* [PATCH] AArch64: Improve rotate patterns
@ 2021-12-08 16:44 Wilco Dijkstra
2021-12-08 17:18 ` Richard Sandiford
0 siblings, 1 reply; 2+ messages in thread
From: Wilco Dijkstra @ 2021-12-08 16:44 UTC (permalink / raw)
To: GCC Patches; +Cc: Richard Sandiford, Kyrylo Tkachov
Improve and generalize rotate patterns. Rotates by more than half the
bitwidth of a register are canonicalized to rotate left. Many existing
shift patterns don't handle this case correctly, so add rotate left to
the shift iterator and convert rotate left into ror during assembly
output. Add missing zero_extend patterns for shifted BIC, ORN and EON.
Passes bootstrap and regress. OK for commit?
ChangeLog:
2021-12-07 Wilco Dijkstra <wdijkstr@arm.com>
* config/aarch64/aarch64.md
(and_<SHIFT:optab><mode>3_compare0): Support rotate left.
(and_<SHIFT:optab>si3_compare0_uxtw): Likewise.
(<LOGICAL:optab>_<SHIFT:optab><mode>3): Likewise.
(<LOGICAL:optab>_<SHIFT:optab>si3_uxtw): Likewise.
(one_cmpl_<optab><mode>2): Likewise.
(<LOGICAL:optab>_one_cmpl_<SHIFT:optab><mode>3): Likewise.
(<LOGICAL:optab>_one_cmpl_<SHIFT:optab>sidi_uxtw): New pattern.
(eor_one_cmpl_<SHIFT:optab><mode>3_alt): Support rotate left.
(eor_one_cmpl_<SHIFT:optab>sidi3_alt_ze): Likewise.
(and_one_cmpl_<SHIFT:optab><mode>3_compare0): Likewise.
(and_one_cmpl_<SHIFT:optab>si3_compare0_uxtw): Likewise.
(and_one_cmpl_<SHIFT:optab><mode>3_compare0_no_reuse): Likewise.
(and_<SHIFT:optab><mode>3nr_compare0): Likewise.
(*<optab>si3_insn_uxtw): Use SHIFT_no_rotate.
(rolsi3_insn_uxtw): New pattern.
* config/aarch64/iterators.md (SHIFT): Add rotate left.
(SHIFT_no_rotate): Add new iterator.
(SHIFT:shift): Print rotate left as ror.
(is_rotl): Add test for left rotate.
* gcc.target/aarch64/ror_2.c: New test.
* gcc.target/aarch64/ror_3.c: New test.
---
diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
index 5297b2d3f95744ac72e36814c6676cc97478d48b..f80679bcb34ea07918b1a0304b32be436568095d 100644
--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -4419,7 +4419,11 @@ (define_insn "*and_<SHIFT:optab><mode>3_compare0"
(set (match_operand:GPI 0 "register_operand" "=r")
(and:GPI (SHIFT:GPI (match_dup 1) (match_dup 2)) (match_dup 3)))]
""
- "ands\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
+ return "ands\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logics_shift_imm")]
)
@@ -4436,7 +4440,11 @@ (define_insn "*and_<SHIFT:optab>si3_compare0_uxtw"
(zero_extend:DI (and:SI (SHIFT:SI (match_dup 1) (match_dup 2))
(match_dup 3))))]
""
- "ands\\t%w0, %w3, %w1, <SHIFT:shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
+ return "ands\\t%w0, %w3, %w1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logics_shift_imm")]
)
@@ -4447,7 +4455,11 @@ (define_insn "*<LOGICAL:optab>_<SHIFT:optab><mode>3"
(match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))
(match_operand:GPI 3 "register_operand" "r")))]
""
- "<LOGICAL:logical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
+ return "<LOGICAL:logical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logic_shift_imm")]
)
@@ -4504,17 +4516,6 @@ (define_split
"operands[3] = gen_reg_rtx (<MODE>mode);"
)
-(define_insn "*<optab>_rol<mode>3"
- [(set (match_operand:GPI 0 "register_operand" "=r")
- (LOGICAL:GPI (rotate:GPI
- (match_operand:GPI 1 "register_operand" "r")
- (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))
- (match_operand:GPI 3 "register_operand" "r")))]
- ""
- "<logical>\\t%<w>0, %<w>3, %<w>1, ror #(<sizen> - %2)"
- [(set_attr "type" "logic_shift_imm")]
-)
-
;; zero_extend versions of above
(define_insn "*<LOGICAL:optab>_<SHIFT:optab>si3_uxtw"
[(set (match_operand:DI 0 "register_operand" "=r")
@@ -4524,19 +4525,11 @@ (define_insn "*<LOGICAL:optab>_<SHIFT:optab>si3_uxtw"
(match_operand:QI 2 "aarch64_shift_imm_si" "n"))
(match_operand:SI 3 "register_operand" "r"))))]
""
- "<LOGICAL:logical>\\t%w0, %w3, %w1, <SHIFT:shift> %2"
- [(set_attr "type" "logic_shift_imm")]
-)
-
-(define_insn "*<optab>_rolsi3_uxtw"
- [(set (match_operand:DI 0 "register_operand" "=r")
- (zero_extend:DI
- (LOGICAL:SI (rotate:SI
- (match_operand:SI 1 "register_operand" "r")
- (match_operand:QI 2 "aarch64_shift_imm_si" "n"))
- (match_operand:SI 3 "register_operand" "r"))))]
- ""
- "<logical>\\t%w0, %w3, %w1, ror #(32 - %2)"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
+ return "<LOGICAL:logical>\\t%w0, %w3, %w1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logic_shift_imm")]
)
@@ -4565,7 +4558,11 @@ (define_insn "*one_cmpl_<optab><mode>2"
(not:GPI (SHIFT:GPI (match_operand:GPI 1 "register_operand" "r")
(match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))))]
""
- "mvn\\t%<w>0, %<w>1, <shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
+ return "mvn\\t%<w>0, %<w>1, <shift> %2";
+}
[(set_attr "type" "logic_shift_imm")]
)
@@ -4672,7 +4669,28 @@ (define_insn "<LOGICAL:optab>_one_cmpl_<SHIFT:optab><mode>3"
(match_operand:QI 2 "aarch64_shift_imm_<mode>" "n")))
(match_operand:GPI 3 "register_operand" "r")))]
""
- "<LOGICAL:nlogical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
+ return "<LOGICAL:nlogical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
+}
+ [(set_attr "type" "logic_shift_imm")]
+)
+
+;; Zero-extend version of the above.
