Add rules for lowering `cbranch<mode>4` to CBB<cond>/CBH<cond>/CB<cond> when
CMPBR extension is enabled.
gcc/ChangeLog:
* config/aarch64/aarch64-protos.h (aarch64_cb_rhs): New function.
* config/aarch64/aarch64.cc (aarch64_cb_rhs): Likewise.
* config/aarch64/aarch64.md (cbranch<mode>4): Rename to ...
(cbranch<GPI:mode>4): ...here, and emit CMPBR if possible.
(cbranch<SHORT:mode>4): New expand rule.
(aarch64_cb<INT_CMP:code><GPI:mode>): New insn rule.
(aarch64_cb<INT_CMP:code><SHORT:mode>): Likewise.
* config/aarch64/constraints.md (Uc0): New constraint.
(Uc1): Likewise.
(Uc2): Likewise.
* config/aarch64/iterators.md (cmpbr_suffix): New mode attr.
(INT_CMP): New code iterator.
(cmpbr_imm_constraint): New code attr.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/cmpbr.c:
---
gcc/config/aarch64/aarch64-protos.h | 2 +
gcc/config/aarch64/aarch64.cc | 33 ++
gcc/config/aarch64/aarch64.md | 95 +++-
gcc/config/aarch64/constraints.md | 18 +
gcc/config/aarch64/iterators.md | 30 ++
gcc/testsuite/gcc.target/aarch64/cmpbr.c | 587 ++++++++---------------
6 files changed, 379 insertions(+), 386 deletions(-)
diff --git a/gcc/config/aarch64/aarch64-protos.h
b/gcc/config/aarch64/aarch64-protos.h
index 31f2f5b8bd2..e946e8da11d 100644
--- a/gcc/config/aarch64/aarch64-protos.h
+++ b/gcc/config/aarch64/aarch64-protos.h
@@ -1135,6 +1135,8 @@ bool aarch64_general_check_builtin_call (location_t,
vec<location_t>,
unsigned int, tree, unsigned int,
tree *);
+bool aarch64_cb_rhs (rtx_code op_code, rtx rhs);
+
namespace aarch64 {
void report_non_ice (location_t, tree, unsigned int);
void report_out_of_range (location_t, tree, unsigned int, HOST_WIDE_INT,
diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
index 2cd03b941bd..f3ce3a15b09 100644
--- a/gcc/config/aarch64/aarch64.cc
+++ b/gcc/config/aarch64/aarch64.cc
@@ -959,6 +959,39 @@ svpattern_token (enum aarch64_svpattern pattern)
gcc_unreachable ();
}
+/* Return true if RHS is an operand suitable for a CB<cc> (immediate)
+ instruction. OP_CODE determines the type of the comparison. */
+bool
+aarch64_cb_rhs (rtx_code op_code, rtx rhs)
+{
+ if (!CONST_INT_P (rhs))
+ return REG_P (rhs);
+
+ HOST_WIDE_INT rhs_val = INTVAL (rhs);
+
+ switch (op_code)
+ {
+ case EQ:
+ case NE:
+ case GT:
+ case GTU:
+ case LT:
+ case LTU:
+ return IN_RANGE (rhs_val, 0, 63);
+
+ case GE: /* CBGE: signed greater than or equal */
+ case GEU: /* CBHS: unsigned greater than or equal */
+ return IN_RANGE (rhs_val, 1, 64);
+
+ case LE: /* CBLE: signed less than or equal */
+ case LEU: /* CBLS: unsigned less than or equal */
+ return IN_RANGE (rhs_val, -1, 62);
+
+ default:
+ return false;
+ }
+}
+
/* Return the location of a piece that is known to be passed or returned
in registers. FIRST_ZR is the first unused vector argument register
and FIRST_PR is the first unused predicate argument register. */
diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
index 1ff887a977e..32e0f739ae5 100644
--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -717,6 +717,10 @@ (define_constants
;; +/- 32KiB. Used by TBZ, TBNZ.
(BRANCH_LEN_P_32KiB 32764)
(BRANCH_LEN_N_32KiB -32768)
+
+ ;; +/- 1KiB. Used by CBB<cond>, CBH<cond>, CB<cond>.
+ (BRANCH_LEN_P_1Kib 1020)
+ (BRANCH_LEN_N_1Kib -1024)
]
)
@@ -724,18 +728,35 @@ (define_constants
;; Conditional jumps
;; -------------------------------------------------------------------
-(define_expand "cbranch<mode>4"
+(define_expand "cbranch<GPI:mode>4"
[(set (pc) (if_then_else (match_operator 0 "aarch64_comparison_operator"
[(match_operand:GPI 1 "register_operand")
(match_operand:GPI 2 "aarch64_plus_operand")])
(label_ref (match_operand 3))
(pc)))]
""
- "
- operands[1] = aarch64_gen_compare_reg (GET_CODE (operands[0]), operands[1],
- operands[2]);
- operands[2] = const0_rtx;
- "
+ {
+ if (TARGET_CMPBR && aarch64_cb_rhs (GET_CODE (operands[0]), operands[2]))
+ {
+ /* Fall-through to `aarch64_cb<INT_CMP:code><GPI:mode>`. */
+ }
+ else
+ {
+ operands[1] = aarch64_gen_compare_reg (GET_CODE (operands[0]),
+ operands[1], operands[2]);
+ operands[2] = const0_rtx;
+ }
+ }
+)
+
+(define_expand "cbranch<SHORT:mode>4"
+ [(set (pc) (if_then_else (match_operator 0 "aarch64_comparison_operator"
+ [(match_operand:SHORT 1 "register_operand")
+ (match_operand:SHORT 2 "aarch64_reg_or_zero")])
+ (label_ref (match_operand 3))
+ (pc)))]
+ "TARGET_CMPBR"
+ ""
)
(define_expand "cbranch<mode>4"
@@ -763,6 +784,68 @@ (define_expand "cbranchcc4"
""
)
+;; Emit a `CB<cond> (register)` or `CB<cond> (immediate)` instruction.
