From 817c2b205c4101d495fecb678a8db4ba28561d2c Mon Sep 17 00:00:00 2001
From: Trevor Elliott <telliott@fastly.com>
Date: Mon, 5 Dec 2022 12:54:14 -0800
Subject: [PATCH] riscv64: Use a temporary when translating shift amount
 (#5375)

Use a temporary when translating the shift amount, instead of reusing the destination register.
---
 .../codegen/src/isa/riscv64/lower/isle.rs     |   8 +-
 .../filetests/isa/riscv64/bitops.clif         | 115 +++++------
 .../filetests/isa/riscv64/shift-rotate.clif   | 186 +++++++++---------
 3 files changed, 154 insertions(+), 155 deletions(-)

diff --git a/cranelift/codegen/src/isa/riscv64/lower/isle.rs b/cranelift/codegen/src/isa/riscv64/lower/isle.rs
index 0d37942099..723bd01edc 100644
--- a/cranelift/codegen/src/isa/riscv64/lower/isle.rs
+++ b/cranelift/codegen/src/isa/riscv64/lower/isle.rs
@@ -293,15 +293,13 @@ impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6>
             tmp.to_reg()
         };
         let len_sub_shamt = {
+            let tmp = self.temp_writable_reg(I64);
+            self.emit(&MInst::load_imm12(tmp, Imm12::from_bits(ty.bits() as i16)));
             let len_sub_shamt = self.temp_writable_reg(I64);
-            self.emit(&MInst::load_imm12(
-                len_sub_shamt,
-                Imm12::from_bits(ty.bits() as i16),
-            ));
             self.emit(&MInst::AluRRR {
                 alu_op: AluOPRRR::Sub,
                 rd: len_sub_shamt,
-                rs1: len_sub_shamt.to_reg(),
+                rs1: tmp.to_reg(),
                 rs2: shamt,
             });
             len_sub_shamt.to_reg()
diff --git a/cranelift/filetests/filetests/isa/riscv64/bitops.clif b/cranelift/filetests/filetests/isa/riscv64/bitops.clif
index 8115cf5522..c4d0575666 100644
--- a/cranelift/filetests/filetests/isa/riscv64/bitops.clif
+++ b/cranelift/filetests/filetests/isa/riscv64/bitops.clif
@@ -795,18 +795,19 @@ block0(v0: i128, v1: i8):
 }
 
 ; block0:
-;   mv a4,a1
+;   mv t2,a1
 ;   andi a1,a2,127
 ;   li a3,128
-;   sub a3,a3,a1
-;   sll a6,a0,a1
-;   srl t3,a0,a3
-;   select_reg t0,zero,t3##condition=(a1 eq zero)
-;   sll t2,a4,a1
-;   or a2,t0,t2
-;   li a3,64
-;   select_reg a0,zero,a6##condition=(a1 uge a3)
-;   select_reg a1,a6,a2##condition=(a1 uge a3)
+;   sub a5,a3,a1
+;   sll a7,a0,a1
+;   srl t4,a0,a5
+;   select_reg t1,zero,t4##condition=(a1 eq zero)
+;   mv a5,t2
+;   sll a0,a5,a1
+;   or a2,t1,a0
+;   li a4,64
+;   select_reg a0,zero,a7##condition=(a1 uge a4)
+;   select_reg a1,a7,a2##condition=(a1 uge a4)
 ;   ret
 
 function %ishl_i128_i128(i128, i128) -> i128 {
@@ -818,15 +819,15 @@ block0(v0: i128, v1: i128):
 ; block0:
 ;   andi a2,a2,127
 ;   li a4,128
-;   sub a4,a4,a2
-;   sll a7,a0,a2
-;   srl t4,a0,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   sll a0,a1,a2
-;   or a3,t1,a0
-;   li a4,64
-;   select_reg a0,zero,a7##condition=(a2 uge a4)
-;   select_reg a1,a7,a3##condition=(a2 uge a4)
+;   sub a6,a4,a2
+;   sll t3,a0,a2
+;   srl t0,a0,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   sll a1,a1,a2
+;   or a3,t2,a1
+;   li a5,64
+;   select_reg a0,zero,t3##condition=(a2 uge a5)
+;   select_reg a1,t3,a3##condition=(a2 uge a5)
 ;   ret
 
 function %ushr_i128_i8(i128, i8) -> i128 {
@@ -836,18 +837,18 @@ block0(v0: i128, v1: i8):
 }
 
 ; block0:
-;   mv t1,a1
+;   mv t2,a1
 ;   andi a1,a2,127
 ;   li a3,128
-;   sub a3,a3,a1
-;   mv a2,t1
-;   sll a6,a2,a3
-;   select_reg t3,zero,a6##condition=(a1 eq zero)
-;   srl t0,a0,a1
-;   or t2,t3,t0
+;   sub a5,a3,a1
+;   mv a2,t2
+;   sll a7,a2,a5
+;   select_reg t4,zero,a7##condition=(a1 eq zero)
+;   srl t1,a0,a1
+;   or a0,t4,t1
 ;   li a3,64
 ;   srl a4,a2,a1
-;   select_reg a0,a4,t2##condition=(a1 uge a3)
+;   select_reg a0,a4,a0##condition=(a1 uge a3)
 ;   select_reg a1,zero,a4##condition=(a1 uge a3)
 ;   ret
 
