cranelift: Implement missing i128 rotates on AArch64 (#4866)

cranelift/codegen/src/isa/aarch64/lower.isle

@@ -1282,7 +1282,8 @@
 
 ;; General 8/16-bit case.
 (rule (lower (has_type (fits_in_16 ty) (rotl x y)))
-      (let ((neg_shift Reg (sub $I32 (zero_reg) y)))
+      (let ((amt Reg (value_regs_get y 0))
+            (neg_shift Reg (sub $I32 (zero_reg) amt)))
         (small_rotr ty (put_in_reg_zext32 x) neg_shift)))
 
 ;; Specialization for the 8/16-bit case when the rotation amount is an immediate.
@@ -1300,12 +1301,14 @@
 
 ;; General 32-bit case.
 (rule (lower (has_type $I32 (rotl x y)))
-      (let ((neg_shift Reg (sub $I32 (zero_reg) y)))
+      (let ((amt Reg (value_regs_get y 0))
+            (neg_shift Reg (sub $I32 (zero_reg) amt)))
         (a64_rotr $I32 x neg_shift)))
 
 ;; General 64-bit case.
 (rule (lower (has_type $I64 (rotl x y)))
-      (let ((neg_shift Reg (sub $I64 (zero_reg) y)))
+      (let ((amt Reg (value_regs_get y 0))
+            (neg_shift Reg (sub $I64 (zero_reg) amt)))
         (a64_rotr $I64 x neg_shift)))
 
 ;; Specialization for the 32-bit case when the rotation amount is an immediate.
@@ -1338,15 +1341,15 @@
 
 ;; General 8/16-bit case.
 (rule (lower (has_type (fits_in_16 ty) (rotr x y)))
-      (small_rotr ty (put_in_reg_zext32 x) y))
+      (small_rotr ty (put_in_reg_zext32 x) (value_regs_get y 0)))
 
 ;; General 32-bit case.
 (rule (lower (has_type $I32 (rotr x y)))
-      (a64_rotr $I32 x y))
+      (a64_rotr $I32 x (value_regs_get y 0)))
 
 ;; General 64-bit case.
 (rule (lower (has_type $I64 (rotr x y)))
-      (a64_rotr $I64 x y))
+      (a64_rotr $I64 x (value_regs_get y 0)))
 
 ;; Specialization for the 8/16-bit case when the rotation amount is an immediate.
 (rule (lower (has_type (fits_in_16 ty) (rotr x (iconst k))))