+;; The immediate range depends on the comparison code.
+;; Comparisons against immediates outside this range fall back to
+;; CMP + B<cond>.
+(define_insn "aarch64_cb<INT_CMP:code><GPI:mode>"
+ [(set (pc) (if_then_else (INT_CMP
+ (match_operand:GPI 0 "register_operand" "r")
+ (match_operand:GPI 1 "nonmemory_operand"
+ "r<INT_CMP:cmpbr_imm_constraint>"))
+ (label_ref (match_operand 2))
+ (pc)))]
+ "TARGET_CMPBR && aarch64_cb_rhs (<INT_CMP:CODE>, operands[1])"
+ {
+ if (get_attr_far_branch (insn) == FAR_BRANCH_YES)
+ return aarch64_gen_far_branch (operands, 2, "L",
+ "cb<INT_CMP:inv_cmp_op>\\t%<w>0, %<w>1,
");
+ else
+ return "cb<INT_CMP:cmp_op>\\t%<w>0, %<w>1, %l2";
+ }
+ [(set_attr "type" "branch")
+ (set (attr "length")
+ (if_then_else (and (ge (minus (match_dup 2) (pc))
+ (const_int BRANCH_LEN_N_1Kib))
+ (lt (minus (match_dup 2) (pc))
+ (const_int BRANCH_LEN_P_1Kib)))
+ (const_int 4)
+ (const_int 8)))
+ (set (attr "far_branch")
+ (if_then_else (and (ge (minus (match_dup 2) (pc))
+ (const_int BRANCH_LEN_N_1Kib))
+ (lt (minus (match_dup 2) (pc))
+ (const_int BRANCH_LEN_P_1Kib)))
+ (const_string "no")
+ (const_string "yes")))]
+)
+
+;; Emit a `CBB<cond> (register)` or `CBH<cond> (register)` instruction.
+(define_insn "aarch64_cb<INT_CMP:code><SHORT:mode>"
+ [(set (pc) (if_then_else (INT_CMP
+ (match_operand:SHORT 0 "register_operand" "r")
+ (match_operand:SHORT 1 "aarch64_reg_or_zero" "rZ"))
+ (label_ref (match_operand 2))
+ (pc)))]
+ "TARGET_CMPBR"
+ "cb<SHORT:cmpbr_suffix><INT_CMP:cmp_op>\\t%<w>0, %<w>1, %l2"
+ [(set_attr "type" "branch")
+ (set (attr "length")
+ (if_then_else (and (ge (minus (match_dup 2) (pc))
+ (const_int BRANCH_LEN_N_1Kib))
+ (lt (minus (match_dup 2) (pc))
+ (const_int BRANCH_LEN_P_1Kib)))
+ (const_int 4)
+ (const_int 8)))
+ (set (attr "far_branch")
+ (if_then_else (and (ge (minus (match_dup 2) (pc))
+ (const_int BRANCH_LEN_N_1Kib))
+ (lt (minus (match_dup 2) (pc))
+ (const_int BRANCH_LEN_P_1Kib)))
+ (const_string "no")
+ (const_string "yes")))]
+)
+
;; Emit `B<cond>`, assuming that the condition is already in the CC register.
(define_insn "aarch64_bcond"
[(set (pc) (if_then_else (match_operator 0 "aarch64_comparison_operator"
diff --git a/gcc/config/aarch64/constraints.md
b/gcc/config/aarch64/constraints.md
index e9f69f823a6..dc1925dfb6c 100644
--- a/gcc/config/aarch64/constraints.md
+++ b/gcc/config/aarch64/constraints.md
@@ -304,6 +304,24 @@ (define_constraint "Ui7"
(and (match_code "const_int")
(match_test "(unsigned HOST_WIDE_INT) ival <= 7")))
+(define_constraint "Uc0"
+ "@internal
+ A constraint that matches the integers 0...63."
+ (and (match_code "const_int")
+ (match_test "IN_RANGE (ival, 0, 63)")))
+
+(define_constraint "Uc1"
+ "@internal
+ A constraint that matches the integers 1...64."
+ (and (match_code "const_int")
+ (match_test "IN_RANGE (ival, 1, 64)")))
+
+(define_constraint "Uc2"
+ "@internal
+ A constraint that matches the integers -1...62."
+ (and (match_code "const_int")
+ (match_test "IN_RANGE (ival, -1, 62)")))
+
(define_constraint "Up3"
"@internal
A constraint that matches the integers 2^(0...4)."
diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
index a8957681357..c59fcd679d7 100644
--- a/gcc/config/aarch64/iterators.md
+++ b/gcc/config/aarch64/iterators.md
@@ -2961,6 +2961,36 @@ (define_code_attr cmp_op [(lt "lt")
(geu "hs")
(gtu "hi")])
+(define_code_attr inv_cmp_op [(lt "ge")
+ (le "gt")
+ (eq "ne")
+ (ne "eq")
+ (ge "lt")
+ (gt "le")
+ (ltu "hs")
+ (leu "hi")
+ (geu "lo")
+ (gtu "ls")])
+
+(define_mode_attr cmpbr_suffix [(QI "b") (HI "h")])
+
+(define_code_iterator INT_CMP [lt le eq ne ge gt ltu leu geu gtu])
+
+(define_code_attr cmpbr_imm_constraint [
+ (eq "Uc0")
+ (ne "Uc0")
+ (gt "Uc0")
+ (gtu "Uc0")
+ (lt "Uc0")
+ (ltu "Uc0")
+
+ (ge "Uc1")
+ (geu "Uc1")
+
+ (le "Uc2")
+ (leu "Uc2")
+])
+
(define_code_attr fix_trunc_optab [(fix "fix_trunc")
(unsigned_fix "fixuns_trunc")])
diff --git a/gcc/testsuite/gcc.target/aarch64/cmpbr.c
b/gcc/testsuite/gcc.target/aarch64/cmpbr.c
index 9ca376a8f33..0fe5641fc07 100644
--- a/gcc/testsuite/gcc.target/aarch64/cmpbr.c
+++ b/gcc/testsuite/gcc.target/aarch64/cmpbr.c
@@ -108,9 +108,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_eq_x1:
-** and w1, w1, 255
-** cmp w1, w0, uxtb
-** beq .L4
+** cbbeq w1, w0, .L4
** b not_taken
** .L4:
** b taken
@@ -118,9 +116,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ne_x1:
-** and w1, w1, 255
-** cmp w1, w0, uxtb
-** beq .L6
+** cbbeq w1, w0, .L6
** b taken
** .L6:
** b not_taken
@@ -128,9 +124,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ult_x1:
-** and w1, w1, 255
-** cmp w1, w0, uxtb
-** bls .L8
+** cbbls w1, w0, .L8
** b taken
** .L8:
** b not_taken
@@ -138,9 +132,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ule_x1:
-** and w1, w1, 255
-** cmp w1, w0, uxtb
-** bcc .L10
+** cbblo w1, w0, .L10
** b taken
** .L10:
** b not_taken
@@ -148,9 +140,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ugt_x1:
-** and w1, w1, 255
-** cmp w1, w0, uxtb
-** bcs .L12
+** cbbhs w1, w0, .L12
** b taken
** .L12:
** b not_taken
@@ -158,9 +148,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_uge_x1:
-** and w1, w1, 255
-** cmp w1, w0, uxtb
-** bhi .L14
+** cbbhi w1, w0, .L14
** b taken
** .L14:
** b not_taken
@@ -168,9 +156,7 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_slt_x1:
-** sxtb w1, w1
-** cmp w1, w0, sxtb
-** ble .L16
+** cbble w1, w0, .L16
** b taken
** .L16:
** b not_taken
@@ -178,9 +164,7 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sle_x1:
-** sxtb w1, w1
-** cmp w1, w0, sxtb
-** blt .L18
+** cbblt w1, w0, .L18
** b taken
** .L18:
** b not_taken
@@ -188,9 +172,7 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sgt_x1:
-** sxtb w1, w1
-** cmp w1, w0, sxtb
-** bge .L20
+** cbbge w1, w0, .L20
** b taken
** .L20:
** b not_taken
@@ -198,9 +180,7 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sge_x1:
-** sxtb w1, w1
-** cmp w1, w0, sxtb
-** bgt .L22
+** cbbgt w1, w0, .L22
** b taken
** .L22:
** b not_taken
@@ -208,9 +188,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_eq_x1:
-** and w1, w1, 65535
-** cmp w1, w0, uxth
-** beq .L25
+** cbheq w1, w0, .L25
** b not_taken
** .L25:
** b taken
@@ -218,9 +196,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ne_x1:
-** and w1, w1, 65535
-** cmp w1, w0, uxth
-** beq .L27
+** cbheq w1, w0, .L27
** b taken
** .L27:
** b not_taken
@@ -228,9 +204,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ult_x1:
-** and w1, w1, 65535
-** cmp w1, w0, uxth
-** bls .L29
+** cbhls w1, w0, .L29
** b taken
** .L29:
** b not_taken
@@ -238,9 +212,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ule_x1:
-** and w1, w1, 65535
-** cmp w1, w0, uxth
-** bcc .L31
+** cbhlo w1, w0, .L31
** b taken
** .L31:
** b not_taken
@@ -248,9 +220,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ugt_x1:
-** and w1, w1, 65535
-** cmp w1, w0, uxth
-** bcs .L33
+** cbhhs w1, w0, .L33
** b taken
** .L33:
** b not_taken
@@ -258,9 +228,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_uge_x1:
-** and w1, w1, 65535
-** cmp w1, w0, uxth
-** bhi .L35
+** cbhhi w1, w0, .L35
** b taken
** .L35:
** b not_taken
@@ -268,9 +236,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_slt_x1:
-** sxth w1, w1
-** cmp w1, w0, sxth
-** ble .L37
+** cbhle w1, w0, .L37
** b taken
** .L37:
** b not_taken
@@ -278,9 +244,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sle_x1:
-** sxth w1, w1
-** cmp w1, w0, sxth
-** blt .L39
+** cbhlt w1, w0, .L39
** b taken
** .L39:
** b not_taken
@@ -288,9 +252,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sgt_x1:
-** sxth w1, w1
-** cmp w1, w0, sxth
-** bge .L41
+** cbhge w1, w0, .L41
** b taken
** .L41:
** b not_taken
@@ -298,9 +260,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sge_x1:
-** sxth w1, w1
-** cmp w1, w0, sxth
-** bgt .L43
+** cbhgt w1, w0, .L43
** b taken
** .L43:
** b not_taken
@@ -308,8 +268,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_eq_x1:
-** cmp w0, w1
-** beq .L46
+** cbeq w0, w1, .L46
** b not_taken
** .L46:
** b taken
@@ -317,8 +276,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ne_x1:
-** cmp w0, w1
-** beq .L48
+** cbeq w0, w1, .L48
