diff mbox series

[committed] i386: Generate strict_low_part QImode insn with high input register

Message ID CAFULd4YSM2r-qeO6Vesy+3Z7Bh_0X=_fn++3KBPONAzvU8Z1DA@mail.gmail.com
State New
Headers show
Series [committed] i386: Generate strict_low_part QImode insn with high input register | expand

Commit Message

Uros Bizjak Nov. 14, 2023, 5:37 p.m. UTC
Following testcase:

struct S1
{
  unsigned char val;
  unsigned char pad1;
  unsigned short pad2;
};

struct S2
{
  unsigned char pad1;
  unsigned char val;
  unsigned short pad2;
};

struct S1 test_and (struct S1 a, struct S2 b)
{
  a.val &= b.val;

  return a;
}

compiles with -O2 to:

    movl    %esi, %edx
    movl    %edi, %eax
    movzbl  %dh, %esi
    andb    %sil, %al

ANDB could use high register %dh instead of %sil:

    movl    %edi, %eax
    movl    %esi, %edx
    andb    %dh, %al

Patch introduces strict_low_part QImode insn patterns with one of
its input arguments extracted from high register.

    PR target/78904

gcc/ChangeLog:

    * config/i386/i386.md (*addqi_ext<mode>_1_slp):
    New define_insn_and_split pattern.
    (*subqi_ext<mode>_1_slp): Ditto.
    (*<any_logic:code>qi_ext<mode>_1_slp): Ditto.

gcc/testsuite/ChangeLog:

    * gcc.target/i386/pr78904-7.c: New test.
    * gcc.target/i386/pr78904-7a.c: New test.
    * gcc.target/i386/pr78904-7b.c: New test.

Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.

Uros.

Comments

Uros Bizjak Nov. 15, 2023, 8:10 a.m. UTC | #1
On Tue, Nov 14, 2023 at 6:37 PM Uros Bizjak <ubizjak@gmail.com> wrote:

>     PR target/78904
>
> gcc/ChangeLog:
>
>     * config/i386/i386.md (*addqi_ext<mode>_1_slp):
>     New define_insn_and_split pattern.
>     (*subqi_ext<mode>_1_slp): Ditto.
>     (*<any_logic:code>qi_ext<mode>_1_slp): Ditto.
>
> gcc/testsuite/ChangeLog:
>
>     * gcc.target/i386/pr78904-7.c: New test.
>     * gcc.target/i386/pr78904-7a.c: New test.
>     * gcc.target/i386/pr78904-7b.c: New test.
>
> Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.

Eh, I somehow managed to mix up patches. The attached patch is also
needed to avoid testsuite ICE in gcc.c-torture/execute/pr82524.c.

Will commit it later today.

Uros.
diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index 6136e46b1bc..0a9d14e9c08 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -6626,7 +6626,7 @@ (define_insn_and_split "*addqi_ext<mode>_1_slp"
 	       (const_int 8)]) 0)
 	  (match_operand:QI 1 "nonimmediate_operand" "0,!Q")))
    (clobber (reg:CC FLAGS_REG))]
-  ""
+  "!TARGET_PARTIAL_REG_STALL || optimize_function_for_size_p (cfun)"
   "@
    add{b}\t{%h2, %0|%0, %h2}
    #"
@@ -6638,8 +6638,8 @@ (define_insn_and_split "*addqi_ext<mode>_1_slp"
 	   (plus:QI
 	     (subreg:QI
 	       (match_op_dup 3
-		 [(match_dup 0) (const_int 8) (const_int 8)]) 0)
-	   (match_dup 1)))
+		 [(match_dup 2) (const_int 8) (const_int 8)]) 0)
+	   (match_dup 0)))
       (clobber (reg:CC FLAGS_REG))])]
   ""
   [(set_attr "type" "alu")
@@ -7669,7 +7669,7 @@ (define_insn_and_split "*subqi_ext<mode>_1_slp"
 	       (const_int 8)
 	       (const_int 8)]) 0)))
    (clobber (reg:CC FLAGS_REG))]
-  ""
+  "!TARGET_PARTIAL_REG_STALL || optimize_function_for_size_p (cfun)"
   "@
    sub{b}\t{%h2, %0|%0, %h2}
    #"
@@ -7679,10 +7679,10 @@ (define_insn_and_split "*subqi_ext<mode>_1_slp"
    (parallel
      [(set (strict_low_part (match_dup 0))
 	   (minus:QI
-	   (match_dup 1)
+	     (match_dup 0)
 	     (subreg:QI
 	       (match_op_dup 3
-		 [(match_dup 0) (const_int 8) (const_int 8)]) 0)))
+		 [(match_dup 2) (const_int 8) (const_int 8)]) 0)))
       (clobber (reg:CC FLAGS_REG))])]
   ""
   [(set_attr "type" "alu")
@@ -11494,7 +11494,7 @@ (define_insn_and_split "*<code>qi_ext<mode>_1_slp"
 	       (const_int 8)]) 0)
 	  (match_operand:QI 1 "nonimmediate_operand" "0,!Q")))
    (clobber (reg:CC FLAGS_REG))]
-  ""
+  "!TARGET_PARTIAL_REG_STALL || optimize_function_for_size_p (cfun)"
   "@
    <logic>{b}\t{%h2, %0|%0, %h2}
    #"
@@ -11504,10 +11504,10 @@ (define_insn_and_split "*<code>qi_ext<mode>_1_slp"
    (parallel
      [(set (strict_low_part (match_dup 0))
 	   (any_logic:QI
-	   (match_dup 1)
 	     (subreg:QI
 	       (match_op_dup 3
-		 [(match_dup 0) (const_int 8) (const_int 8)]) 0)))
+		 [(match_dup 2) (const_int 8) (const_int 8)]) 0)))
+	   (match_dup 0)
       (clobber (reg:CC FLAGS_REG))])]
   ""
   [(set_attr "type" "alu")
diff mbox series

