Merge pull request #3610 from cfallin/fix-narrow-type-rotate

Fix some 16- and 8-bit behavior in x64 backend related to rotates.

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

@@ -1,4 +1,4 @@
 src/clif.isle be1359b4b6b153f378517c1dd95cd80f4a6bed0c7b86eaba11c088fd71b7bfe80a3c868ace245b2da0bfbbd6ded262ea9576c8e0eeacbf89d03c34a17a709602
-src/prelude.isle d3d2a6a42fb778231a4cdca30995324e1293a9ca8073c5a27a501535759eb51f84a6718322a93dfba4b66ee4f0c9afce7dcec0428516ef0c5bc96e8c8b76925d
+src/prelude.isle 75a46b97817ad6a4c34e618b81e60876eec6fd1c83ac3ee174851e42045c951644663b2cbc31f1749ce2bc3ad9eb94fb0b877eb2c3bc4885cab7d7e87e9df1d6
 src/isa/aarch64/inst.isle 8e4b8e452cf06a368c2e1d930042027a5d3bd690ab46d498d959257e9b4461d17abf244838395cd80da1fe5e2e86fc43855fb5753ca4f1643538c2ae4b3b4a1e
 src/isa/aarch64/lower.isle bc3db9c1e6ac186b918cc04f4d26af398f99ec36c8cdc20ec4d02d18dd57dba12e3184fea031b4ac97051c5e194a69666afb5e204807c818e6688c177f9c1b91

cranelift/codegen/src/isa/x64/inst.isle

@@ -77,9 +77,13 @@
             Size32
             Size64))
 
-;; Get the `OperandSize` for a given `Type`.
-(decl operand_size_of_type (Type) OperandSize)
-(extern constructor operand_size_of_type operand_size_of_type)
+;; Get the `OperandSize` for a given `Type`, rounding smaller types up to 32 bits.
+(decl operand_size_of_type_32_64 (Type) OperandSize)
+(extern constructor operand_size_of_type_32_64 operand_size_of_type_32_64)
+
+;; Get the true `OperandSize` for a given `Type`, with no rounding.
+(decl raw_operand_size_of_type (Type) OperandSize)
+(extern constructor raw_operand_size_of_type raw_operand_size_of_type)
 
 ;; Get the bit width of an `OperandSize`.
 (decl operand_size_bits (OperandSize) u16)
@@ -379,6 +383,10 @@
 (decl imm8_from_value (Imm8Reg) Value)
 (extern extractor imm8_from_value imm8_from_value)
 
+;; Mask an `Imm8Reg.Imm8`.
+(decl mask_imm8_const (Imm8Reg u64) Imm8Reg)
+(extern constructor mask_imm8_const mask_imm8_const)
+
 ;; Extract a constant `RegMemImm.Imm` from a value operand.
 (decl simm32_from_value (RegMemImm) Value)
 (extern extractor simm32_from_value simm32_from_value)
@@ -426,7 +434,7 @@
       (let ((from_bits u16 (ty_bits_u16 from_ty))
             ;; Use `operand_size_of_type` so that the we clamp the output to 32-
             ;; or 64-bit width types.
-            (to_bits u16 (operand_size_bits (operand_size_of_type to_ty))))
+            (to_bits u16 (operand_size_bits (operand_size_of_type_32_64 to_ty))))
         (extend kind
                 to_ty
                 (ext_mode from_bits to_bits)
@@ -503,7 +511,7 @@
 (decl alu_rmi_r (Type AluRmiROpcode Reg RegMemImm) Reg)
 (rule (alu_rmi_r ty opcode src1 src2)
       (let ((dst WritableReg (temp_writable_reg ty))
-            (size OperandSize (operand_size_of_type ty))
+            (size OperandSize (operand_size_of_type_32_64 ty))
             (_ Unit (emit (MInst.AluRmiR size opcode src1 src2 dst))))
         (writable_reg_to_reg dst)))
 