** b taken
** .L48:
** b not_taken
@@ -326,8 +284,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ult_x1:
-** cmp w0, w1
-** bcs .L50
+** cbhs w0, w1, .L50
** b taken
** .L50:
** b not_taken
@@ -335,8 +292,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ule_x1:
-** cmp w0, w1
-** bhi .L52
+** cbhi w0, w1, .L52
** b taken
** .L52:
** b not_taken
@@ -344,8 +300,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ugt_x1:
-** cmp w0, w1
-** bls .L54
+** cbls w0, w1, .L54
** b taken
** .L54:
** b not_taken
@@ -353,8 +308,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_uge_x1:
-** cmp w0, w1
-** bcc .L56
+** cblo w0, w1, .L56
** b taken
** .L56:
** b not_taken
@@ -362,8 +316,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_slt_x1:
-** cmp w0, w1
-** bge .L58
+** cbge w0, w1, .L58
** b taken
** .L58:
** b not_taken
@@ -371,8 +324,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sle_x1:
-** cmp w0, w1
-** bgt .L60
+** cbgt w0, w1, .L60
** b taken
** .L60:
** b not_taken
@@ -380,8 +332,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sgt_x1:
-** cmp w0, w1
-** ble .L62
+** cble w0, w1, .L62
** b taken
** .L62:
** b not_taken
@@ -389,8 +340,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sge_x1:
-** cmp w0, w1
-** blt .L64
+** cblt w0, w1, .L64
** b taken
** .L64:
** b not_taken
@@ -398,8 +348,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_eq_x1:
-** cmp x0, x1
-** beq .L67
+** cbeq x0, x1, .L67
** b not_taken
** .L67:
** b taken
@@ -407,8 +356,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ne_x1:
-** cmp x0, x1
-** beq .L69
+** cbeq x0, x1, .L69
** b taken
** .L69:
** b not_taken
@@ -416,8 +364,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ult_x1:
-** cmp x0, x1
-** bcs .L71
+** cbhs x0, x1, .L71
** b taken
** .L71:
** b not_taken
@@ -425,8 +372,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ule_x1:
-** cmp x0, x1
-** bhi .L73
+** cbhi x0, x1, .L73
** b taken
** .L73:
** b not_taken
@@ -434,8 +380,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ugt_x1:
-** cmp x0, x1
-** bls .L75
+** cbls x0, x1, .L75
** b taken
** .L75:
** b not_taken
@@ -443,8 +388,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_uge_x1:
-** cmp x0, x1
-** bcc .L77
+** cblo x0, x1, .L77
** b taken
** .L77:
** b not_taken
@@ -452,8 +396,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_slt_x1:
-** cmp x0, x1
-** bge .L79
+** cbge x0, x1, .L79
** b taken
** .L79:
** b not_taken
@@ -461,8 +404,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sle_x1:
-** cmp x0, x1
-** bgt .L81
+** cbgt x0, x1, .L81
** b taken
** .L81:
** b not_taken
@@ -470,8 +412,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sgt_x1:
-** cmp x0, x1
-** ble .L83
+** cble x0, x1, .L83
** b taken
** .L83:
** b not_taken
@@ -479,8 +420,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sge_x1:
-** cmp x0, x1
-** blt .L85
+** cblt x0, x1, .L85
** b taken
** .L85:
** b not_taken
@@ -488,8 +428,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_eq_42:
-** cmp w0, 42
-** beq .L88
+** cbeq w0, 42, .L88
** b not_taken
** .L88:
** b taken
@@ -497,8 +436,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ne_42:
-** cmp w0, 42
-** beq .L90
+** cbeq w0, 42, .L90
** b taken
** .L90:
** b not_taken
@@ -506,8 +444,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ult_42:
-** cmp w0, 41
-** bhi .L92
+** cbhi w0, 41, .L92
** b taken
** .L92:
** b not_taken
@@ -515,8 +452,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ule_42:
-** cmp w0, 42
-** bhi .L94
+** cbhi w0, 42, .L94
** b taken
** .L94:
** b not_taken
@@ -524,8 +460,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ugt_42:
-** cmp w0, 42
-** bls .L96
+** cbls w0, 42, .L96
** b taken
** .L96:
** b not_taken
@@ -533,8 +468,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_uge_42:
-** cmp w0, 41
-** bls .L98
+** cbls w0, 41, .L98
** b taken
** .L98:
** b not_taken
@@ -542,8 +476,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_slt_42:
-** cmp w0, 41
-** bgt .L100
+** cbgt w0, 41, .L100
** b taken
** .L100:
** b not_taken
@@ -551,8 +484,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sle_42:
-** cmp w0, 42
-** bgt .L102
+** cbgt w0, 42, .L102
** b taken
** .L102:
** b not_taken
@@ -560,8 +492,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sgt_42:
-** cmp w0, 42
-** ble .L104
+** cble w0, 42, .L104
** b taken
** .L104:
** b not_taken
@@ -569,8 +500,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sge_42:
-** cmp w0, 41
-** ble .L106
+** cble w0, 41, .L106
** b taken
** .L106:
** b not_taken
@@ -578,8 +508,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_eq_42:
-** cmp x0, 42
-** beq .L109
+** cbeq x0, 42, .L109
** b not_taken
** .L109:
** b taken
@@ -587,8 +516,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ne_42:
-** cmp x0, 42
-** beq .L111
+** cbeq x0, 42, .L111
** b taken
** .L111:
** b not_taken
@@ -596,8 +524,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ult_42:
-** cmp x0, 41
-** bhi .L113
+** cbhi x0, 41, .L113
** b taken
** .L113:
** b not_taken
@@ -605,8 +532,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ule_42:
-** cmp x0, 42
-** bhi .L115
+** cbhi x0, 42, .L115
