S/390: Don't emit prefetch instructions for clrmem

gcc/ChangeLog:

2018-07-31  Andreas Krebbel  <krebbel@linux.ibm.com>

	* config/s390/s390.c (s390_expand_setmem): Make the unrolling to
	depend on whether prefetch instructions will be emitted or not.
	Use TARGET_SETMEM_PFD for checking whether prefetch instructions
	will be emitted or not.
	* config/s390/s390.h (TARGET_SETMEM_PREFETCH_DISTANCE)
	(TARGET_SETMEM_PFD): New macros.

gcc/testsuite/ChangeLog:

2018-07-31  Andreas Krebbel  <krebbel@linux.ibm.com>

	* gcc.target/s390/memset-1.c: Improve testcase.

From-SVN: r263165

gcc/ChangeLog

+2018-07-31  Andreas Krebbel  <krebbel@linux.ibm.com>
+
+	* config/s390/s390.c (s390_expand_setmem): Make the unrolling to
+	depend on whether prefetch instructions will be emitted or not.
+	Use TARGET_SETMEM_PFD for checking whether prefetch instructions
+	will be emitted or not.
+	* config/s390/s390.h (TARGET_SETMEM_PREFETCH_DISTANCE)
+	(TARGET_SETMEM_PFD): New macros.
+
 2018-07-31  Richard Sandiford  <richard.sandiford@arm.com>
 
 	* tree-vectorizer.h (stmt_vec_info): Turn back into a typedef.

gcc/config/s390/s390.c

@@ -5499,12 +5499,15 @@ s390_expand_setmem (rtx dst, rtx len, rtx val)
 
   /* Expand setmem/clrmem for a constant length operand without a
      loop if it will be shorter that way.
-     With a constant length and without pfd argument a
-     clrmem loop is 32 bytes -> 5.3 * xc
-     setmem loop is 36 bytes -> 3.6 * (mvi/stc + mvc) */
+     clrmem loop (with PFD)    is 30 bytes -> 5 * xc
+     clrmem loop (without PFD) is 24 bytes -> 4 * xc
+     setmem loop (with PFD)    is 38 bytes -> ~4 * (mvi/stc + mvc)
+     setmem loop (without PFD) is 32 bytes -> ~4 * (mvi/stc + mvc) */
   if (GET_CODE (len) == CONST_INT
-      && ((INTVAL (len) <= 256 * 5 && val == const0_rtx)
-	  || INTVAL (len) <= 257 * 3)
+      && ((val == const0_rtx
+	   && (INTVAL (len) <= 256 * 4
+	       || (INTVAL (len) <= 256 * 5 && TARGET_SETMEM_PFD(val,len))))
+	  || (val != const0_rtx && INTVAL (len) <= 257 * 4))
       && (!TARGET_MVCLE || INTVAL (len) <= 256))
     {
       HOST_WIDE_INT o, l;
@@ -5618,12 +5621,11 @@ s390_expand_setmem (rtx dst, rtx len, rtx val)
 
       emit_label (loop_start_label);
 
-      if (TARGET_Z10
-	  && (GET_CODE (len) != CONST_INT || INTVAL (len) > 1024))
+      if (TARGET_SETMEM_PFD (val, len))
 	{
-	  /* Issue a write prefetch for the +4 cache line.  */
-	  rtx prefetch = gen_prefetch (gen_rtx_PLUS (Pmode, dst_addr,
-						     GEN_INT (1024)),
+	  /* Issue a write prefetch.  */
+	  rtx distance = GEN_INT (TARGET_SETMEM_PREFETCH_DISTANCE);
+	  rtx prefetch = gen_prefetch (gen_rtx_PLUS (Pmode, dst_addr, distance),
 				       const1_rtx, const0_rtx);
 	  emit_insn (prefetch);
 	  PREFETCH_SCHEDULE_BARRIER_P (prefetch) = true;

gcc/config/s390/s390.h

@@ -181,6 +181,16 @@ enum processor_flags
 
 #define TARGET_AVOID_CMP_AND_BRANCH (s390_tune == PROCESSOR_2817_Z196)
 
+/* Issue a write prefetch for the +4 cache line.  */
+#define TARGET_SETMEM_PREFETCH_DISTANCE 1024
+
+/* Expand to a C expressions evaluating to true if a setmem to VAL of
+   length LEN should be emitted using prefetch instructions.  */
+#define TARGET_SETMEM_PFD(VAL,LEN)					\
+  (TARGET_Z10								\
+   && (s390_tune < PROCESSOR_2964_Z13 || (VAL) != const0_rtx)		\
+   && (!CONST_INT_P (LEN) || INTVAL ((LEN)) > TARGET_SETMEM_PREFETCH_DISTANCE))
+
 /* Run-time target specification.  */
 