Patch

diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index 84cc50c7bb2..6136e46b1bc 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -6615,6 +6615,36 @@  (define_insn_and_split "*add<mode>_1_slp"
 	(const_string "alu")))
    (set_attr "mode" "<MODE>")])
 
+;; Alternative 1 is needed to work around LRA limitation, see PR82524.
+(define_insn_and_split "*addqi_ext<mode>_1_slp"
+  [(set (strict_low_part (match_operand:QI 0 "register_operand" "+Q,&Q"))
+	(plus:QI
+	  (subreg:QI
+	    (match_operator:SWI248 3 "extract_operator"
+	      [(match_operand 2 "int248_register_operand" "Q,Q")
+	       (const_int 8)
+	       (const_int 8)]) 0)
+	  (match_operand:QI 1 "nonimmediate_operand" "0,!Q")))
+   (clobber (reg:CC FLAGS_REG))]
+  ""
+  "@
+   add{b}\t{%h2, %0|%0, %h2}
+   #"
+  "reload_completed
+   && !rtx_equal_p (operands[0], operands[1])"
+  [(set (strict_low_part (match_dup 0)) (match_dup 1))
+   (parallel
+     [(set (strict_low_part (match_dup 0))
+	   (plus:QI
+	     (subreg:QI
+	       (match_op_dup 3
+		 [(match_dup 0) (const_int 8) (const_int 8)]) 0)
+	   (match_dup 1)))
+      (clobber (reg:CC FLAGS_REG))])]
+  ""
+  [(set_attr "type" "alu")
+   (set_attr "mode" "QI")])
+
 ;; Split non destructive adds if we cannot use lea.
 (define_split
   [(set (match_operand:SWI48 0 "register_operand")
@@ -7628,6 +7658,36 @@  (define_insn_and_split "*sub<mode>_1_slp"
   [(set_attr "type" "alu")
    (set_attr "mode" "<MODE>")])
 
+;; Alternative 1 is needed to work around LRA limitation, see PR82524.
+(define_insn_and_split "*subqi_ext<mode>_1_slp"
+  [(set (strict_low_part (match_operand:QI 0 "register_operand" "+Q,&Q"))
+	(minus:QI
+	  (match_operand:QI 1 "nonimmediate_operand" "0,!Q")
+	  (subreg:QI
+	    (match_operator:SWI248 3 "extract_operator"
+	      [(match_operand 2 "int248_register_operand" "Q,Q")
+	       (const_int 8)
+	       (const_int 8)]) 0)))
+   (clobber (reg:CC FLAGS_REG))]
+  ""
+  "@
+   sub{b}\t{%h2, %0|%0, %h2}
+   #"
+  "reload_completed
+   && !rtx_equal_p (operands[0], operands[1])"
+  [(set (strict_low_part (match_dup 0)) (match_dup 1))
+   (parallel
+     [(set (strict_low_part (match_dup 0))
+	   (minus:QI
+	   (match_dup 1)
+	     (subreg:QI
+	       (match_op_dup 3
+		 [(match_dup 0) (const_int 8) (const_int 8)]) 0)))
+      (clobber (reg:CC FLAGS_REG))])]
+  ""
+  [(set_attr "type" "alu")
+   (set_attr "mode" "QI")])
+
 (define_insn "*sub<mode>_2"
   [(set (reg FLAGS_REG)
 	(compare
@@ -11423,6 +11483,36 @@  (define_insn_and_split "*<code><mode>_1_slp"
   [(set_attr "type" "alu")
    (set_attr "mode" "<MODE>")])
 