cranelift/filetests/filetests/runtests/i128-rotate.clif

@@ -51,3 +51,117 @@ block0(v0: i128, v1: i8):
 ; run: %rotr_amt_i128(0x01010101_01010101_01010101_01010101, 73) == 0x80808080_80808080_80808080_80808080
 ; run: %rotr_amt_i128(0x02020202_02020202_01010101_01010101, 0) == 0x02020202_02020202_01010101_01010101
 ; run: %rotr_amt_i128(0x03030303_03030303_01010101_01010101, 128) == 0x03030303_03030303_01010101_01010101
+
+
+function %rotl_i64_i128(i64, i128) -> i64 {
+block0(v0: i64, v1: i128):
+    v2 = rotl.i64 v0, v1
+    return v2
+}
+; run: %rotl_i64_i128(0xe0000000_00000000, 0x00000000_00000000_00000000_00000000) == 0xe0000000_00000000
+; run: %rotl_i64_i128(0xe0000000_00000000, 0x00000000_00000002_00000000_00000001) == 0xc0000000_00000001
+; run: %rotl_i64_i128(0xe000000f_0000000f, 0x00000000_00000002_00000000_00000000) == 0xe000000f_0000000f
+; run: %rotl_i64_i128(0xe000000f_0000000f, 0x00000000_00000002_00000000_00000004) == 0x000000f0_000000fe
+; run: %rotl_i64_i128(0xe0000000_00000004, 0x00000000_00000002_00000000_00000040) == 0xe0000000_00000004
+; run: %rotl_i64_i128(0xe0000000_00000004, 0x00000000_00000002_00000000_00000041) == 0xc0000000_00000009
+; run: %rotl_i64_i128(0xe0000000_00000004, 0x00000000_00000002_00000000_00000042) == 0x80000000_00000013
+; run: %rotl_i64_i128(0xe0000000_00000004, 0x00000000_00000002_00000000_00000101) == 0xc0000000_00000009
+
+function %rotl_i32_i128(i32, i128) -> i32 {
+block0(v0: i32, v1: i128):
+    v2 = rotl.i32 v0, v1
+    return v2
+}
+; run: %rotl_i32_i128(0xe0000000, 0x00000000_00000000_00000000_00000000) == 0xe0000000
+; run: %rotl_i32_i128(0xe0000000, 0x00000000_00000002_00000000_00000001) == 0xc0000001
+; run: %rotl_i32_i128(0xe00f000f, 0x00000000_00000002_00000000_00000000) == 0xe00f000f
+; run: %rotl_i32_i128(0xe00f000f, 0x00000000_00000002_00000000_00000004) == 0x00f000fe
+; run: %rotl_i32_i128(0xe0000004, 0x00000000_00000002_00000000_00000020) == 0xe0000004
+; run: %rotl_i32_i128(0xe0000004, 0x00000000_00000002_00000000_00000021) == 0xc0000009
+; run: %rotl_i32_i128(0xe0000004, 0x00000000_00000002_00000000_00000022) == 0x80000013
+; run: %rotl_i32_i128(0xe0000004, 0x00000000_00000002_00000000_00000101) == 0xc0000009
+
+function %rotl_i16_i128(i16, i128) -> i16 {
+block0(v0: i16, v1: i128):
+    v2 = rotl.i16 v0, v1
+    return v2
+}
+; run: %rotl_i16_i128(0xe000, 0x00000000_00000000_00000000_00000000) == 0xe000
+; run: %rotl_i16_i128(0xe000, 0x00000000_00000002_00000000_00000001) == 0xc001
+; run: %rotl_i16_i128(0xef0f, 0x00000000_00000002_00000000_00000000) == 0xef0f
+; run: %rotl_i16_i128(0xef0f, 0x00000000_00000002_00000000_00000004) == 0xf0fe
+; run: %rotl_i16_i128(0xe004, 0x00000000_00000002_00000000_00000010) == 0xe004
+; run: %rotl_i16_i128(0xe004, 0x00000000_00000002_00000000_00000011) == 0xc009
+; run: %rotl_i16_i128(0xe004, 0x00000000_00000002_00000000_00000012) == 0x8013
+; run: %rotl_i16_i128(0xe004, 0x00000000_00000002_00000000_00000101) == 0xc009
+
+function %rotl_i8_i128(i8, i128) -> i8 {
+block0(v0: i8, v1: i128):
+    v2 = rotl.i8 v0, v1
+    return v2
+}
+; run: %rotl_i8_i128(0xe0, 0x00000000_00000000_00000000_00000000) == 0xe0
+; run: %rotl_i8_i128(0xe0, 0x00000000_00000002_00000000_00000001) == 0xc1
+; run: %rotl_i8_i128(0xef, 0x00000000_00000002_00000000_00000000) == 0xef
+; run: %rotl_i8_i128(0xef, 0x00000000_00000002_00000000_00000004) == 0xfe
+; run: %rotl_i8_i128(0xe4, 0x00000000_00000002_00000000_00000008) == 0xe4