@@ -519,7 +527,7 @@
 (decl add_with_flags (Type Reg RegMemImm) ProducesFlags)
 (rule (add_with_flags ty src1 src2)
       (let ((dst WritableReg (temp_writable_reg ty)))
-        (ProducesFlags.ProducesFlags (MInst.AluRmiR (operand_size_of_type ty)
+        (ProducesFlags.ProducesFlags (MInst.AluRmiR (operand_size_of_type_32_64 ty)
                                                     (AluRmiROpcode.Add)
                                                     src1
                                                     src2
@@ -530,7 +538,7 @@
 (decl adc (Type Reg RegMemImm) ConsumesFlags)
 (rule (adc ty src1 src2)
       (let ((dst WritableReg (temp_writable_reg ty)))
-        (ConsumesFlags.ConsumesFlags (MInst.AluRmiR (operand_size_of_type ty)
+        (ConsumesFlags.ConsumesFlags (MInst.AluRmiR (operand_size_of_type_32_64 ty)
                                                     (AluRmiROpcode.Adc)
                                                     src1
                                                     src2
@@ -549,7 +557,7 @@
 (decl sub_with_flags (Type Reg RegMemImm) ProducesFlags)
 (rule (sub_with_flags ty src1 src2)
       (let ((dst WritableReg (temp_writable_reg ty)))
-        (ProducesFlags.ProducesFlags (MInst.AluRmiR (operand_size_of_type ty)
+        (ProducesFlags.ProducesFlags (MInst.AluRmiR (operand_size_of_type_32_64 ty)
                                                     (AluRmiROpcode.Sub)
                                                     src1
                                                     src2
@@ -560,7 +568,7 @@
 (decl sbb (Type Reg RegMemImm) ConsumesFlags)
 (rule (sbb ty src1 src2)
       (let ((dst WritableReg (temp_writable_reg ty)))
-        (ConsumesFlags.ConsumesFlags (MInst.AluRmiR (operand_size_of_type ty)
+        (ConsumesFlags.ConsumesFlags (MInst.AluRmiR (operand_size_of_type_32_64 ty)
                                                     (AluRmiROpcode.Sbb)
                                                     src1
                                                     src2
@@ -608,7 +616,7 @@
 ;; Integer immediates.
 (rule (imm ty simm64)
       (let ((dst WritableReg (temp_writable_reg ty))
-            (size OperandSize (operand_size_of_type ty))
+            (size OperandSize (operand_size_of_type_32_64 ty))
             (_ Unit (emit (MInst.Imm size simm64 dst))))
         (writable_reg_to_reg dst)))
 
@@ -634,7 +642,7 @@
 (rule (imm ty 0)
       (let ((wr WritableReg (temp_writable_reg ty))
             (r Reg (writable_reg_to_reg wr))
-            (size OperandSize (operand_size_of_type ty))
+            (size OperandSize (operand_size_of_type_32_64 ty))
             (_ Unit (emit (MInst.AluRmiR size
                                          (AluRmiROpcode.Xor)
                                          r
@@ -681,7 +689,9 @@
 (decl shift_r (Type ShiftKind Reg Imm8Reg) Reg)
 (rule (shift_r ty kind src1 src2)
       (let ((dst WritableReg (temp_writable_reg ty))
-            (size OperandSize (operand_size_of_type ty))
+            ;; Use actual 8/16-bit instructions when appropriate: we
+            ;; rely on their shift-amount-masking semantics.
+            (size OperandSize (raw_operand_size_of_type ty))
             (_ Unit (emit (MInst.ShiftR size kind src1 src2 dst))))
         (writable_reg_to_reg dst)))
 
@@ -729,7 +739,7 @@
 (decl cmove (Type CC RegMem Reg) ConsumesFlags)
 (rule (cmove ty cc consequent alternative)
       (let ((dst WritableReg (temp_writable_reg ty))
-            (size OperandSize (operand_size_of_type ty)))
+            (size OperandSize (operand_size_of_type_32_64 ty)))
         (ConsumesFlags.ConsumesFlags (MInst.Cmove size cc consequent alternative dst)
                                      (writable_reg_to_reg dst))))
 
@@ -1155,7 +1165,7 @@
 (rule (mul_hi ty signed src1 src2)
       (let ((dst_lo WritableReg (temp_writable_reg ty))
             (dst_hi WritableReg (temp_writable_reg ty))
-            (size OperandSize (operand_size_of_type ty))
+            (size OperandSize (operand_size_of_type_32_64 ty))
             (_ Unit (emit (MInst.MulHi size
                                        signed
                                        src1
@@ -1224,6 +1234,6 @@
 (decl not (Type Reg) Reg)
 (rule (not ty src)
       (let ((dst WritableReg (temp_writable_reg ty))
-            (size OperandSize (operand_size_of_type ty))
+            (size OperandSize (operand_size_of_type_32_64 ty))
             (_ Unit (emit (MInst.Not size src dst))))
         (writable_reg_to_reg dst)))

cranelift/codegen/src/isa/x64/lower.isle

@@ -633,15 +633,26 @@
 
 ;;;; Rules for `rotl` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
-;; `i64` and smaller.
+;; `i16` and `i8`: we need to extend the shift amount, or mask the
+;; constant.
+
+(rule (lower (has_type (ty_8_or_16 ty) (rotl src amt)))
+      (let ((amt_ Reg (extend_to_reg amt $I32 (ExtendKind.Zero))))
+        (value_reg (m_rotl ty (put_in_reg src) (Imm8Reg.Reg amt_)))))
+
+(rule (lower (has_type (ty_8_or_16 ty) (rotl src (imm8_from_value amt))))
+      (value_reg (m_rotl ty (put_in_reg src) (mask_imm8_const amt (ty_bits_mask ty)))))
+
+;; `i64` and `i32`: we can rely on x86's rotate-amount masking since
+;;  we operate on the whole register.
 