** b taken
** .L115:
** b not_taken
@@ -614,8 +540,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ugt_42:
-** cmp x0, 42
-** bls .L117
+** cbls x0, 42, .L117
** b taken
** .L117:
** b not_taken
@@ -623,8 +548,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_uge_42:
-** cmp x0, 41
-** bls .L119
+** cbls x0, 41, .L119
** b taken
** .L119:
** b not_taken
@@ -632,8 +556,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_slt_42:
-** cmp x0, 41
-** bgt .L121
+** cbgt x0, 41, .L121
** b taken
** .L121:
** b not_taken
@@ -641,8 +564,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sle_42:
-** cmp x0, 42
-** bgt .L123
+** cbgt x0, 42, .L123
** b taken
** .L123:
** b not_taken
@@ -650,8 +572,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sgt_42:
-** cmp x0, 42
-** ble .L125
+** cble x0, 42, .L125
** b taken
** .L125:
** b not_taken
@@ -659,8 +580,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sge_42:
-** cmp x0, 41
-** ble .L127
+** cble x0, 41, .L127
** b taken
** .L127:
** b not_taken
@@ -668,8 +588,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_eq_0:
-** tst w0, 255
-** bne .L129
+** cbbne w0, wzr, .L129
** b taken
** .L129:
** b not_taken
@@ -677,8 +596,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ne_0:
-** tst w0, 255
-** beq .L131
+** cbbeq w0, wzr, .L131
** b taken
** .L131:
** b not_taken
@@ -691,8 +609,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ule_0:
-** tst w0, 255
-** bne .L134
+** cbbne w0, wzr, .L134
** b taken
** .L134:
** b not_taken
@@ -700,8 +617,7 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ugt_0:
-** tst w0, 255
-** beq .L136
+** cbbeq w0, wzr, .L136
** b taken
** .L136:
** b not_taken
@@ -714,7 +630,7 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_slt_0:
-** tbnz w0, 7, .L140
+** cbblt w0, wzr, .L140
** b not_taken
** .L140:
** b taken
@@ -722,9 +638,7 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sle_0:
-** sxtb w0, w0
-** cmp w0, 0
-** ble .L143
+** cbble w0, wzr, .L143
** b not_taken
** .L143:
** b taken
@@ -732,9 +646,7 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sgt_0:
-** sxtb w0, w0
-** cmp w0, 0
-** ble .L145
+** cbble w0, wzr, .L145
** b taken
** .L145:
** b not_taken
@@ -742,7 +654,7 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sge_0:
-** tbnz w0, 7, .L147
+** cbblt w0, wzr, .L147
** b taken
** .L147:
** b not_taken
@@ -750,8 +662,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_eq_0:
-** tst w0, 65535
-** bne .L149
+** cbhne w0, wzr, .L149
** b taken
** .L149:
** b not_taken
@@ -759,8 +670,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ne_0:
-** tst w0, 65535
-** beq .L151
+** cbheq w0, wzr, .L151
** b taken
** .L151:
** b not_taken
@@ -773,8 +683,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ule_0:
-** tst w0, 65535
-** bne .L154
+** cbhne w0, wzr, .L154
** b taken
** .L154:
** b not_taken
@@ -782,8 +691,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ugt_0:
-** tst w0, 65535
-** beq .L156
+** cbheq w0, wzr, .L156
** b taken
** .L156:
** b not_taken
@@ -796,7 +704,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_slt_0:
-** tbnz w0, 15, .L160
+** cbhlt w0, wzr, .L160
** b not_taken
** .L160:
** b taken
@@ -804,9 +712,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sle_0:
-** sxth w0, w0
-** cmp w0, 0
-** ble .L163
+** cbhle w0, wzr, .L163
** b not_taken
** .L163:
** b taken
@@ -814,9 +720,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sgt_0:
-** sxth w0, w0
-** cmp w0, 0
-** ble .L165
+** cbhle w0, wzr, .L165
** b taken
** .L165:
** b not_taken
@@ -824,7 +728,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sge_0:
-** tbnz w0, 15, .L167
+** cbhlt w0, wzr, .L167
** b taken
** .L167:
** b not_taken
@@ -832,7 +736,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_eq_0:
-** cbnz w0, .L169
+** cbne w0, wzr, .L169
** b taken
** .L169:
** b not_taken
@@ -840,7 +744,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ne_0:
-** cbz w0, .L171
+** cbeq w0, wzr, .L171
** b taken
** .L171:
** b not_taken
@@ -853,7 +757,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ule_0:
-** cbnz w0, .L174
+** cbne w0, wzr, .L174
** b taken
** .L174:
** b not_taken
@@ -861,7 +765,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ugt_0:
-** cbz w0, .L176
+** cbeq w0, wzr, .L176
** b taken
** .L176:
** b not_taken
@@ -882,8 +786,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sle_0:
-** cmp w0, 0
-** ble .L183
+** cble w0, wzr, .L183
** b not_taken
** .L183:
** b taken
@@ -891,8 +794,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sgt_0:
-** cmp w0, 0
-** ble .L185
+** cble w0, wzr, .L185
** b taken
** .L185:
** b not_taken
@@ -900,7 +802,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sge_0:
-** tbnz w0, #31, .L187
+** cblt w0, wzr, .L187
** b taken
** .L187:
** b not_taken
@@ -908,7 +810,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_eq_0:
-** cbnz x0, .L189
+** cbne x0, xzr, .L189
** b taken
** .L189:
** b not_taken
@@ -916,7 +818,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ne_0:
-** cbz x0, .L191
+** cbeq x0, xzr, .L191