+(define_insn "<LOGICAL:optab>_one_cmpl_<SHIFT:optab>sidi_uxtw"
+ [(set (match_operand:DI 0 "register_operand" "=r")
+ (zero_extend:DI (LOGICAL:SI (not:SI
+ (SHIFT:SI
+ (match_operand:SI 1 "register_operand" "r")
+ (match_operand:QI 2 "aarch64_shift_imm_si" "n")))
+ (match_operand:SI 3 "register_operand" "r"))))]
+ ""
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
+ return "<LOGICAL:nlogical>\\t%w0, %w3, %w1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logic_shift_imm")]
)
@@ -4684,7 +4702,11 @@ (define_insn "*eor_one_cmpl_<SHIFT:optab><mode>3_alt"
(match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))
(match_operand:GPI 3 "register_operand" "r"))))]
""
- "eon\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
+ return "eon\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logic_shift_imm")]
)
@@ -4698,7 +4720,11 @@ (define_insn "*eor_one_cmpl_<SHIFT:optab>sidi3_alt_ze"
(match_operand:QI 2 "aarch64_shift_imm_si" "n"))
(match_operand:SI 3 "register_operand" "r")))))]
""
- "eon\\t%w0, %w3, %w1, <SHIFT:shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
+ return "eon\\t%w0, %w3, %w1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logic_shift_imm")]
)
@@ -4716,7 +4742,11 @@ (define_insn "*and_one_cmpl_<SHIFT:optab><mode>3_compare0"
(SHIFT:GPI
(match_dup 1) (match_dup 2))) (match_dup 3)))]
""
- "bics\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
+ return "bics\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logics_shift_imm")]
)
@@ -4735,7 +4765,11 @@ (define_insn "*and_one_cmpl_<SHIFT:optab>si3_compare0_uxtw"
(not:SI
(SHIFT:SI (match_dup 1) (match_dup 2))) (match_dup 3))))]
""
- "bics\\t%w0, %w3, %w1, <SHIFT:shift> %2"
+{
+ if (<SHIFT:is_rotl>)
+ operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
+ return "bics\\t%w0, %w3, %w1, <SHIFT:shift> %2";
+}
[(set_attr "type" "logics_shift_imm")]
)
@@ -4749,7 +4783,11 @@ (define_insn "*and_one_cmpl_<SHIFT:optab><mode>3_compare0_no_reuse"
(match_operand:GPI 2 "register_operand" "r"))
(const_int 0)))]
""
- "bics\\t<w>zr, %<w>2, %<w>0, <SHIFT:shift> %1"
+{
+ if (<SHIFT:is_rotl>)
+ operands[1] = GEN_INT (<sizen> - UINTVAL (operands[1]));
+ return "bics\\t<w>zr, %<w>2, %<w>0, <SHIFT:shift> %1";
+}
[(set_attr "type" "logics_shift_imm")]
)
@@ -4923,7 +4961,11 @@ (define_insn "*and_<SHIFT:optab><mode>3nr_compare0"
(match_operand:GPI 2 "register_operand" "r"))
(const_int 0)))]
""
- "tst\\t%<w>2, %<w>0, <SHIFT:shift> %1"
+{
+ if (<SHIFT:is_rotl>)
+ operands[1] = GEN_INT (<sizen> - UINTVAL (operands[1]));
+ return "tst\\t%<w>2, %<w>0, <SHIFT:shift> %1";
+}
[(set_attr "type" "logics_shift_imm")]
)
@@ -5324,10 +5366,22 @@ (define_insn "*ror<mode>3_insn"
[(set_attr "type" "rotate_imm,shift_reg")]
)
-;; zero_extend version of above
+(define_insn "*rol<mode>3_insn"
+ [(set (match_operand:GPI 0 "register_operand" "=r")
+ (rotate:GPI (match_operand:GPI 1 "register_operand" "r")
+ (match_operand 2 "const_int_operand" "n")))]
+ "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<MODE>mode)"
+{
+ operands[3] = GEN_INT (<sizen> - UINTVAL (operands[2]));
+ return "ror\\t%<w>0, %<w>1, %3";
+}
+ [(set_attr "type" "rotate_imm")]
+)
+
+;; zero_extend version of shifts
(define_insn "*<optab>si3_insn_uxtw"
[(set (match_operand:DI 0 "register_operand" "=r,r")
- (zero_extend:DI (SHIFT:SI
+ (zero_extend:DI (SHIFT_no_rotate:SI
(match_operand:SI 1 "register_operand" "r,r")
(match_operand:QI 2 "aarch64_reg_or_shift_imm_si" "Uss,r"))))]
""
@@ -5337,6 +5391,31 @@ (define_insn "*<optab>si3_insn_uxtw"
[(set_attr "type" "bfx,shift_reg")]
)
+;; zero_extend version of rotate right
+(define_insn "*rorsi3_insn_uxtw"
+ [(set (match_operand:DI 0 "register_operand" "=r")
+ (zero_extend:DI
+ (rotatert:SI (match_operand:SI 1 "register_operand" "r")
+ (match_operand 2 "const_int_operand" "n"))))]
+ "UINTVAL (operands[2]) < 32"
+ "ror\\t%w0, %w1, %2"
+ [(set_attr "type" "rotate_imm")]
+)
+
+;; zero_extend version of rotate left
+(define_insn "*rolsi3_insn_uxtw"
+ [(set (match_operand:DI 0 "register_operand" "=r")
+ (zero_extend:DI
+ (rotate:SI (match_operand:SI 1 "register_operand" "r")
+ (match_operand 2 "const_int_operand" "n"))))]
+ "UINTVAL (operands[2]) < 32"
+{
+ operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
+ return "ror\\t%w0, %w1, %2";
+}
+ [(set_attr "type" "rotate_imm")]
+)
+
(define_insn "*<optab><mode>3_insn"
[(set (match_operand:SHORT 0 "register_operand" "=r")
(ASHIFT:SHORT (match_operand:SHORT 1 "register_operand" "r")
@@ -5419,32 +5498,6 @@ (define_insn "*extrsi5_insn_di"
[(set_attr "type" "rotate_imm")]
)
-(define_insn "*ror<mode>3_insn"
- [(set (match_operand:GPI 0 "register_operand" "=r")
- (rotate:GPI (match_operand:GPI 1 "register_operand" "r")
- (match_operand 2 "const_int_operand" "n")))]
- "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<MODE>mode)"
-{
- operands[3] = GEN_INT (<sizen> - UINTVAL (operands[2]));
- return "ror\\t%<w>0, %<w>1, %3";
-}
- [(set_attr "type" "rotate_imm")]
-)
-
-;; zero_extend version of the above
-(define_insn "*rorsi3_insn_uxtw"
- [(set (match_operand:DI 0 "register_operand" "=r")
- (zero_extend:DI
- (rotate:SI (match_operand:SI 1 "register_operand" "r")
- (match_operand 2 "const_int_operand" "n"))))]
- "UINTVAL (operands[2]) < 32"
-{
- operands[3] = GEN_INT (32 - UINTVAL (operands[2]));
- return "ror\\t%w0, %w1, %3";
-}
- [(set_attr "type" "rotate_imm")]
-)
-
(define_insn "*<ANY_EXTEND:optab><GPI:mode>_ashl<SHORT:mode>"
[(set (match_operand:GPI 0 "register_operand" "=r")
(ANY_EXTEND:GPI
diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
index 0b34e7f7d16c2cc08e69026543e93858eee0f5dd..4bbe8057f406d15019319ab73f6573ee11237a57 100644