+;; Alternative 1 is needed to work around LRA limitation, see PR82524.
+(define_insn_and_split "*<code>qi_ext<mode>_1_slp"
+  [(set (strict_low_part (match_operand:QI 0 "register_operand" "+Q,&Q"))
+	(any_logic:QI
+	  (subreg:QI
+	    (match_operator:SWI248 3 "extract_operator"
+	      [(match_operand 2 "int248_register_operand" "Q,Q")
+	       (const_int 8)
+	       (const_int 8)]) 0)
+	  (match_operand:QI 1 "nonimmediate_operand" "0,!Q")))
+   (clobber (reg:CC FLAGS_REG))]
+  ""
+  "@
+   <logic>{b}\t{%h2, %0|%0, %h2}
+   #"
+  "reload_completed
+   && !rtx_equal_p (operands[0], operands[1])"
+  [(set (strict_low_part (match_dup 0)) (match_dup 1))
+   (parallel
+     [(set (strict_low_part (match_dup 0))
+	   (any_logic:QI
+	   (match_dup 1)
+	     (subreg:QI
+	       (match_op_dup 3
+		 [(match_dup 0) (const_int 8) (const_int 8)]) 0)))
+      (clobber (reg:CC FLAGS_REG))])]
+  ""
+  [(set_attr "type" "alu")
+   (set_attr "mode" "QI")])
+
 (define_split
   [(set (match_operand:SWI248 0 "register_operand")
 	(and:SWI248 (match_operand:SWI248 1 "nonimmediate_operand")
diff --git a/gcc/testsuite/gcc.target/i386/pr78904-7.c b/gcc/testsuite/gcc.target/i386/pr78904-7.c
new file mode 100644
index 00000000000..e52a26992ab
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr78904-7.c
@@ -0,0 +1,64 @@ 
+/* PR target/78904 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -masm=att" } */
+/* { dg-final { scan-assembler-not "movzbl" } } */
+/* { dg-final { scan-assembler-not "movb" } } */
+
+struct S1
+{
+  unsigned char val;
+  unsigned char pad1;
+  unsigned short pad2;
+};
+
+struct S2
+{
+  unsigned char pad1;
+  unsigned char val;
+  unsigned short pad2;
+};
+
+struct S1 test_and (struct S1 a, struct S2 b)
+{
+  a.val &= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]andb" } } */
+
+struct S1 test_or (struct S1 a, struct S2 b)
+{
+  a.val |= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]orb" } } */
+
+struct S1 test_xor (struct S1 a, struct S2 b)
+{
+  a.val ^= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]xorb" } } */
+
+struct S1 test_add (struct S1 a, struct S2 b)
+{
+  a.val += b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]addb" } } */
+
+struct S1 test_sub (struct S1 a, struct S2 b)
+{
+  a.val -= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]subb" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr78904-7a.c b/gcc/testsuite/gcc.target/i386/pr78904-7a.c
new file mode 100644
index 00000000000..c060a482ba2
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr78904-7a.c
@@ -0,0 +1,62 @@ 
+/* PR target/78904 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -masm=att" } */
+/* { dg-final { scan-assembler-not "movzbl" } } */
+/* { dg-final { scan-assembler-not "movb" } } */
+
+struct S1
+{
+  unsigned char val;
+  unsigned char pad1;
+};
+
+struct S2
+{
+  unsigned char pad1;
+  unsigned char val;
+};
+
+struct S1 test_and (struct S1 a, struct S2 b)
+{
+  a.val &= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]andb" } } */
+
+struct S1 test_or (struct S1 a, struct S2 b)
+{
+  a.val |= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]orb" } } */
+
+struct S1 test_xor (struct S1 a, struct S2 b)
+{
+  a.val ^= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]xorb" } } */
+
+struct S1 test_add (struct S1 a, struct S2 b)
+{
+  a.val += b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]addb" } } */
+
+struct S1 test_sub (struct S1 a, struct S2 b)
+{
+  a.val -= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]subb" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr78904-7b.c b/gcc/testsuite/gcc.target/i386/pr78904-7b.c
new file mode 100644
index 00000000000..76dc048e73f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr78904-7b.c
@@ -0,0 +1,66 @@ 
+/* PR target/78904 */
+/* { dg-do compile { target { ! ia32 } } } */
+/* { dg-options "-O2 -masm=att" } */
+/* { dg-final { scan-assembler-not "movzbl" } } */
+/* { dg-final { scan-assembler-not "movb" } } */
+
+struct S1
+{
+  unsigned char val;
+  unsigned char pad1;
+  unsigned short pad2;
+  unsigned int pad3;
+};
+
+struct S2
+{
+  unsigned char pad1;
+  unsigned char val;
+  unsigned short pad2;
+  unsigned int pad3;
+};
+
+struct S1 test_and (struct S1 a, struct S2 b)
+{
+  a.val &= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]andb" } } */
+
+struct S1 test_or (struct S1 a, struct S2 b)
+{
+  a.val |= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]orb" } } */
+
+struct S1 test_xor (struct S1 a, struct S2 b)
+{
+  a.val ^= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]xorb" } } */
+
+struct S1 test_add (struct S1 a, struct S2 b)
+{
+  a.val += b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]addb" } } */
+
+struct S1 test_sub (struct S1 a, struct S2 b)
+{
+  a.val -= b.val;
+
+  return a;
+}
+
+/* { dg-final { scan-assembler "\[ \t\]subb" } } */