** b taken
** .L191:
** b not_taken
@@ -929,7 +831,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ule_0:
-** cbnz x0, .L194
+** cbne x0, xzr, .L194
** b taken
** .L194:
** b not_taken
@@ -937,7 +839,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ugt_0:
-** cbz x0, .L196
+** cbeq x0, xzr, .L196
** b taken
** .L196:
** b not_taken
@@ -958,8 +860,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sle_0:
-** cmp x0, 0
-** ble .L203
+** cble x0, xzr, .L203
** b not_taken
** .L203:
** b taken
@@ -967,8 +868,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sgt_0:
-** cmp x0, 0
-** ble .L205
+** cble x0, xzr, .L205
** b taken
** .L205:
** b not_taken
@@ -976,7 +876,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sge_0:
-** tbnz x0, #63, .L207
+** cblt x0, xzr, .L207
** b taken
** .L207:
** b not_taken
@@ -984,9 +884,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_eq_42:
-** and w0, w0, 255
-** cmp w0, 42
-** beq .L210
+** mov w1, 42
+** cbbeq w0, w1, .L210
** b not_taken
** .L210:
** b taken
@@ -994,9 +893,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ne_42:
-** and w0, w0, 255
-** cmp w0, 42
-** beq .L212
+** mov w1, 42
+** cbbeq w0, w1, .L212
** b taken
** .L212:
** b not_taken
@@ -1004,9 +902,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ult_42:
-** and w0, w0, 255
-** cmp w0, 41
-** bhi .L214
+** mov w1, 41
+** cbbhi w0, w1, .L214
** b taken
** .L214:
** b not_taken
@@ -1014,9 +911,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ule_42:
-** and w0, w0, 255
-** cmp w0, 42
-** bhi .L216
+** mov w1, 42
+** cbbhi w0, w1, .L216
** b taken
** .L216:
** b not_taken
@@ -1024,9 +920,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ugt_42:
-** and w0, w0, 255
-** cmp w0, 42
-** bls .L218
+** mov w1, 42
+** cbbls w0, w1, .L218
** b taken
** .L218:
** b not_taken
@@ -1034,9 +929,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_uge_42:
-** and w0, w0, 255
-** cmp w0, 41
-** bls .L220
+** mov w1, 41
+** cbbls w0, w1, .L220
** b taken
** .L220:
** b not_taken
@@ -1044,9 +938,8 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_slt_42:
-** sxtb w0, w0
-** cmp w0, 41
-** bgt .L222
+** mov w1, 41
+** cbbgt w0, w1, .L222
** b taken
** .L222:
** b not_taken
@@ -1054,9 +947,8 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sle_42:
-** sxtb w0, w0
-** cmp w0, 42
-** bgt .L224
+** mov w1, 42
+** cbbgt w0, w1, .L224
** b taken
** .L224:
** b not_taken
@@ -1064,9 +956,8 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sgt_42:
-** sxtb w0, w0
-** cmp w0, 42
-** ble .L226
+** mov w1, 42
+** cbble w0, w1, .L226
** b taken
** .L226:
** b not_taken
@@ -1074,9 +965,8 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sge_42:
-** sxtb w0, w0
-** cmp w0, 41
-** ble .L228
+** mov w1, 41
+** cbble w0, w1, .L228
** b taken
** .L228:
** b not_taken
@@ -1084,9 +974,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_eq_42:
-** and w0, w0, 65535
-** cmp w0, 42
-** beq .L231
+** mov w1, 42
+** cbheq w0, w1, .L231
** b not_taken
** .L231:
** b taken
@@ -1094,9 +983,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ne_42:
-** and w0, w0, 65535
-** cmp w0, 42
-** beq .L233
+** mov w1, 42
+** cbheq w0, w1, .L233
** b taken
** .L233:
** b not_taken
@@ -1104,9 +992,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ult_42:
-** and w0, w0, 65535
-** cmp w0, 41
-** bhi .L235
+** mov w1, 41
+** cbhhi w0, w1, .L235
** b taken
** .L235:
** b not_taken
@@ -1114,9 +1001,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ule_42:
-** and w0, w0, 65535
-** cmp w0, 42
-** bhi .L237
+** mov w1, 42
+** cbhhi w0, w1, .L237
** b taken
** .L237:
** b not_taken
@@ -1124,9 +1010,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ugt_42:
-** and w0, w0, 65535
-** cmp w0, 42
-** bls .L239
+** mov w1, 42
+** cbhls w0, w1, .L239
** b taken
** .L239:
** b not_taken
@@ -1134,9 +1019,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_uge_42:
-** and w0, w0, 65535
-** cmp w0, 41
-** bls .L241
+** mov w1, 41
+** cbhls w0, w1, .L241
** b taken
** .L241:
** b not_taken
@@ -1144,9 +1028,8 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_slt_42:
-** sxth w0, w0
-** cmp w0, 41
-** bgt .L243
+** mov w1, 41
+** cbhgt w0, w1, .L243
** b taken
** .L243:
** b not_taken
@@ -1154,9 +1037,8 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sle_42:
-** sxth w0, w0
-** cmp w0, 42
-** bgt .L245
+** mov w1, 42
+** cbhgt w0, w1, .L245
** b taken
** .L245:
** b not_taken
@@ -1164,9 +1046,8 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sgt_42:
-** sxth w0, w0
-** cmp w0, 42
-** ble .L247
+** mov w1, 42
+** cbhle w0, w1, .L247
** b taken
** .L247:
** b not_taken
@@ -1174,9 +1055,8 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sge_42:
-** sxth w0, w0
-** cmp w0, 41
-** ble .L249
+** mov w1, 41
+** cbhle w0, w1, .L249
** b taken
** .L249:
** b not_taken
@@ -1184,9 +1064,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_eq_64:
-** and w0, w0, 255
-** cmp w0, 64
-** beq .L252
+** mov w1, 64
+** cbbeq w0, w1, .L252
** b not_taken
** .L252:
** b taken
@@ -1194,9 +1073,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ne_64:
-** and w0, w0, 255
-** cmp w0, 64
-** beq .L254
+** mov w1, 64
+** cbbeq w0, w1, .L254
** b taken
** .L254:
** b not_taken
@@ -1204,9 +1082,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ult_64:
-** and w0, w0, 255
-** cmp w0, 63
-** bhi .L256
+** mov w1, 63
+** cbbhi w0, w1, .L256
** b taken
** .L256:
** b not_taken
@@ -1214,9 +1091,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ule_64:
-** and w0, w0, 255
-** cmp w0, 64
-** bhi .L258
+** mov w1, 64
+** cbbhi w0, w1, .L258
** b taken
** .L258:
** b not_taken
@@ -1224,9 +1100,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_ugt_64:
-** and w0, w0, 255
-** cmp w0, 64
-** bls .L260
+** mov w1, 64
+** cbbls w0, w1, .L260
** b taken
** .L260:
** b not_taken
@@ -1234,9 +1109,8 @@ int far_branch(i32 x, i32 y) {
/*
** u8_x0_uge_64:
-** and w0, w0, 255
-** cmp w0, 63
-** bls .L262
+** mov w1, 63
+** cbbls w0, w1, .L262
** b taken
** .L262:
** b not_taken
@@ -1244,9 +1118,8 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_slt_64:
-** sxtb w0, w0
-** cmp w0, 63
-** bgt .L264
+** mov w1, 63
+** cbbgt w0, w1, .L264
** b taken
** .L264:
** b not_taken
@@ -1254,9 +1127,8 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sle_64:
-** sxtb w0, w0
-** cmp w0, 64
-** bgt .L266
+** mov w1, 64
+** cbbgt w0, w1, .L266
** b taken
** .L266:
** b not_taken
@@ -1264,9 +1136,8 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sgt_64:
-** sxtb w0, w0
-** cmp w0, 64
-** ble .L268
+** mov w1, 64
+** cbble w0, w1, .L268
** b taken
** .L268:
** b not_taken
@@ -1274,9 +1145,8 @@ int far_branch(i32 x, i32 y) {
/*
** i8_x0_sge_64:
-** sxtb w0, w0
-** cmp w0, 63
-** ble .L270
+** mov w1, 63
+** cbble w0, w1, .L270
** b taken
** .L270:
** b not_taken
@@ -1284,9 +1154,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_eq_64:
-** and w0, w0, 65535
-** cmp w0, 64
-** beq .L273
+** mov w1, 64
+** cbheq w0, w1, .L273
** b not_taken
** .L273:
** b taken
@@ -1294,9 +1163,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ne_64:
-** and w0, w0, 65535
-** cmp w0, 64
-** beq .L275
+** mov w1, 64
+** cbheq w0, w1, .L275
** b taken
** .L275:
** b not_taken
@@ -1304,9 +1172,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ult_64:
-** and w0, w0, 65535
-** cmp w0, 63
-** bhi .L277
+** mov w1, 63
+** cbhhi w0, w1, .L277
** b taken
** .L277:
** b not_taken
@@ -1314,9 +1181,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ule_64:
-** and w0, w0, 65535
-** cmp w0, 64
-** bhi .L279
+** mov w1, 64
+** cbhhi w0, w1, .L279
** b taken
** .L279:
** b not_taken
@@ -1324,9 +1190,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ugt_64:
-** and w0, w0, 65535
-** cmp w0, 64
-** bls .L281
+** mov w1, 64
+** cbhls w0, w1, .L281
** b taken
** .L281:
** b not_taken
@@ -1334,9 +1199,8 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_uge_64:
-** and w0, w0, 65535
-** cmp w0, 63
-** bls .L283
+** mov w1, 63
+** cbhls w0, w1, .L283
** b taken
** .L283:
** b not_taken
@@ -1344,9 +1208,8 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_slt_64:
-** sxth w0, w0
-** cmp w0, 63
-** bgt .L285
+** mov w1, 63
+** cbhgt w0, w1, .L285
** b taken
** .L285:
** b not_taken
@@ -1354,9 +1217,8 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sle_64:
-** sxth w0, w0
-** cmp w0, 64
-** bgt .L287
+** mov w1, 64
+** cbhgt w0, w1, .L287
** b taken
** .L287:
** b not_taken
@@ -1364,9 +1226,8 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sgt_64:
-** sxth w0, w0
-** cmp w0, 64
-** ble .L289
+** mov w1, 64
+** cbhle w0, w1, .L289
** b taken
** .L289:
** b not_taken
@@ -1374,9 +1235,8 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sge_64:
-** sxth w0, w0
-** cmp w0, 63
-** ble .L291
+** mov w1, 63
+** cbhle w0, w1, .L291
** b taken
** .L291:
** b not_taken
@@ -1402,8 +1262,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ult_64:
-** cmp w0, 63
-** bhi .L298
+** cbhi w0, 63, .L298
** b taken
** .L298:
** b not_taken
@@ -1438,8 +1297,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_slt_64:
-** cmp w0, 63
-** bgt .L306
+** cbgt w0, 63, .L306
** b taken
** .L306:
** b not_taken
@@ -1492,8 +1350,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ult_64:
-** cmp x0, 63
-** bhi .L319
+** cbhi x0, 63, .L319
** b taken
** .L319:
** b not_taken
@@ -1528,8 +1385,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_slt_64:
-** cmp x0, 63
-** bgt .L327
+** cbgt x0, 63, .L327
** b taken
** .L327:
** b not_taken
@@ -1565,8 +1421,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_eq_4098:
** mov w1, 4098
-** cmp w1, w0, uxth
-** beq .L336
+** cbheq w0, w1, .L336
** b not_taken
** .L336:
** b taken
@@ -1575,8 +1430,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ne_4098:
** mov w1, 4098
-** cmp w1, w0, uxth
-** beq .L338
+** cbheq w0, w1, .L338
** b taken
** .L338:
** b not_taken
@@ -1585,8 +1439,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ult_4098:
** mov w1, 4097
-** cmp w1, w0, uxth
-** bcc .L340
+** cbhhi w0, w1, .L340
** b taken
** .L340:
** b not_taken
@@ -1595,8 +1448,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ule_4098:
** mov w1, 4098
-** cmp w1, w0, uxth
-** bcc .L342