+; run: %rotl_i8_i128(0xe4, 0x00000000_00000002_00000000_00000009) == 0xc9
+; run: %rotl_i8_i128(0xe4, 0x00000000_00000002_00000000_0000000A) == 0x93
+; run: %rotl_i8_i128(0xe4, 0x00000000_00000002_00000000_00000101) == 0xc9
+
+
+function %rotr_i64_i128(i64, i128) -> i64 {
+block0(v0: i64, v1: i128):
+    v2 = rotr.i64 v0, v1
+    return v2
+}
+; run: %rotr_i64_i128(0xe0000000_00000000, 0x00000000_00000000_00000000_00000000) == 0xe0000000_00000000
+; run: %rotr_i64_i128(0xe0000000_00000000, 0x00000000_00000002_00000000_00000001) == 0x70000000_00000000
+; run: %rotr_i64_i128(0xe000000f_0000000f, 0x00000000_00000002_00000000_00000000) == 0xe000000f_0000000f
+; run: %rotr_i64_i128(0xe000000f_0000000f, 0x00000000_00000002_00000000_00000004) == 0xfe000000_f0000000
+; run: %rotr_i64_i128(0xe0000000_00000004, 0x00000000_00000002_00000000_00000040) == 0xe0000000_00000004
+; run: %rotr_i64_i128(0xe0000000_00000004, 0x00000000_00000002_00000000_00000041) == 0x70000000_00000002
+; run: %rotr_i64_i128(0xe0000000_00000004, 0x00000000_00000002_00000000_00000042) == 0x38000000_00000001
+; run: %rotr_i64_i128(0xe0000000_00000004, 0x00000000_00000002_00000000_00000101) == 0x70000000_00000002
+
+function %rotr_i32_i128(i32, i128) -> i32 {
+block0(v0: i32, v1: i128):
+    v2 = rotr.i32 v0, v1
+    return v2
+}
+; run: %rotr_i32_i128(0xe0000000, 0x00000000_00000000_00000000_00000000) == 0xe0000000
+; run: %rotr_i32_i128(0xe0000000, 0x00000000_00000002_00000000_00000001) == 0x70000000
+; run: %rotr_i32_i128(0xe00f000f, 0x00000000_00000002_00000000_00000000) == 0xe00f000f
+; run: %rotr_i32_i128(0xe00f000f, 0x00000000_00000002_00000000_00000004) == 0xfe00f000
+; run: %rotr_i32_i128(0xe0000004, 0x00000000_00000002_00000000_00000020) == 0xe0000004
+; run: %rotr_i32_i128(0xe0000004, 0x00000000_00000002_00000000_00000021) == 0x70000002
+; run: %rotr_i32_i128(0xe0000004, 0x00000000_00000002_00000000_00000022) == 0x38000001
+; run: %rotr_i32_i128(0xe0000004, 0x00000000_00000002_00000000_00000101) == 0x70000002
+
+function %rotr_i16_i128(i16, i128) -> i16 {
+block0(v0: i16, v1: i128):
+    v2 = rotr.i16 v0, v1
+    return v2
+}
+; run: %rotr_i16_i128(0xe000, 0x00000000_00000000_00000000_00000000) == 0xe000
+; run: %rotr_i16_i128(0xe000, 0x00000000_00000002_00000000_00000001) == 0x7000
+; run: %rotr_i16_i128(0xef0f, 0x00000000_00000002_00000000_00000000) == 0xef0f
+; run: %rotr_i16_i128(0xef0f, 0x00000000_00000002_00000000_00000004) == 0xfef0
+; run: %rotr_i16_i128(0xe004, 0x00000000_00000002_00000000_00000010) == 0xe004
+; run: %rotr_i16_i128(0xe004, 0x00000000_00000002_00000000_00000011) == 0x7002
+; run: %rotr_i16_i128(0xe004, 0x00000000_00000002_00000000_00000012) == 0x3801
+; run: %rotr_i16_i128(0xe004, 0x00000000_00000002_00000000_00000101) == 0x7002
+
+function %rotr_i8_i128(i8, i128) -> i8 {
+block0(v0: i8, v1: i128):
+    v2 = rotr.i8 v0, v1
+    return v2
+}
+; run: %rotr_i8_i128(0xe0, 0x00000000_00000000_00000000_00000000) == 0xe0
+; run: %rotr_i8_i128(0xe0, 0x00000000_00000002_00000000_00000001) == 0x70
+; run: %rotr_i8_i128(0xef, 0x00000000_00000002_00000000_00000000) == 0xef
+; run: %rotr_i8_i128(0xef, 0x00000000_00000002_00000000_00000004) == 0xfe
+; run: %rotr_i8_i128(0xe0, 0x00000000_00000002_00000000_00000008) == 0xe0
+; run: %rotr_i8_i128(0xe0, 0x00000000_00000002_00000000_00000009) == 0x70
+; run: %rotr_i8_i128(0xe0, 0x00000000_00000002_00000000_0000000A) == 0x38
+; run: %rotr_i8_i128(0xe0, 0x00000000_00000002_00000000_00000101) == 0x70