-(rule (lower (has_type (fits_in_64 ty) (rotl src amt)))
+(rule (lower (has_type (ty_32_or_64 ty) (rotl src amt)))
       ;; NB: Only the low bits of `amt` matter since we logically mask the
       ;; shift amount to the value's bit width.
       (let ((amt_ Reg (lo_reg amt)))
         (value_reg (m_rotl ty (put_in_reg src) (Imm8Reg.Reg amt_)))))
 
-(rule (lower (has_type (fits_in_64 ty) (rotl src (imm8_from_value amt))))
+(rule (lower (has_type (ty_32_or_64 ty) (rotl src (imm8_from_value amt))))
       (value_reg (m_rotl ty (put_in_reg src) amt)))
 
 ;; `i128`.

cranelift/codegen/src/isa/x64/lower/isle.rs

@@ -56,14 +56,19 @@ where
     isle_prelude_methods!();
 
     #[inline]
-    fn operand_size_of_type(&mut self, ty: Type) -> OperandSize {
+    fn operand_size_of_type_32_64(&mut self, ty: Type) -> OperandSize {
         if ty.bits() == 64 {
             OperandSize::Size64
         } else {
             OperandSize::Size32
         }
     }
 
+    #[inline]
+    fn raw_operand_size_of_type(&mut self, ty: Type) -> OperandSize {
+        OperandSize::from_ty(ty)
+    }
+
     fn put_in_reg_mem(&mut self, val: Value) -> RegMem {
         let inputs = self.lower_ctx.get_value_as_source_or_const(val);
 
@@ -125,6 +130,16 @@ where
         Some(Imm8Reg::Imm8 { imm })
     }
 
+    #[inline]
+    fn mask_imm8_const(&mut self, imm8: &Imm8Reg, mask: u64) -> Imm8Reg {
+        match imm8 {
+            &Imm8Reg::Reg { reg } => Imm8Reg::Reg { reg },
+            &Imm8Reg::Imm8 { imm } => Imm8Reg::Imm8 {
+                imm: imm & (mask as u8),
+            },
+        }
+    }
+
     #[inline]
     fn simm32_from_value(&mut self, val: Value) -> Option<RegMemImm> {
         let inst = self.lower_ctx.dfg().value_def(val).inst()?;

cranelift/codegen/src/isa/x64/lower/isle/generated_code.manifest

@@ -1,4 +1,4 @@
 src/clif.isle be1359b4b6b153f378517c1dd95cd80f4a6bed0c7b86eaba11c088fd71b7bfe80a3c868ace245b2da0bfbbd6ded262ea9576c8e0eeacbf89d03c34a17a709602
-src/prelude.isle d3d2a6a42fb778231a4cdca30995324e1293a9ca8073c5a27a501535759eb51f84a6718322a93dfba4b66ee4f0c9afce7dcec0428516ef0c5bc96e8c8b76925d
-src/isa/x64/inst.isle b151120df3c356ac697122a8557becd8857eb725851506e844edeb85d831d461322a96d280ad84f9a23518e1e4efb607aebc0e249004148675e4cc19e89f0655
-src/isa/x64/lower.isle c9b408df0a089fb4f207838973ac775b0f9b56c86f056867c28e6bae317873d3844f74f713f9acd6fed98d3d11a2f9d19d392fe5049169dad33b1fc703b9b766
+src/prelude.isle 75a46b97817ad6a4c34e618b81e60876eec6fd1c83ac3ee174851e42045c951644663b2cbc31f1749ce2bc3ad9eb94fb0b877eb2c3bc4885cab7d7e87e9df1d6
+src/isa/x64/inst.isle 1a44ccc0c2cad90447762848461fcae714216ef058d42bdba89330a6008061526e92bbf1c17055c465b20fc75d98d1faa34feda8b22fa7ae0504a0f808798b41
+src/isa/x64/lower.isle c7943201b32e9eb9726466e8cc417f7e84c4c4052de31e05ab6e0ad7502a587cf1d7d9835703c4ff5a506390f7a0668741e7f3feaa1edda6396571a425949fc9