+** cbhhi w0, w1, .L342
** b taken
** .L342:
** b not_taken
@@ -1605,8 +1457,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_ugt_4098:
** mov w1, 4098
-** cmp w1, w0, uxth
-** bcs .L344
+** cbhls w0, w1, .L344
** b taken
** .L344:
** b not_taken
@@ -1615,8 +1466,7 @@ int far_branch(i32 x, i32 y) {
/*
** u16_x0_uge_4098:
** mov w1, 4097
-** cmp w1, w0, uxth
-** bcs .L346
+** cbhls w0, w1, .L346
** b taken
** .L346:
** b not_taken
@@ -1625,8 +1475,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_slt_4098:
** mov w1, 4097
-** cmp w1, w0, sxth
-** blt .L348
+** cbhgt w0, w1, .L348
** b taken
** .L348:
** b not_taken
@@ -1635,8 +1484,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sle_4098:
** mov w1, 4098
-** cmp w1, w0, sxth
-** blt .L350
+** cbhgt w0, w1, .L350
** b taken
** .L350:
** b not_taken
@@ -1645,8 +1493,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sgt_4098:
** mov w1, 4098
-** cmp w1, w0, sxth
-** bge .L352
+** cbhle w0, w1, .L352
** b taken
** .L352:
** b not_taken
@@ -1655,8 +1502,7 @@ int far_branch(i32 x, i32 y) {
/*
** i16_x0_sge_4098:
** mov w1, 4097
-** cmp w1, w0, sxth
-** bge .L354
+** cbhle w0, w1, .L354
** b taken
** .L354:
** b not_taken
@@ -1665,8 +1511,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_eq_4098:
** mov w1, 4098
-** cmp w0, w1
-** beq .L357
+** cbeq w0, w1, .L357
** b not_taken
** .L357:
** b taken
@@ -1675,8 +1520,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ne_4098:
** mov w1, 4098
-** cmp w0, w1
-** beq .L359
+** cbeq w0, w1, .L359
** b taken
** .L359:
** b not_taken
@@ -1685,8 +1529,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ult_4098:
** mov w1, 4097
-** cmp w0, w1
-** bhi .L361
+** cbhi w0, w1, .L361
** b taken
** .L361:
** b not_taken
@@ -1695,8 +1538,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ule_4098:
** mov w1, 4098
-** cmp w0, w1
-** bhi .L363
+** cbhi w0, w1, .L363
** b taken
** .L363:
** b not_taken
@@ -1705,8 +1547,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_ugt_4098:
** mov w1, 4098
-** cmp w0, w1
-** bls .L365
+** cbls w0, w1, .L365
** b taken
** .L365:
** b not_taken
@@ -1715,8 +1556,7 @@ int far_branch(i32 x, i32 y) {
/*
** u32_x0_uge_4098:
** mov w1, 4097
-** cmp w0, w1
-** bls .L367
+** cbls w0, w1, .L367
** b taken
** .L367:
** b not_taken
@@ -1725,8 +1565,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_slt_4098:
** mov w1, 4097
-** cmp w0, w1
-** bgt .L369
+** cbgt w0, w1, .L369
** b taken
** .L369:
** b not_taken
@@ -1735,8 +1574,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sle_4098:
** mov w1, 4098
-** cmp w0, w1
-** bgt .L371
+** cbgt w0, w1, .L371
** b taken
** .L371:
** b not_taken
@@ -1745,8 +1583,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sgt_4098:
** mov w1, 4098
-** cmp w0, w1
-** ble .L373
+** cble w0, w1, .L373
** b taken
** .L373:
** b not_taken
@@ -1755,8 +1592,7 @@ int far_branch(i32 x, i32 y) {
/*
** i32_x0_sge_4098:
** mov w1, 4097
-** cmp w0, w1
-** ble .L375
+** cble w0, w1, .L375
** b taken
** .L375:
** b not_taken
@@ -1765,8 +1601,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_eq_4098:
** mov x1, 4098
-** cmp x0, x1
-** beq .L378
+** cbeq x0, x1, .L378
** b not_taken
** .L378:
** b taken
@@ -1775,8 +1610,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ne_4098:
** mov x1, 4098
-** cmp x0, x1
-** beq .L380
+** cbeq x0, x1, .L380
** b taken
** .L380:
** b not_taken
@@ -1785,8 +1619,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ult_4098:
** mov x1, 4097
-** cmp x0, x1
-** bhi .L382
+** cbhi x0, x1, .L382
** b taken
** .L382:
** b not_taken
@@ -1795,8 +1628,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ule_4098:
** mov x1, 4098
-** cmp x0, x1
-** bhi .L384
+** cbhi x0, x1, .L384
** b taken
** .L384:
** b not_taken
@@ -1805,8 +1637,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_ugt_4098:
** mov x1, 4098
-** cmp x0, x1
-** bls .L386
+** cbls x0, x1, .L386
** b taken
** .L386:
** b not_taken
@@ -1815,8 +1646,7 @@ int far_branch(i32 x, i32 y) {
/*
** u64_x0_uge_4098:
** mov x1, 4097
-** cmp x0, x1
-** bls .L388
+** cbls x0, x1, .L388
** b taken
** .L388:
** b not_taken
@@ -1825,8 +1655,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_slt_4098:
** mov x1, 4097
-** cmp x0, x1
-** bgt .L390
+** cbgt x0, x1, .L390
** b taken
** .L390:
** b not_taken
@@ -1835,8 +1664,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sle_4098:
** mov x1, 4098
-** cmp x0, x1
-** bgt .L392
+** cbgt x0, x1, .L392
** b taken
** .L392:
** b not_taken
@@ -1845,8 +1673,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sgt_4098:
** mov x1, 4098
-** cmp x0, x1
-** ble .L394
+** cble x0, x1, .L394
** b taken
** .L394:
** b not_taken
@@ -1855,8 +1682,7 @@ int far_branch(i32 x, i32 y) {
/*
** i64_x0_sge_4098:
** mov x1, 4097
-** cmp x0, x1
-** ble .L396
+** cble x0, x1, .L396
** b taken
** .L396:
** b not_taken
@@ -1866,8 +1692,9 @@ int far_branch(i32 x, i32 y) {
** far_branch:
** sub sp, sp, #16
** str wzr, \[sp, 12\]
-** cmp w0, w1
-** bne .L398
+** cbeq w0, w1, .L400
+** b .L398
+** .L400:
** str wzr, \[sp, 12\]
** ...
** str wzr, \[sp, 12\]
--
2.45.2