--- a/gcc/config/aarch64/iterators.md
+++ b/gcc/config/aarch64/iterators.md
@@ -2083,7 +2083,10 @@ (define_mode_attr sve_lane_pair_con [(VNx8HF "y") (VNx4SF "x")])
;; -------------------------------------------------------------------
;; This code iterator allows the various shifts supported on the core
-(define_code_iterator SHIFT [ashift ashiftrt lshiftrt rotatert])
+(define_code_iterator SHIFT [ashift ashiftrt lshiftrt rotatert rotate])
+
+;; This code iterator allows all shifts except for rotates.
+(define_code_iterator SHIFT_no_rotate [ashift ashiftrt lshiftrt])
;; This code iterator allows the shifts supported in arithmetic instructions
(define_code_iterator ASHIFT [ashift ashiftrt lshiftrt])
@@ -2210,6 +2213,7 @@ (define_code_attr optab [(ashift "ashl")
(ashiftrt "ashr")
(lshiftrt "lshr")
(rotatert "rotr")
+ (rotate "rotl")
(sign_extend "extend")
(zero_extend "zero_extend")
(sign_extract "extv")
@@ -2299,7 +2303,10 @@ (define_code_attr su_optab [(sign_extend "") (zero_extend "u")
;; Similar for the instruction mnemonics
(define_code_attr shift [(ashift "lsl") (ashiftrt "asr")
- (lshiftrt "lsr") (rotatert "ror")])
+ (lshiftrt "lsr") (rotatert "ror") (rotate "ror")])
+;; True if shift is rotate left.
+(define_code_attr is_rotl [(ashift "0") (ashiftrt "0")
+ (lshiftrt "0") (rotatert "0") (rotate "1")])
;; Op prefix for shift right and accumulate.
(define_code_attr sra_op [(ashiftrt "s") (lshiftrt "u")])
diff --git a/gcc/testsuite/gcc.target/aarch64/ror_2.c b/gcc/testsuite/gcc.target/aarch64/ror_2.c
new file mode 100644
index 0000000000000000000000000000000000000000..796c1222230e49f81999f3913f483288ff9ff6e8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/ror_2.c
@@ -0,0 +1,205 @@
+/* { dg-options "-O2 --save-temps" } */
+/* { dg-do assemble } */
+
+
+#define ROR(X,Y) ((X >> Y) | (X << (32 - Y)))
+
+unsigned
+ror1 (unsigned x)
+{
+ /* { dg-final { scan-assembler "ror\tw\[0-9\]+, w\[0-9\]+, 3\n" } } */
+ return ROR (x, 3);
+}
+
+unsigned
+ror2 (unsigned x)
+{
+ /* { dg-final { scan-assembler "ror\tw\[0-9\]+, w\[0-9\]+, 17\n" } } */
+ return ROR (x, 17);
+}
+
+unsigned long
+ror3 (unsigned x)
+{
+ /* { dg-final { scan-assembler "ror\tw\[0-9\]+, w\[0-9\]+, 2\n" } } */
+ return (unsigned long) ROR (x, 2);
+}
+
+unsigned long
+ror4 (unsigned x)
+{
+ /* { dg-final { scan-assembler "ror\tw\[0-9\]+, w\[0-9\]+, 26\n" } } */
+ return (unsigned long) ROR (x, 26);
+}
+
+unsigned
+and1 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "and\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 13\n" } } */
+ return x & ROR (y, 13);
+}
+
+unsigned
+and2 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "and\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 23\n" } } */
+ return x & ROR (y, 23);
+}
+
+unsigned long
+and3 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "and\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 12\n" } } */
+ return x & (unsigned long) ROR (y, 12);
+}
+
+unsigned
+bic1 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "bic\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 11\n" } } */
+ return x & ~ROR (y, 11);
+}
+
+unsigned
+bic2 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "bic\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 17\n" } } */
+ return x & ~ROR (y, 17);
+}
+
+unsigned long
+bic3 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "bic\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 22\n" } } */
+ return (unsigned long) x & ~ROR (y, 22);
+}
+
+unsigned
+orr1 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "orr\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 5\n" } } */
+ return x | ROR (y, 5);
+}
+
+unsigned
+orr2 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "orr\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 25\n" } } */
+ return x | ROR (y, 25);
+}
+
+unsigned long
+orr3 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "orr\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 24\n" } } */
+ return (unsigned long)x | ROR (y, 24);
+}
+
+unsigned
+orn1 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "orn\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 3\n" } } */
+ return x | ~ROR (y, 3);
+}
+
+unsigned
+orn2 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "orn\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 30\n" } } */
+ return x | ~ROR (y, 30);
+}
+
+unsigned long
+orn3 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "orn\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 9\n" } } */
+ return x | (unsigned long) ~ROR (y, 9);
+}
+
+unsigned
+eor1 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "eor\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 9\n" } } */
+ return x ^ ROR (y, 9);
+}
+
+unsigned
+eor2 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "eor\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 31\n" } } */
+ return x ^ ROR (y, 31);
+}
+
+unsigned long
+eor3 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "eor\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 28\n" } } */
+ return (unsigned long) x ^ ROR (y, 28);
+}
+
+unsigned
+eon1 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "eon\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 1\n" } } */
+ return x ^ ~ROR (y, 1);
+}
+
+unsigned
+eon2 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "eon\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 18\n" } } */
+ return x ^ ~ROR (y, 18);
+}
+
+unsigned long
+eon3 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "eon\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 19\n" } } */
