[PATCH, i386]: Fix PR 53712, Does not combine unaligned load with _mm_cmpistri

[Uros Bizjak]

Hello!

Attached patch teaches gcc how to merge unaligned load (UNSPEC_MOVU)
with pcmpistr/pcmpestr instructions.

2012-06-18  Uros Bizjak  <ubizjak@gmail.com>

	PR target/53712
	* config/i386/sse.md (*sse4_2_pcmpestr_unaligned): New.
	(*sse4_2_pcmpistr_unaligned): New.

testsuite/ChangeLog:

2012-06-18  Uros Bizjak  <ubizjak@gmail.com>

	PR target/53712
	* gcc.target/i386/pr53712.c: New test.

Bootstrapped and regression tested on x86_64-pc-linux-gnu {,-m32},
also with followup lex.c patch that substitutes asm with builtins.

Committed to mainline SVN.

Uros.
-------------- next part --------------
Index: testsuite/gcc.target/i386/pr53712.c
===================================================================
--- testsuite/gcc.target/i386/pr53712.c	(revision 0)
+++ testsuite/gcc.target/i386/pr53712.c	(working copy)
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -msse4.2" } */
+
+typedef char v16qi __attribute__ ((__vector_size__ (16)));
+
+int test (const char *s1, const char *s2)
+{
+  v16qi s1chars = __builtin_ia32_loaddqu ((const char *) s2);
+  v16qi s2chars = __builtin_ia32_loaddqu ((const char *) s1);
+  return __builtin_ia32_pcmpistri128 (s1chars, s2chars, 0);
+}
+
+/* { dg-final { scan-assembler-times "movdqu" 1 } } */
Index: config/i386/sse.md
===================================================================
--- config/i386/sse.md	(revision 188736)
+++ config/i386/sse.md	(working copy)
@@ -9679,6 +9679,68 @@
    (set_attr "memory" "none,load")
    (set_attr "mode" "TI")])
 
+(define_insn_and_split "*sse4_2_pcmpestr_unaligned"
+  [(set (match_operand:SI 0 "register_operand" "=c")
+	(unspec:SI
+	  [(match_operand:V16QI 2 "reg_not_xmm0_operand" "x")
+	   (match_operand:SI 3 "register_operand" "a")
+	   (unspec:V16QI
+	     [(match_operand:V16QI 4 "memory_operand" "m")]
+	     UNSPEC_MOVU)
+	   (match_operand:SI 5 "register_operand" "d")
+	   (match_operand:SI 6 "const_0_to_255_operand" "n")]
+	  UNSPEC_PCMPESTR))
+   (set (match_operand:V16QI 1 "register_operand" "=Yz")
+	(unspec:V16QI
+	  [(match_dup 2)
+	   (match_dup 3)
+	   (unspec:V16QI [(match_dup 4)] UNSPEC_MOVU)
+	   (match_dup 5)
+	   (match_dup 6)]
+	  UNSPEC_PCMPESTR))
+   (set (reg:CC FLAGS_REG)
+	(unspec:CC
+	  [(match_dup 2)
+	   (match_dup 3)
+	   (unspec:V16QI [(match_dup 4)] UNSPEC_MOVU)
+	   (match_dup 5)
+	   (match_dup 6)]
+	  UNSPEC_PCMPESTR))]
+  "TARGET_SSE4_2
+   && can_create_pseudo_p ()"
+  "#"
+  "&& 1"
+  [(const_int 0)]
+{
+  int ecx = !find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[0]));
+  int xmm0 = !find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[1]));
+  int flags = !find_regno_note (curr_insn, REG_UNUSED, FLAGS_REG);
+
+  if (ecx)
+    emit_insn (gen_sse4_2_pcmpestri (operands[0], operands[2],
+				     operands[3], operands[4],
+				     operands[5], operands[6]));
+  if (xmm0)
+    emit_insn (gen_sse4_2_pcmpestrm (operands[1], operands[2],
+				     operands[3], operands[4],
+				     operands[5], operands[6]));
+  if (flags && !(ecx || xmm0))
+    emit_insn (gen_sse4_2_pcmpestr_cconly (NULL, NULL,
+					   operands[2], operands[3],
+					   operands[4], operands[5],
+					   operands[6]));
+  if (!(flags || ecx || xmm0))
+    emit_note (NOTE_INSN_DELETED);
+
+  DONE;
+}
+  [(set_attr "type" "sselog")
+   (set_attr "prefix_data16" "1")
+   (set_attr "prefix_extra" "1")
+   (set_attr "length_immediate" "1")
+   (set_attr "memory" "load")
+   (set_attr "mode" "TI")])
+
 (define_insn "sse4_2_pcmpestri"
   [(set (match_operand:SI 0 "register_operand" "=c,c")
 	(unspec:SI
@@ -9809,6 +9871,59 @@
    (set_attr "memory" "none,load")
    (set_attr "mode" "TI")])
 
+(define_insn_and_split "*sse4_2_pcmpistr_unaligned"
+  [(set (match_operand:SI 0 "register_operand" "=c")
+	(unspec:SI
+	  [(match_operand:V16QI 2 "reg_not_xmm0_operand" "x")
+	   (unspec:V16QI
+	     [(match_operand:V16QI 3 "memory_operand" "m")]
+	     UNSPEC_MOVU)
+	   (match_operand:SI 4 "const_0_to_255_operand" "n")]
+	  UNSPEC_PCMPISTR))
+   (set (match_operand:V16QI 1 "register_operand" "=Yz")
+	(unspec:V16QI
+	  [(match_dup 2)
+	   (unspec:V16QI [(match_dup 3)] UNSPEC_MOVU)
+	   (match_dup 4)]
+	  UNSPEC_PCMPISTR))
+   (set (reg:CC FLAGS_REG)
+	(unspec:CC
+	  [(match_dup 2)
+	   (unspec:V16QI [(match_dup 3)] UNSPEC_MOVU)
+	   (match_dup 4)]
+	  UNSPEC_PCMPISTR))]
+  "TARGET_SSE4_2
+   && can_create_pseudo_p ()"
+  "#"
+  "&& 1"
+  [(const_int 0)]
+{
+  int ecx = !find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[0]));
+  int xmm0 = !find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[1]));
+  int flags = !find_regno_note (curr_insn, REG_UNUSED, FLAGS_REG);
+
+  if (ecx)
+    emit_insn (gen_sse4_2_pcmpistri (operands[0], operands[2],
+				     operands[3], operands[4]));
+  if (xmm0)
+    emit_insn (gen_sse4_2_pcmpistrm (operands[1], operands[2],
+				     operands[3], operands[4]));
+  if (flags && !(ecx || xmm0))
+    emit_insn (gen_sse4_2_pcmpistr_cconly (NULL, NULL,
+					   operands[2], operands[3],
+					   operands[4]));
+  if (!(flags || ecx || xmm0))
+    emit_note (NOTE_INSN_DELETED);
+
+  DONE;
+}
+  [(set_attr "type" "sselog")
+   (set_attr "prefix_data16" "1")
+   (set_attr "prefix_extra" "1")
+   (set_attr "length_immediate" "1")
+   (set_attr "memory" "load")
+   (set_attr "mode" "TI")])
+
 (define_insn "sse4_2_pcmpistri"
   [(set (match_operand:SI 0 "register_operand" "=c,c")
 	(unspec:SI