-dump-times {\madddi3\M} 2 "final" } } */
@@ -9,8 +9,18 @@
void __attribute__ ((noinline)) foo (unsigned long long *a)
{
/* 2 lis + 2 ori + 1 rldimi for each constant. */
- *a++ = 0x800aabcdc167fa16ULL;
- *a++ = 0x7543a876867f616ULL;
+ {
+ register long long d asm("r0") = 0x800aabcdc167fa16ULL;
+ long long n;
+ asm("mr %0, %1" : "=r"(n) : "r"(d));
+ *a++ = n;
+ }
+ {
+ register long long d asm("r0") = 0x7543a876867f616ULL;
+ long long n;
+ asm("mr %0, %1" : "=r"(n) : "r"(d));
+ *a++ = n;
+ }
}
long long A[] = {0x800aabcdc167fa16ULL, 0x7543a876867f616ULL};
@@ -1,12 +1,25 @@
/* PR target/106550 */
/* { dg-options "-O2 -mdejagnu-cpu=power10" } */
/* { dg-require-effective-target power10_ok } */
+/* { dg-require-effective-target has_arch_ppc64 } */
void
foo (unsigned long long *a)
{
- *a++ = 0x020805006106003; /* pli+pli+rldimi */
- *a++ = 0x2351847027482577;/* pli+pli+rldimi */
+ {
+ /* pli+pli+rldimi */
+ register long long d asm("r0") = 0x020805006106003ULL;
+ long long n;
+ asm("mr %0, %1" : "=r"(n) : "r"(d));
+ *a++ = n;
+ }
+ {
+ /* pli+pli+rldimi */
+ register long long d asm("r0") = 0x2351847027482577ULL;
+ long long n;
+ asm("mr %0, %1" : "=r"(n) : "r"(d));
+ *a++ = n;
+ }
}
/* { dg-final { scan-assembler-times {\mpli\M} 4 } } */
@@ -12,8 +12,19 @@ foo (unsigned long long *a)
asm("cntlzd %0, %1" : "=r"(n) : "r"(d));
*a++ = n;
- *a++ = 0x235a8470a7480000ULL; /* pli+sldi+oris */
- *a++ = 0x23a184700000b677ULL; /* pli+sldi+ori */
+ {
+ register long long d asm("r0") = 0x235a8470a7480000ULL; /* pli+sldi+oris */
+ long long n;
+ asm("cntlzd %0, %1" : "=r"(n) : "r"(d));
+ *a++ = n;
+ }
+
+ {
+ register long long d asm("r0") = 0x23a184700000b677ULL; /* pli+sldi+ori */
+ long long n;
+ asm("cntlzd %0, %1" : "=r"(n) : "r"(d));
+ *a++ = n;
+ }
}
/* { dg-final { scan-assembler-times {\mpli\M} 3 } } */
@@ -25,4 +25,7 @@ test3 (void)
return ((__int128)0xdeadbeefcafebabe << 64) | 0xfacefeedbaaaaaad;
}
-/* { dg-final { scan-assembler-not {\mld\M} } } */
+/* test3 using "ld" to load the value for r3 and r4.
+ test0, test1 and test2 are using "li". */
+/* { dg-final { scan-assembler-times {\mp?ld\M} 2 } } */
+/* { dg-final { scan-assembler-times {\mli\M} 6 } } */
@@ -11,4 +11,6 @@ unsigned long long mskl1() { return 0x2bcdffff2bcdffffULL; }
unsigned long long mskse() { return 0xffff1234ffff1234ULL; }
/* { dg-final { scan-assembler-times {\mpli\M} 4 { target has_arch_pwr10 }} } */
-/* { dg-final { scan-assembler-times {\mrldimi\M} 7 } } */
+/* { dg-final { scan-assembler-times {\mrldimi\M} 7 { target has_arch_pwr10 } } } */
+/* { dg-final { scan-assembler-times {\mrldimi\M} 3 { target { ! has_arch_pwr10 } } } } */
+/* { dg-final { scan-assembler-times {\mld\M} 4 { target { ! has_arch_pwr10 } } } } */