+ return x ^ (unsigned long) ~ROR (y, 19);
+}
+
+int
+tst1 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "tst\tw\[0-9\]+, w\[0-9\]+, ror 8\n" } } */
+ return (x & ROR (y, 8)) == 0;
+}
+
+int
+tst2 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "tst\tw\[0-9\]+, w\[0-9\]+, ror 20\n" } } */
+ return (x & ROR (y, 20)) == 0;
+}
+
+int
+tst3 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "tst\tw\[0-9\]+, w\[0-9\]+, ror 20\n" } } */
+ return ((unsigned long)x & ROR (y, 20)) == 0;
+}
+
+int
+bics1 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "bics\twzr, w\[0-9\]+, w\[0-9\]+, ror 10\n" } } */
+ return (x & ~ROR (y, 10)) == 0;
+}
+
+int
+bics2 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "bics\twzr, w\[0-9\]+, w\[0-9\]+, ror 21\n" } } */
+ return (x & ~ROR (y, 21)) == 0;
+}
+
+int
+bics3 (unsigned x, unsigned y)
+{
+ /* { dg-final { scan-assembler "bics\twzr, w\[0-9\]+, w\[0-9\]+, ror 21\n" } } */
+ return (x & (unsigned long)~ROR (y, 21)) == 0;
+}
+
+/* { dg-final { scan-assembler-not "cmp" } } */
+/* { dg-final { scan-assembler-not "mvn" } } */
+/* { dg-final { scan-assembler-not "uxtw" } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/ror_3.c b/gcc/testsuite/gcc.target/aarch64/ror_3.c
new file mode 100644
index 0000000000000000000000000000000000000000..484beecf886b6ff94d84b12cfdffe5e7d858f437
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/ror_3.c
@@ -0,0 +1,131 @@
+/* { dg-options "-O2 --save-temps" } */
+/* { dg-do assemble } */
+
+
+#define ROR(X,Y) ((X >> Y) | (X << (64 - Y)))
+
+unsigned long
+ror1 (unsigned long x)
+{
+ /* { dg-final { scan-assembler "ror\tx\[0-9\]+, x\[0-9\]+, 3\n" } } */
+ return ROR (x, 3);
+}
+
+unsigned long
+ror2 (unsigned long x)
+{
+ /* { dg-final { scan-assembler "ror\tx\[0-9\]+, x\[0-9\]+, 37\n" } } */
+ return ROR (x, 37);
+}
+
+unsigned long
+and1 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "and\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 13\n" } } */
+ return x & ROR (y, 13);
+}
+
+unsigned long
+and2 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "and\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 33\n" } } */
+ return x & ROR (y, 33);
+}
+
+unsigned long
+bic1 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "bic\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 11\n" } } */
+ return x & ~ROR (y, 11);
+}
+
+unsigned long
+bic2 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "bic\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 47\n" } } */
+ return x & ~ROR (y, 47);
+}
+
+unsigned long
+orr1 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "orr\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 5\n" } } */
+ return x | ROR (y, 5);
+}
+
+unsigned long
+orr2 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "orr\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 35\n" } } */
+ return x | ROR (y, 35);
+}
+
+unsigned long
+orn1 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "orn\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 3\n" } } */
+ return x | ~ROR (y, 3);
+}
+
+unsigned long
+orn2 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "orn\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 39\n" } } */
+ return x | ~ROR (y, 39);
+}
+
+unsigned long
+eor1 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "eor\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 9\n" } } */
+ return x ^ ROR (y, 9);
+}
+
+unsigned long
+eor2 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "eor\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 41\n" } } */
+ return x ^ ROR (y, 41);
+}
+
+unsigned long
+eon1 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "eon\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 1\n" } } */
+ return x ^ ~ROR (y, 1);
+}
+
+unsigned long
+eon2 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "eon\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 38\n" } } */
+ return x ^ ~ROR (y, 38);
+}
+
+unsigned long
+tst1 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "tst\tx\[0-9\]+, x\[0-9\]+, ror 8\n" } } */
+ return (x & ROR (y, 8)) == 0;
+}
+
+unsigned long
+tst2 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "tst\tx\[0-9\]+, x\[0-9\]+, ror 50\n" } } */
+ return (x & ROR (y, 50)) == 0;
+}
+
+unsigned long
+bics1 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "bics\txzr, x\[0-9\]+, x\[0-9\]+, ror 10\n" } } */
+ return (x & ~ROR (y, 10)) == 0;
+}
+
+unsigned long
+bics2 (unsigned long x, unsigned long y)
+{
+ /* { dg-final { scan-assembler "bics\txzr, x\[0-9\]+, x\[0-9\]+, ror 62\n" } } */
+ return (x & ~ROR (y, 62)) == 0;
+}
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] AArch64: Improve rotate patterns
2021-12-08 16:44 [PATCH] AArch64: Improve rotate patterns Wilco Dijkstra
@ 2021-12-08 17:18 ` Richard Sandiford
0 siblings, 0 replies; 2+ messages in thread
From: Richard Sandiford @ 2021-12-08 17:18 UTC (permalink / raw)
To: Wilco Dijkstra; +Cc: GCC Patches, Kyrylo Tkachov
Wilco Dijkstra <Wilco.Dijkstra@arm.com> writes:
> Improve and generalize rotate patterns. Rotates by more than half the
> bitwidth of a register are canonicalized to rotate left. Many existing
> shift patterns don't handle this case correctly, so add rotate left to
> the shift iterator and convert rotate left into ror during assembly
> output. Add missing zero_extend patterns for shifted BIC, ORN and EON.
>
> Passes bootstrap and regress. OK for commit?
OK, thanks. I agree it should go in during stage 3 since handling
rotatert by a constant without rotate by a constant is a bug,
like you say. And the port is supposed to be complete wrt.
zero_extend patterns.
