i386: x r<< (c - y) to x r>> y etc. optimization [PR117930]

The following patch optimizes x r<< (c - y) to x r>> y,
x r>> (c - y) to x r<< y, x r<< (c + y) to x r<< y and
x r>> (c + y) to x r>> y if c is a multiple of x's bitsize.

2024-12-07  Jakub Jelinek  <jakub@redhat.com>

	PR target/117930
	* config/i386/i386.md (crotate): New define_code_attr.
	(*<insn><mode>3_add, *<insn><mode>3_add_1,
	*<insn><mode>3_sub, *<insn><mode>3_sub_1): New define_insn_and_split
	patterns plus following define_split for constant first input
	operand.

	* gcc.target/i386/pr117930.c: New test.

2 files changed, 259 insertions, 0 deletions

diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index ec816be..6edcb6d 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -1079,6 +1079,9 @@
 ;; Base name for insn mnemonic.
 (define_code_attr rotate [(rotate "rol") (rotatert "ror")])
 
+;; Counter rotate.
+(define_code_attr crotate [(rotate "rotatert") (rotatert "rotate")])
+
 ;; Mapping of abs neg operators
 (define_code_iterator absneg [abs neg])
 
@@ -18216,6 +18219,144 @@
        (any_rotate:SWI (match_dup 4) (match_dup 2)))]
  "operands[4] = gen_reg_rtx (<MODE>mode);")
 
+(define_insn_and_split "*<insn><mode>3_add"
+  [(set (match_operand:SWI 0 "nonimmediate_operand")
+	(any_rotate:SWI
+	  (match_operand:SWI 1 "nonimmediate_operand")
+	  (subreg:QI
+	    (plus
+	      (match_operand 2 "int_nonimmediate_operand")
+	      (match_operand 3 "const_int_operand")) 0)))
+   (clobber (reg:CC FLAGS_REG))]
+  "ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)
+   && (INTVAL (operands[3]) & (<MODE_SIZE> * BITS_PER_UNIT - 1)) == 0
+   && ix86_pre_reload_split ()"
+  "#"
+  "&& 1"
+  [(parallel
+     [(set (match_dup 0)
+	   (any_rotate:SWI (match_dup 1) (match_dup 2)))
+      (clobber (reg:CC FLAGS_REG))])]
+{
+  operands[2] = force_reg (GET_MODE (operands[2]), operands[2]);
+  operands[2] = gen_lowpart (QImode, operands[2]);
+})
+
+(define_split
+  [(set (match_operand:SWI 0 "register_operand")
+	(any_rotate:SWI
+	  (match_operand:SWI 1 "const_int_operand")
+	  (subreg:QI
+	    (plus
+	      (match_operand 2 "int248_register_operand")
+	      (match_operand 3 "const_int_operand")) 0)))]
+ "(INTVAL (operands[3]) & (<MODE_SIZE> * BITS_PER_UNIT - 1)) == 0"
+ [(set (match_dup 4) (match_dup 1))
+  (set (match_dup 0)
+       (any_rotate:SWI (match_dup 4) (subreg:QI (match_dup 2) 0)))]
+ "operands[4] = gen_reg_rtx (<MODE>mode);")
+
+(define_insn_and_split "*<insn><mode>3_add_1"
+  [(set (match_operand:SWI 0 "nonimmediate_operand")
+	(any_rotate:SWI
+	  (match_operand:SWI 1 "nonimmediate_operand")
+	  (plus:QI
+	    (match_operand:QI 2 "nonimmediate_operand")
+	    (match_operand:QI 3 "const_int_operand"))))
+   (clobber (reg:CC FLAGS_REG))]
+  "ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)
+   && (INTVAL (operands[3]) & (<MODE_SIZE> * BITS_PER_UNIT - 1)) == 0
+   && ix86_pre_reload_split ()"
+  "#"
+  "&& 1"
+  [(parallel
+     [(set (match_dup 0)
+	   (any_rotate:SWI (match_dup 1) (match_dup 2)))
+      (clobber (reg:CC FLAGS_REG))])]
+ "operands[2] = force_reg (GET_MODE (operands[2]), operands[2]);")
+
+(define_split
+  [(set (match_operand:SWI 0 "register_operand")
+	(any_rotate:SWI
+	  (match_operand:SWI 1 "const_int_operand")
+	  (plus:QI
+	    (match_operand:QI 2 "register_operand")
+	    (match_operand:QI 3 "const_int_operand"))))]
+ "(INTVAL (operands[3]) & (<MODE_SIZE> * BITS_PER_UNIT - 1)) == 0"
+ [(set (match_dup 4) (match_dup 1))
+  (set (match_dup 0)
+       (any_rotate:SWI (match_dup 4) (match_dup 2)))]
+ "operands[4] = gen_reg_rtx (<MODE>mode);")
+
+(define_insn_and_split "*<insn><mode>3_sub"
+  [(set (match_operand:SWI 0 "nonimmediate_operand")
+	(any_rotate:SWI
+	  (match_operand:SWI 1 "nonimmediate_operand")
+	  (subreg:QI
+	    (minus
+	      (match_operand 3 "const_int_operand")
+	      (match_operand 2 "int_nonimmediate_operand")) 0)))
+   (clobber (reg:CC FLAGS_REG))]
+  "ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)
+   && (INTVAL (operands[3]) & (<MODE_SIZE> * BITS_PER_UNIT - 1)) == 0
+   && ix86_pre_reload_split ()"
+  "#"
+  "&& 1"
+  [(parallel
+     [(set (match_dup 0)
+	   (<crotate>:SWI (match_dup 1) (match_dup 2)))
+      (clobber (reg:CC FLAGS_REG))])]
+{
+  operands[2] = force_reg (GET_MODE (operands[2]), operands[2]);
+  operands[2] = gen_lowpart (QImode, operands[2]);
+})
+
+(define_split
+  [(set (match_operand:SWI 0 "register_operand")
+	(any_rotate:SWI
+	  (match_operand:SWI 1 "const_int_operand")
+	  (subreg:QI
+	    (minus
+	      (match_operand 3 "const_int_operand")
+	      (match_operand 2 "int248_register_operand")) 0)))]
+ "(INTVAL (operands[3]) & (<MODE_SIZE> * BITS_PER_UNIT - 1)) == 0"
+ [(set (match_dup 4) (match_dup 1))
+  (set (match_dup 0)
+       (<crotate>:SWI (match_dup 4) (subreg:QI (match_dup 2) 0)))]
+ "operands[4] = gen_reg_rtx (<MODE>mode);")
+
+(define_insn_and_split "*<insn><mode>3_sub_1"
+  [(set (match_operand:SWI 0 "nonimmediate_operand")
+	(any_rotate:SWI
+	  (match_operand:SWI 1 "nonimmediate_operand")
+	  (minus:QI
+	    (match_operand:QI 3 "const_int_operand")
+	    (match_operand:QI 2 "nonimmediate_operand"))))
+   (clobber (reg:CC FLAGS_REG))]
+  "ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)
+   && (INTVAL (operands[3]) & (<MODE_SIZE> * BITS_PER_UNIT - 1)) == 0
+   && ix86_pre_reload_split ()"
+  "#"
+  "&& 1"
+  [(parallel
+     [(set (match_dup 0)
+	   (<crotate>:SWI (match_dup 1) (match_dup 2)))
+      (clobber (reg:CC FLAGS_REG))])]
+ "operands[2] = force_reg (GET_MODE (operands[2]), operands[2]);")
+
+(define_split
+  [(set (match_operand:SWI 0 "register_operand")
+	(any_rotate:SWI
+	  (match_operand:SWI 1 "const_int_operand")
+	  (minus:QI
+	    (match_operand:QI 3 "const_int_operand")
+	    (match_operand:QI 2 "register_operand"))))]
+ "(INTVAL (operands[3]) & (<MODE_SIZE> * BITS_PER_UNIT - 1)) == 0"
+ [(set (match_dup 4) (match_dup 1))
+  (set (match_dup 0)
+       (<crotate>:SWI (match_dup 4) (match_dup 2)))]
+ "operands[4] = gen_reg_rtx (<MODE>mode);")
+
 ;; Implement rotation using two double-precision
 ;; shift instructions and a scratch register.
 