 /* Defaults for option flags defined only on some subtargets.  */

gcc/testsuite/ChangeLog

+2018-07-31  Andreas Krebbel  <krebbel@linux.ibm.com>
+
+	* gcc.target/s390/memset-1.c: Improve testcase.
+
 2018-07-31  Tom de Vries  <tdevries@suse.de>
 
 	PR debug/86687

gcc/testsuite/gcc.target/s390/memset-1.c

@@ -2,16 +2,23 @@
    without loop statements.  */
 
 /* { dg-do compile } */
-/* { dg-options "-O3 -mzarch" } */
+/* { dg-options "-O3 -mzarch -march=z13" } */
 
-/* 1 mvc */
+/* 1 stc */
+void
+*memset0(void *s, int c)
+{
+  return __builtin_memset (s, c, 1);
+}
+
+/* 1 stc 1 mvc */
 void
 *memset1(void *s, int c)
 {
   return __builtin_memset (s, c, 42);
 }
 
-/* 3 mvc */
+/* 3 stc 3 mvc */
 void
 *memset2(void *s, int c)
 {
@@ -25,55 +32,62 @@ void
   return __builtin_memset (s, c, 0);
 }
 
-/* mvc */
+/* 1 stc 1 mvc */
 void
 *memset4(void *s, int c)
 {
   return __builtin_memset (s, c, 256);
 }
 
-/* 2 mvc */
+/* 2 stc 2 mvc */
 void
 *memset5(void *s, int c)
 {
   return __builtin_memset (s, c, 512);
 }
 
-/* still 2 mvc through the additional first byte  */
+/* 2 stc 2 mvc - still due to the stc bytes */
 void
 *memset6(void *s, int c)
 {
   return __builtin_memset (s, c, 514);
 }
 
-/* 3 mvc */
+/* 3 stc 2 mvc */
 void
 *memset7(void *s, int c)
 {
   return __builtin_memset (s, c, 515);
 }
 
-/* still 3 mvc through the additional first byte  */
+/* 4 stc 4 mvc - 4 * 256 + 4 stc bytes */
 void
 *memset8(void *s, int c)
 {
-  return __builtin_memset (s, c, 771);
+  return __builtin_memset (s, c, 1028);
 }
 
-/* Use mvc loop: 2 mvc */
+/* 2 stc 1 pfd 2 mvc - start using mvc loop */
 void
 *memset9(void *s, int c)
 {
-  return __builtin_memset (s, c, 772);
+  return __builtin_memset (s, c, 1029);
 }
 
-/* 3 mvc with displacement overflow after the first */
+/* 2 stc 1 stcy 3 mvc - displacement overflow after the first */
 void
 *memset10(void *s, int c)
 {
   return __builtin_memset ((char*)s + 4000, c, 700);
 }
 
+/* 1 mvi */
+void
+*clrmem0(void *s)
+{
+  return __builtin_memset (s, 0, 1);
+}
+
 /* 1 xc */
 void
 *clrmem1(void *s)
@@ -109,26 +123,55 @@ void
   return __builtin_memset (s, 0, 512);
 }
 
-/* 3 xc */
+/* 4 xc */
 void
 *clrmem6(void *s)
 {
-  return __builtin_memset (s, 0, 768);
+  return __builtin_memset (s, 0, 1024);
 }
 
-/* start using xc loop */
+/* 2 xc - start using xc loop*/
 void
 *clrmem7(void *s)
 {
+  return __builtin_memset (s, 0, 1025);
+}
+
+/* 5 xc - on z10 PFD would be used in the loop body so the unrolled
+   variant would still be shorter.  */
+__attribute__ ((target("tune=z10")))
+void
+*clrmem7_z10(void *s)
+{
+  return __builtin_memset (s, 0, 1025);
+}
+
+/* 5 xc */
+__attribute__ ((target("tune=z10")))
+void
+*clrmem8_z10(void *s)
+{
+  return __builtin_memset (s, 0, 1280);
+}
+
+/* 1 pfd 2 xc - start using xc loop also on z10 */
+__attribute__ ((target("tune=z10")))
+void
+*clrmem9_z10(void *s)
+{
   return __builtin_memset (s, 0, 1281);
 }
 
-/* 3 xc with displacement overflow after the first */
+/* 3 xc - displacement overflow after the first */
 void
-*clrmem8(void *s)
+*clrmem10(void *s)
 {
   return __builtin_memset (s + 4000, 0, 700);
 }
 
-/* { dg-final { scan-assembler-times "mvc" 19 } } */
-/* { dg-final { scan-assembler-times "xc" 15 } } */
+/* { dg-final { scan-assembler-times "mvi\\s" 1 } } */
+/* { dg-final { scan-assembler-times "mvc\\s" 20 } } */
+/* { dg-final { scan-assembler-times "xc\\s" 28 } } */
+/* { dg-final { scan-assembler-times "stc\\s" 21 } } */
+/* { dg-final { scan-assembler-times "stcy\\s" 1 } } */
+/* { dg-final { scan-assembler-times "pfd\\s" 2 } } */