FTR, the is_rotl isn't strictly necessary, since we could I think do:
if (<SHIFT:CODE> == ROTATE)
But the fact that ROTATE is the left shift isn't very mnemonic,
so I agree the version in the patch is better.
Richard
>
> ChangeLog:
> 2021-12-07 Wilco Dijkstra <wdijkstr@arm.com>
>
> * config/aarch64/aarch64.md
> (and_<SHIFT:optab><mode>3_compare0): Support rotate left.
> (and_<SHIFT:optab>si3_compare0_uxtw): Likewise.
> (<LOGICAL:optab>_<SHIFT:optab><mode>3): Likewise.
> (<LOGICAL:optab>_<SHIFT:optab>si3_uxtw): Likewise.
> (one_cmpl_<optab><mode>2): Likewise.
> (<LOGICAL:optab>_one_cmpl_<SHIFT:optab><mode>3): Likewise.
> (<LOGICAL:optab>_one_cmpl_<SHIFT:optab>sidi_uxtw): New pattern.
> (eor_one_cmpl_<SHIFT:optab><mode>3_alt): Support rotate left.
> (eor_one_cmpl_<SHIFT:optab>sidi3_alt_ze): Likewise.
> (and_one_cmpl_<SHIFT:optab><mode>3_compare0): Likewise.
> (and_one_cmpl_<SHIFT:optab>si3_compare0_uxtw): Likewise.
> (and_one_cmpl_<SHIFT:optab><mode>3_compare0_no_reuse): Likewise.
> (and_<SHIFT:optab><mode>3nr_compare0): Likewise.
> (*<optab>si3_insn_uxtw): Use SHIFT_no_rotate.
> (rolsi3_insn_uxtw): New pattern.
> * config/aarch64/iterators.md (SHIFT): Add rotate left.
> (SHIFT_no_rotate): Add new iterator.
> (SHIFT:shift): Print rotate left as ror.
> (is_rotl): Add test for left rotate.
>
> * gcc.target/aarch64/ror_2.c: New test.
> * gcc.target/aarch64/ror_3.c: New test.
>
> ---
>
> diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
> index 5297b2d3f95744ac72e36814c6676cc97478d48b..f80679bcb34ea07918b1a0304b32be436568095d 100644
> --- a/gcc/config/aarch64/aarch64.md
> +++ b/gcc/config/aarch64/aarch64.md
> @@ -4419,7 +4419,11 @@ (define_insn "*and_<SHIFT:optab><mode>3_compare0"
> (set (match_operand:GPI 0 "register_operand" "=r")
> (and:GPI (SHIFT:GPI (match_dup 1) (match_dup 2)) (match_dup 3)))]
> ""
> - "ands\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
> + return "ands\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logics_shift_imm")]
> )
>
> @@ -4436,7 +4440,11 @@ (define_insn "*and_<SHIFT:optab>si3_compare0_uxtw"
> (zero_extend:DI (and:SI (SHIFT:SI (match_dup 1) (match_dup 2))
> (match_dup 3))))]
> ""
> - "ands\\t%w0, %w3, %w1, <SHIFT:shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
> + return "ands\\t%w0, %w3, %w1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logics_shift_imm")]
> )
>
> @@ -4447,7 +4455,11 @@ (define_insn "*<LOGICAL:optab>_<SHIFT:optab><mode>3"
> (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))
> (match_operand:GPI 3 "register_operand" "r")))]
> ""
> - "<LOGICAL:logical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
> + return "<LOGICAL:logical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logic_shift_imm")]
> )
>
> @@ -4504,17 +4516,6 @@ (define_split
> "operands[3] = gen_reg_rtx (<MODE>mode);"
> )
>
> -(define_insn "*<optab>_rol<mode>3"
> - [(set (match_operand:GPI 0 "register_operand" "=r")
> - (LOGICAL:GPI (rotate:GPI
> - (match_operand:GPI 1 "register_operand" "r")
> - (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))
> - (match_operand:GPI 3 "register_operand" "r")))]
> - ""
> - "<logical>\\t%<w>0, %<w>3, %<w>1, ror #(<sizen> - %2)"
> - [(set_attr "type" "logic_shift_imm")]
> -)
> -
> ;; zero_extend versions of above
> (define_insn "*<LOGICAL:optab>_<SHIFT:optab>si3_uxtw"
> [(set (match_operand:DI 0 "register_operand" "=r")
> @@ -4524,19 +4525,11 @@ (define_insn "*<LOGICAL:optab>_<SHIFT:optab>si3_uxtw"
> (match_operand:QI 2 "aarch64_shift_imm_si" "n"))
> (match_operand:SI 3 "register_operand" "r"))))]
> ""
> - "<LOGICAL:logical>\\t%w0, %w3, %w1, <SHIFT:shift> %2"
> - [(set_attr "type" "logic_shift_imm")]
> -)
> -
> -(define_insn "*<optab>_rolsi3_uxtw"
> - [(set (match_operand:DI 0 "register_operand" "=r")
> - (zero_extend:DI
> - (LOGICAL:SI (rotate:SI
> - (match_operand:SI 1 "register_operand" "r")
> - (match_operand:QI 2 "aarch64_shift_imm_si" "n"))
> - (match_operand:SI 3 "register_operand" "r"))))]
> - ""
> - "<logical>\\t%w0, %w3, %w1, ror #(32 - %2)"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
> + return "<LOGICAL:logical>\\t%w0, %w3, %w1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logic_shift_imm")]
> )
>
> @@ -4565,7 +4558,11 @@ (define_insn "*one_cmpl_<optab><mode>2"
> (not:GPI (SHIFT:GPI (match_operand:GPI 1 "register_operand" "r")
> (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))))]
> ""
> - "mvn\\t%<w>0, %<w>1, <shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
> + return "mvn\\t%<w>0, %<w>1, <shift> %2";
> +}
> [(set_attr "type" "logic_shift_imm")]
> )
>
> @@ -4672,7 +4669,28 @@ (define_insn "<LOGICAL:optab>_one_cmpl_<SHIFT:optab><mode>3"
> (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n")))
> (match_operand:GPI 3 "register_operand" "r")))]
> ""
> - "<LOGICAL:nlogical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
> + return "<LOGICAL:nlogical>\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
> +}
> + [(set_attr "type" "logic_shift_imm")]
> +)
> +
> +;; Zero-extend version of the above.