diff --git a/gcc/testsuite/gcc.target/i386/pr117930.c b/gcc/testsuite/gcc.target/i386/pr117930.c
new file mode 100644
index 0000000..e8dec92
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr117930.c
@@ -0,0 +1,118 @@
+/* PR target/117930 */
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+/* { dg-final { scan-assembler-not "sub\[bwlq\]\t" } } */
+/* { dg-final { scan-assembler-not "add\[bwlq\]\t" } } */
+/* { dg-final { scan-assembler-not "lea\[lq\]\t" } } */
+
+static inline
+unsigned lrotate (unsigned x, int t)
+{
+  unsigned tl = x << t;
+  unsigned th = x >> (-t & 31);
+  return tl | th;
+}
+
+static inline
+unsigned rrotate (unsigned x, int t)
+{
+  unsigned tl = x >> t;
+  unsigned th = x << (-t & 31);
+  return tl | th;
+}
+
+unsigned
+f1 (unsigned x, int t)
+{
+  return lrotate (x, 32 - t);
+}
+
+unsigned
+f2 (unsigned x, int t)
+{
+  return lrotate (x, 64 - t);
+}
+
+unsigned
+f3 (unsigned x, int t)
+{
+  return lrotate (x, 32 + t);
+}
+
+unsigned
+f4 (unsigned x, int t)
+{
+  return lrotate (x, 64 + t);
+}
+
+unsigned
+f5 (unsigned x, int t)
+{
+  return rrotate (x, 32 - t);
+}
+
+unsigned
+f6 (unsigned x, int t)
+{
+  return rrotate (x, 64 - t);
+}
+
+unsigned
+f7 (unsigned x, int t)
+{
+  return rrotate (x, 32 + t);
+}
+
+unsigned
+f8 (unsigned x, int t)
+{
+  return rrotate (x, 64 + t);
+}
+
+unsigned
+f9 (int t)
+{
+  return lrotate (0xdeadbeefU, 32 - t);
+}
+
+unsigned
+f10 (int t)
+{
+  return lrotate (0xdeadbeefU, 64 - t);
+}
+
+unsigned
+f11 (int t)
+{
+  return lrotate (0xdeadbeefU, 32 + t);
+}
+
+unsigned
+f12 (int t)
+{
+  return lrotate (0xdeadbeefU, 64 + t);
+}
+
+unsigned
+f13 (int t)
+{
+  return rrotate (0xdeadbeefU, 32 - t);
+}
+
+unsigned
+f14 (int t)
+{
+  return rrotate (0xdeadbeefU, 64 - t);
+}
+
+unsigned
+f15 (int t)
+{
+  return rrotate (0xdeadbeefU, 32 + t);
+}
+
+unsigned
+f16 (int t)
+{
+  return rrotate (0xdeadbeefU, 64 + t);
+}