x86: Mark scratch operand in ssse3_pshufbv8qi3 as earlyclobber

commit 16ed2601ad0a4aa82f11e9df86ea92183f94f979
Author: H.J. Lu <hongjiu.lu@intel.com>
Date:   Wed May 15 15:26:19 2019 +0000

    i386: Emulate MMX pshufb with SSE version

has

+(define_insn_and_split "ssse3_pshufbv8qi3"
+  [(set (match_operand:V8QI 0 "register_operand" "=y,x,Yv")
+  (unspec:V8QI [(match_operand:V8QI 1 "register_operand" "0,0,Yv")
+           (match_operand:V8QI 2 "register_mmxmem_operand" "ym,x,Yv")]
+          UNSPEC_PSHUFB))
+   (clobber (match_scratch:V4SI 3 "=X,x,Yv"))]
                                       ^^^  There are earlyclobber.
+  "(TARGET_MMX || TARGET_MMX_WITH_SSE) && TARGET_SSSE3"
+  "@
+   pshufb\t{%2, %0|%0, %2}
+   #
+   #"
+  "TARGET_MMX_WITH_SSE && reload_completed"
+  [(set (match_dup 3) (match_dup 5))
+   (set (match_dup 3)
+  (and:V4SI (match_dup 3) (match_dup 2)))
+   (set (match_dup 0)
+  (unspec:V16QI [(match_dup 1) (match_dup 4)] UNSPEC_PSHUFB))]

If input register operand 2 is dead after this insn, RA may choose it
as scratch operand.  Since it isn't marked as earlyclobber, operand 2
becomes unused after split and then it gets optimized out.  Mark scratch
operand as earlyclobber fixes the issue.

gcc/

	PR target/94467
	* config/i386/sse.md (ssse3_pshufbv8qi3): Mark scratch operand
	as earlyclobber.

gcc/testsuite/

	PR target/94467
	* gcc.target/i386/pr94467-1.c: New test.
	* gcc.target/i386/pr94467-2.c: Likewise.

5 files changed, 101 insertions, 1 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index e9dfa71..6317e38 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,9 @@
+2020-04-03  H.J. Lu  <hongjiu.lu@intel.com>
+
+	PR target/94467
+	* config/i386/sse.md (ssse3_pshufbv8qi3): Mark scratch operand
+	as earlyclobber.
+
 2020-04-03  Jeff Law  <law@redhat.com>
 
 	PR rtl-optimization/92264
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index 24b3acd..fef6065 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -16695,7 +16695,7 @@
 	(unspec:V8QI [(match_operand:V8QI 1 "register_operand" "0,0,Yv")
 		      (match_operand:V8QI 2 "register_mmxmem_operand" "ym,x,Yv")]
 		     UNSPEC_PSHUFB))
-   (clobber (match_scratch:V4SI 3 "=X,x,Yv"))]
+   (clobber (match_scratch:V4SI 3 "=X,&x,&Yv"))]
   "(TARGET_MMX || TARGET_MMX_WITH_SSE) && TARGET_SSSE3"
   "@
    pshufb\t{%2, %0|%0, %2}
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 67a14db..aad627d 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,9 @@
+2020-04-03  H.J. Lu  <hongjiu.lu@intel.com>
+
+	PR target/94467
+	* gcc.target/i386/pr94467-1.c: New test.
+	* gcc.target/i386/pr94467-2.c: Likewise.
+
 2020-04-03  Jakub Jelinek  <jakub@redhat.com>
 
 	PR target/94460
diff --git a/gcc/testsuite/gcc.target/i386/pr94467-1.c b/gcc/testsuite/gcc.target/i386/pr94467-1.c
new file mode 100644
index 0000000..a51c3a8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr94467-1.c
@@ -0,0 +1,40 @@
+/* { dg-do run } */
+/* { dg-require-effective-target avx } */
+/* { dg-options "-O -mavx" } */
+
+#include "avx-check.h"
+
+typedef char __attribute__ ((__vector_size__ (8))) v8qi;
+typedef short __attribute__ ((__vector_size__ (8))) v4hi;
+typedef int __attribute__ ((__vector_size__ (8))) v2si;
+typedef long long __attribute__ ((__vector_size__ (8))) v1di;
+typedef unsigned long long u64;
+u64 k, c;
+
+v8qi g, h, p, q;
+v4hi d, e, f, l, n, o;
+v2si j;
+
+u64
+foo (v4hi r)
+{
+  v8qi s;
+  f = (v4hi) j;
+  e = __builtin_ia32_psrlwi ((v4hi) k, c);
+  s = __builtin_ia32_pavgb (h, h);
+  n = __builtin_ia32_pabsw (f);
+  o = __builtin_ia32_psubusw (n, l);
+  p = __builtin_ia32_packsswb (r, o);
+  q = __builtin_ia32_pshufb (p, s);
+  g = __builtin_ia32_punpcklbw (q, (v8qi) r);
+  d = r;
+  return (u64) g + (u64) h + (u64) j;
+}
+
+static void
+avx_test (void)
+{
+  u64 x = foo ((v4hi) { 5 });
+  if (x != 0x0005000500050505)
+    __builtin_abort ();
+}
diff --git a/gcc/testsuite/gcc.target/i386/pr94467-2.c b/gcc/testsuite/gcc.target/i386/pr94467-2.c
new file mode 100644
index 0000000..8128be3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr94467-2.c
@@ -0,0 +1,48 @@
+/* { dg-do run } */
+/* { dg-require-effective-target ssse3 } */
+/* { dg-options "-O -mssse3" } */
+
+#ifndef CHECK_H
+#define CHECK_H "ssse3-check.h"
+#endif
+
+#ifndef TEST
+#define TEST ssse3_test
+#endif
+
+#include CHECK_H
+
+typedef char __attribute__ ((__vector_size__ (8))) v8qi;
+typedef short __attribute__ ((__vector_size__ (8))) v4hi;
+typedef int __attribute__ ((__vector_size__ (8))) v2si;
+typedef long long __attribute__ ((__vector_size__ (8))) v1di;
+typedef unsigned long long u64;
+u64 k, c;
+
+v8qi g, h, p, q;
+v4hi d, e, f, l, n, o;
+v2si j;
+
+u64
+foo (v4hi r)
+{
+  v8qi s;
+  f = (v4hi) j;
+  e = __builtin_ia32_psrlwi ((v4hi) k, c);
+  s = __builtin_ia32_pavgb (h, h);
+  n = __builtin_ia32_pabsw (f);
+  o = __builtin_ia32_psubusw (n, l);
+  p = __builtin_ia32_packsswb (r, o);
+  q = __builtin_ia32_pshufb (p, s);
+  g = __builtin_ia32_punpcklbw (q, (v8qi) r);
+  d = r;
+  return (u64) g + (u64) h + (u64) j;
+}
+
+static void
+ssse3_test (void)
+{
+  u64 x = foo ((v4hi) { 5 });
+  if (x != 0x0005000500050505)
+    __builtin_abort ();
+}