> +(define_insn "<LOGICAL:optab>_one_cmpl_<SHIFT:optab>sidi_uxtw"
> + [(set (match_operand:DI 0 "register_operand" "=r")
> + (zero_extend:DI (LOGICAL:SI (not:SI
> + (SHIFT:SI
> + (match_operand:SI 1 "register_operand" "r")
> + (match_operand:QI 2 "aarch64_shift_imm_si" "n")))
> + (match_operand:SI 3 "register_operand" "r"))))]
> + ""
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
> + return "<LOGICAL:nlogical>\\t%w0, %w3, %w1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logic_shift_imm")]
> )
>
> @@ -4684,7 +4702,11 @@ (define_insn "*eor_one_cmpl_<SHIFT:optab><mode>3_alt"
> (match_operand:QI 2 "aarch64_shift_imm_<mode>" "n"))
> (match_operand:GPI 3 "register_operand" "r"))))]
> ""
> - "eon\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
> + return "eon\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logic_shift_imm")]
> )
>
> @@ -4698,7 +4720,11 @@ (define_insn "*eor_one_cmpl_<SHIFT:optab>sidi3_alt_ze"
> (match_operand:QI 2 "aarch64_shift_imm_si" "n"))
> (match_operand:SI 3 "register_operand" "r")))))]
> ""
> - "eon\\t%w0, %w3, %w1, <SHIFT:shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
> + return "eon\\t%w0, %w3, %w1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logic_shift_imm")]
> )
>
> @@ -4716,7 +4742,11 @@ (define_insn "*and_one_cmpl_<SHIFT:optab><mode>3_compare0"
> (SHIFT:GPI
> (match_dup 1) (match_dup 2))) (match_dup 3)))]
> ""
> - "bics\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (<sizen> - UINTVAL (operands[2]));
> + return "bics\\t%<w>0, %<w>3, %<w>1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logics_shift_imm")]
> )
>
> @@ -4735,7 +4765,11 @@ (define_insn "*and_one_cmpl_<SHIFT:optab>si3_compare0_uxtw"
> (not:SI
> (SHIFT:SI (match_dup 1) (match_dup 2))) (match_dup 3))))]
> ""
> - "bics\\t%w0, %w3, %w1, <SHIFT:shift> %2"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
> + return "bics\\t%w0, %w3, %w1, <SHIFT:shift> %2";
> +}
> [(set_attr "type" "logics_shift_imm")]
> )
>
> @@ -4749,7 +4783,11 @@ (define_insn "*and_one_cmpl_<SHIFT:optab><mode>3_compare0_no_reuse"
> (match_operand:GPI 2 "register_operand" "r"))
> (const_int 0)))]
> ""
> - "bics\\t<w>zr, %<w>2, %<w>0, <SHIFT:shift> %1"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[1] = GEN_INT (<sizen> - UINTVAL (operands[1]));
> + return "bics\\t<w>zr, %<w>2, %<w>0, <SHIFT:shift> %1";
> +}
> [(set_attr "type" "logics_shift_imm")]
> )
>
> @@ -4923,7 +4961,11 @@ (define_insn "*and_<SHIFT:optab><mode>3nr_compare0"
> (match_operand:GPI 2 "register_operand" "r"))
> (const_int 0)))]
> ""
> - "tst\\t%<w>2, %<w>0, <SHIFT:shift> %1"
> +{
> + if (<SHIFT:is_rotl>)
> + operands[1] = GEN_INT (<sizen> - UINTVAL (operands[1]));
> + return "tst\\t%<w>2, %<w>0, <SHIFT:shift> %1";
> +}
> [(set_attr "type" "logics_shift_imm")]
> )
>
> @@ -5324,10 +5366,22 @@ (define_insn "*ror<mode>3_insn"
> [(set_attr "type" "rotate_imm,shift_reg")]
> )
>
> -;; zero_extend version of above
> +(define_insn "*rol<mode>3_insn"
> + [(set (match_operand:GPI 0 "register_operand" "=r")
> + (rotate:GPI (match_operand:GPI 1 "register_operand" "r")
> + (match_operand 2 "const_int_operand" "n")))]
> + "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<MODE>mode)"
> +{
> + operands[3] = GEN_INT (<sizen> - UINTVAL (operands[2]));
> + return "ror\\t%<w>0, %<w>1, %3";
> +}
> + [(set_attr "type" "rotate_imm")]
> +)
> +
> +;; zero_extend version of shifts
> (define_insn "*<optab>si3_insn_uxtw"
> [(set (match_operand:DI 0 "register_operand" "=r,r")
> - (zero_extend:DI (SHIFT:SI
> + (zero_extend:DI (SHIFT_no_rotate:SI
> (match_operand:SI 1 "register_operand" "r,r")
> (match_operand:QI 2 "aarch64_reg_or_shift_imm_si" "Uss,r"))))]
> ""
> @@ -5337,6 +5391,31 @@ (define_insn "*<optab>si3_insn_uxtw"
> [(set_attr "type" "bfx,shift_reg")]
> )
>
> +;; zero_extend version of rotate right
> +(define_insn "*rorsi3_insn_uxtw"
> + [(set (match_operand:DI 0 "register_operand" "=r")
> + (zero_extend:DI
> + (rotatert:SI (match_operand:SI 1 "register_operand" "r")
> + (match_operand 2 "const_int_operand" "n"))))]
> + "UINTVAL (operands[2]) < 32"
> + "ror\\t%w0, %w1, %2"
> + [(set_attr "type" "rotate_imm")]
> +)
> +
> +;; zero_extend version of rotate left
> +(define_insn "*rolsi3_insn_uxtw"
> + [(set (match_operand:DI 0 "register_operand" "=r")
> + (zero_extend:DI
> + (rotate:SI (match_operand:SI 1 "register_operand" "r")
> + (match_operand 2 "const_int_operand" "n"))))]
> + "UINTVAL (operands[2]) < 32"
> +{
> + operands[2] = GEN_INT (32 - UINTVAL (operands[2]));
> + return "ror\\t%w0, %w1, %2";
> +}
> + [(set_attr "type" "rotate_imm")]
> +)
> +
> (define_insn "*<optab><mode>3_insn"
> [(set (match_operand:SHORT 0 "register_operand" "=r")
> (ASHIFT:SHORT (match_operand:SHORT 1 "register_operand" "r")
> @@ -5419,32 +5498,6 @@ (define_insn "*extrsi5_insn_di"
> [(set_attr "type" "rotate_imm")]
> )
>
> -(define_insn "*ror<mode>3_insn"
> - [(set (match_operand:GPI 0 "register_operand" "=r")
> - (rotate:GPI (match_operand:GPI 1 "register_operand" "r")
> - (match_operand 2 "const_int_operand" "n")))]
> - "UINTVAL (operands[2]) < GET_MODE_BITSIZE (<MODE>mode)"
> -{
> - operands[3] = GEN_INT (<sizen> - UINTVAL (operands[2]));
> - return "ror\\t%<w>0, %<w>1, %3";
> -}
> - [(set_attr "type" "rotate_imm")]
> -)
> -
> -;; zero_extend version of the above
> -(define_insn "*rorsi3_insn_uxtw"
> - [(set (match_operand:DI 0 "register_operand" "=r")
> - (zero_extend:DI
> - (rotate:SI (match_operand:SI 1 "register_operand" "r")
> - (match_operand 2 "const_int_operand" "n"))))]
> - "UINTVAL (operands[2]) < 32"
> -{
> - operands[3] = GEN_INT (32 - UINTVAL (operands[2]));
> - return "ror\\t%w0, %w1, %3";
> -}
> - [(set_attr "type" "rotate_imm")]
> -)
> -
> (define_insn "*<ANY_EXTEND:optab><GPI:mode>_ashl<SHORT:mode>"
> [(set (match_operand:GPI 0 "register_operand" "=r")
> (ANY_EXTEND:GPI
> diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
> index 0b34e7f7d16c2cc08e69026543e93858eee0f5dd..4bbe8057f406d15019319ab73f6573ee11237a57 100644
> --- a/gcc/config/aarch64/iterators.md
> +++ b/gcc/config/aarch64/iterators.md
> @@ -2083,7 +2083,10 @@ (define_mode_attr sve_lane_pair_con [(VNx8HF "y") (VNx4SF "x")])
> ;; -------------------------------------------------------------------
>
> ;; This code iterator allows the various shifts supported on the core
> -(define_code_iterator SHIFT [ashift ashiftrt lshiftrt rotatert])
> +(define_code_iterator SHIFT [ashift ashiftrt lshiftrt rotatert rotate])
> +
> +;; This code iterator allows all shifts except for rotates.
> +(define_code_iterator SHIFT_no_rotate [ashift ashiftrt lshiftrt])
>
> ;; This code iterator allows the shifts supported in arithmetic instructions
> (define_code_iterator ASHIFT [ashift ashiftrt lshiftrt])
> @@ -2210,6 +2213,7 @@ (define_code_attr optab [(ashift "ashl")
> (ashiftrt "ashr")
> (lshiftrt "lshr")
> (rotatert "rotr")
> + (rotate "rotl")
> (sign_extend "extend")
> (zero_extend "zero_extend")
> (sign_extract "extv")
> @@ -2299,7 +2303,10 @@ (define_code_attr su_optab [(sign_extend "") (zero_extend "u")
>
> ;; Similar for the instruction mnemonics
> (define_code_attr shift [(ashift "lsl") (ashiftrt "asr")
> - (lshiftrt "lsr") (rotatert "ror")])
> + (lshiftrt "lsr") (rotatert "ror") (rotate "ror")])
> +;; True if shift is rotate left.
> +(define_code_attr is_rotl [(ashift "0") (ashiftrt "0")
> + (lshiftrt "0") (rotatert "0") (rotate "1")])
>
> ;; Op prefix for shift right and accumulate.
> (define_code_attr sra_op [(ashiftrt "s") (lshiftrt "u")])
> diff --git a/gcc/testsuite/gcc.target/aarch64/ror_2.c b/gcc/testsuite/gcc.target/aarch64/ror_2.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..796c1222230e49f81999f3913f483288ff9ff6e8
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/ror_2.c
> @@ -0,0 +1,205 @@
> +/* { dg-options "-O2 --save-temps" } */
> +/* { dg-do assemble } */
> +
> +
> +#define ROR(X,Y) ((X >> Y) | (X << (32 - Y)))
> +
> +unsigned
> +ror1 (unsigned x)
> +{
> + /* { dg-final { scan-assembler "ror\tw\[0-9\]+, w\[0-9\]+, 3\n" } } */
> + return ROR (x, 3);
> +}
> +
> +unsigned
> +ror2 (unsigned x)
> +{
> + /* { dg-final { scan-assembler "ror\tw\[0-9\]+, w\[0-9\]+, 17\n" } } */
> + return ROR (x, 17);
> +}
> +
> +unsigned long
> +ror3 (unsigned x)
> +{
> + /* { dg-final { scan-assembler "ror\tw\[0-9\]+, w\[0-9\]+, 2\n" } } */
> + return (unsigned long) ROR (x, 2);
> +}
> +
> +unsigned long
> +ror4 (unsigned x)
> +{
> + /* { dg-final { scan-assembler "ror\tw\[0-9\]+, w\[0-9\]+, 26\n" } } */
> + return (unsigned long) ROR (x, 26);
> +}
> +
> +unsigned
> +and1 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "and\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 13\n" } } */
> + return x & ROR (y, 13);
> +}
> +
> +unsigned
> +and2 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "and\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 23\n" } } */
> + return x & ROR (y, 23);
> +}
> +
> +unsigned long
> +and3 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "and\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 12\n" } } */
> + return x & (unsigned long) ROR (y, 12);
> +}
> +
> +unsigned
> +bic1 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "bic\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 11\n" } } */
> + return x & ~ROR (y, 11);
> +}
> +
> +unsigned
> +bic2 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "bic\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 17\n" } } */
> + return x & ~ROR (y, 17);
> +}
> +
> +unsigned long
> +bic3 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "bic\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 22\n" } } */
> + return (unsigned long) x & ~ROR (y, 22);
> +}
> +
> +unsigned
> +orr1 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "orr\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 5\n" } } */
> + return x | ROR (y, 5);
> +}
> +
> +unsigned
> +orr2 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "orr\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 25\n" } } */
> + return x | ROR (y, 25);
> +}
> +
> +unsigned long
> +orr3 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "orr\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 24\n" } } */
> + return (unsigned long)x | ROR (y, 24);
> +}
> +
> +unsigned
> +orn1 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "orn\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 3\n" } } */
> + return x | ~ROR (y, 3);
> +}
> +
> +unsigned
> +orn2 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "orn\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 30\n" } } */
> + return x | ~ROR (y, 30);
> +}
> +
> +unsigned long
> +orn3 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "orn\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 9\n" } } */
> + return x | (unsigned long) ~ROR (y, 9);
> +}
> +
> +unsigned
> +eor1 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "eor\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 9\n" } } */
> + return x ^ ROR (y, 9);
> +}
> +
> +unsigned
> +eor2 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "eor\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 31\n" } } */
> + return x ^ ROR (y, 31);
> +}
> +
> +unsigned long
> +eor3 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "eor\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 28\n" } } */
> + return (unsigned long) x ^ ROR (y, 28);
> +}
> +
> +unsigned
> +eon1 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "eon\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 1\n" } } */
> + return x ^ ~ROR (y, 1);
> +}
> +
> +unsigned
> +eon2 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "eon\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 18\n" } } */
> + return x ^ ~ROR (y, 18);
> +}
> +
> +unsigned long
> +eon3 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "eon\tw\[0-9\]+, w\[0-9\]+, w\[0-9\]+, ror 19\n" } } */
> + return x ^ (unsigned long) ~ROR (y, 19);
> +}
> +
> +int
> +tst1 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "tst\tw\[0-9\]+, w\[0-9\]+, ror 8\n" } } */
> + return (x & ROR (y, 8)) == 0;
> +}
> +
> +int
> +tst2 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "tst\tw\[0-9\]+, w\[0-9\]+, ror 20\n" } } */
> + return (x & ROR (y, 20)) == 0;
> +}
> +
> +int
> +tst3 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "tst\tw\[0-9\]+, w\[0-9\]+, ror 20\n" } } */
> + return ((unsigned long)x & ROR (y, 20)) == 0;
> +}
> +
> +int
> +bics1 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "bics\twzr, w\[0-9\]+, w\[0-9\]+, ror 10\n" } } */
> + return (x & ~ROR (y, 10)) == 0;
> +}
> +
> +int
> +bics2 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "bics\twzr, w\[0-9\]+, w\[0-9\]+, ror 21\n" } } */
> + return (x & ~ROR (y, 21)) == 0;
> +}
> +
> +int
> +bics3 (unsigned x, unsigned y)
> +{
> + /* { dg-final { scan-assembler "bics\twzr, w\[0-9\]+, w\[0-9\]+, ror 21\n" } } */
> + return (x & (unsigned long)~ROR (y, 21)) == 0;
> +}
> +
> +/* { dg-final { scan-assembler-not "cmp" } } */
> +/* { dg-final { scan-assembler-not "mvn" } } */
> +/* { dg-final { scan-assembler-not "uxtw" } } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/ror_3.c b/gcc/testsuite/gcc.target/aarch64/ror_3.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..484beecf886b6ff94d84b12cfdffe5e7d858f437
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/ror_3.c
> @@ -0,0 +1,131 @@
> +/* { dg-options "-O2 --save-temps" } */
> +/* { dg-do assemble } */
> +
> +
> +#define ROR(X,Y) ((X >> Y) | (X << (64 - Y)))
> +
> +unsigned long
> +ror1 (unsigned long x)
> +{
> + /* { dg-final { scan-assembler "ror\tx\[0-9\]+, x\[0-9\]+, 3\n" } } */
> + return ROR (x, 3);
> +}
> +
> +unsigned long
> +ror2 (unsigned long x)
> +{
> + /* { dg-final { scan-assembler "ror\tx\[0-9\]+, x\[0-9\]+, 37\n" } } */
> + return ROR (x, 37);
> +}
> +
> +unsigned long
> +and1 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "and\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 13\n" } } */
> + return x & ROR (y, 13);
> +}
> +
> +unsigned long
> +and2 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "and\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 33\n" } } */
> + return x & ROR (y, 33);
> +}
> +
> +unsigned long
> +bic1 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "bic\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 11\n" } } */
> + return x & ~ROR (y, 11);
> +}
> +
> +unsigned long
> +bic2 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "bic\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 47\n" } } */
> + return x & ~ROR (y, 47);
> +}
> +
> +unsigned long
> +orr1 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "orr\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 5\n" } } */
> + return x | ROR (y, 5);
> +}
> +
> +unsigned long
> +orr2 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "orr\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 35\n" } } */
> + return x | ROR (y, 35);
> +}
> +
> +unsigned long
> +orn1 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "orn\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 3\n" } } */
> + return x | ~ROR (y, 3);
> +}
> +
> +unsigned long
> +orn2 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "orn\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 39\n" } } */
> + return x | ~ROR (y, 39);
> +}
> +
> +unsigned long
> +eor1 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "eor\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 9\n" } } */
> + return x ^ ROR (y, 9);
> +}
> +
> +unsigned long
> +eor2 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "eor\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 41\n" } } */
> + return x ^ ROR (y, 41);
> +}
> +
> +unsigned long
> +eon1 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "eon\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 1\n" } } */
> + return x ^ ~ROR (y, 1);
> +}
> +
> +unsigned long
> +eon2 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "eon\tx\[0-9\]+, x\[0-9\]+, x\[0-9\]+, ror 38\n" } } */
> + return x ^ ~ROR (y, 38);
> +}
> +
> +unsigned long
> +tst1 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "tst\tx\[0-9\]+, x\[0-9\]+, ror 8\n" } } */
> + return (x & ROR (y, 8)) == 0;
> +}
> +
> +unsigned long
> +tst2 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "tst\tx\[0-9\]+, x\[0-9\]+, ror 50\n" } } */
> + return (x & ROR (y, 50)) == 0;
> +}
> +
> +unsigned long
> +bics1 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "bics\txzr, x\[0-9\]+, x\[0-9\]+, ror 10\n" } } */
> + return (x & ~ROR (y, 10)) == 0;
> +}
> +
> +unsigned long
> +bics2 (unsigned long x, unsigned long y)
> +{
> + /* { dg-final { scan-assembler "bics\txzr, x\[0-9\]+, x\[0-9\]+, ror 62\n" } } */
> + return (x & ~ROR (y, 62)) == 0;
> +}
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-12-08 17:18 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-12-08 16:44 [PATCH] AArch64: Improve rotate patterns Wilco Dijkstra
2021-12-08 17:18 ` Richard Sandiford
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).