@@ -860,15 +861,15 @@ block0(v0: i128, v1: i128):
 ; block0:
 ;   andi a2,a2,127
 ;   li a4,128
-;   sub a4,a4,a2
-;   sll a7,a1,a4
-;   select_reg t4,zero,a7##condition=(a2 eq zero)
-;   srl t1,a0,a2
-;   or a0,t4,t1
+;   sub a6,a4,a2
+;   sll t3,a1,a6
+;   select_reg t0,zero,t3##condition=(a2 eq zero)
+;   srl t2,a0,a2
+;   or a4,t0,t2
 ;   li a3,64
-;   srl a4,a1,a2
-;   select_reg a0,a4,a0##condition=(a2 uge a3)
-;   select_reg a1,zero,a4##condition=(a2 uge a3)
+;   srl a5,a1,a2
+;   select_reg a0,a5,a4##condition=(a2 uge a3)
+;   select_reg a1,zero,a5##condition=(a2 uge a3)
 ;   ret
 
 function %sshr_i128_i8(i128, i8) -> i128 {
@@ -878,20 +879,20 @@ block0(v0: i128, v1: i8):
 }
 
 ; block0:
-;   mv a3,a1
+;   mv a4,a1
 ;   andi a1,a2,127
-;   li a4,128
-;   sub a4,a4,a1
-;   sll a6,a3,a4
-;   select_reg t3,zero,a6##condition=(a1 eq zero)
-;   srl t0,a0,a1
-;   or t2,t3,t0
+;   li a3,128
+;   sub a5,a3,a1
+;   sll a7,a4,a5
+;   select_reg t4,zero,a7##condition=(a1 eq zero)
+;   srl t1,a0,a1
+;   or a0,t4,t1
 ;   li a2,64
-;   sra a4,a3,a1
-;   li a5,-1
-;   select_reg a7,a5,zero##condition=(a3 slt zero)
-;   select_reg a0,a4,t2##condition=(a1 uge a2)
-;   select_reg a1,a7,a4##condition=(a1 uge a2)
+;   sra a5,a4,a1
+;   li a6,-1
+;   select_reg t3,a6,zero##condition=(a4 slt zero)
+;   select_reg a0,a5,a0##condition=(a1 uge a2)
+;   select_reg a1,t3,a5##condition=(a1 uge a2)
 ;   ret
 
 function %sshr_i128_i128(i128, i128) -> i128 {
@@ -903,16 +904,16 @@ block0(v0: i128, v1: i128):
 ; block0:
 ;   andi a2,a2,127
 ;   li a4,128
-;   sub a4,a4,a2
-;   sll a7,a1,a4
-;   select_reg t4,zero,a7##condition=(a2 eq zero)
-;   srl t1,a0,a2
-;   or a0,t4,t1
+;   sub a6,a4,a2
+;   sll t3,a1,a6
+;   select_reg t0,zero,t3##condition=(a2 eq zero)
+;   srl t2,a0,a2
+;   or a4,t0,t2
 ;   li a3,64
-;   sra a4,a1,a2
-;   li a6,-1
-;   select_reg t3,a6,zero##condition=(a1 slt zero)
-;   select_reg a0,a4,a0##condition=(a2 uge a3)
-;   select_reg a1,t3,a4##condition=(a2 uge a3)
+;   sra a5,a1,a2
+;   li a7,-1
+;   select_reg t4,a7,zero##condition=(a1 slt zero)
+;   select_reg a0,a5,a4##condition=(a2 uge a3)
+;   select_reg a1,t4,a5##condition=(a2 uge a3)
 ;   ret
 
diff --git a/cranelift/filetests/filetests/isa/riscv64/shift-rotate.clif b/cranelift/filetests/filetests/isa/riscv64/shift-rotate.clif
index 9ae1ea4b1f..f6b224beb3 100644
--- a/cranelift/filetests/filetests/isa/riscv64/shift-rotate.clif
+++ b/cranelift/filetests/filetests/isa/riscv64/shift-rotate.clif
@@ -15,18 +15,18 @@ block0(v0: i128, v1: i128):
 ; block0:
 ;   andi a2,a2,127
 ;   li a4,128
-;   sub a4,a4,a2
-;   srl a7,a0,a2
-;   sll t4,a1,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   or a3,a7,t1
-;   srl a5,a1,a2
-;   sll a4,a0,a4
-;   select_reg a6,zero,a4##condition=(a2 eq zero)
-;   or t3,a5,a6
-;   li t0,64
-;   select_reg a0,t3,a3##condition=(a2 uge t0)
-;   select_reg a1,a3,t3##condition=(a2 uge t0)
+;   sub a6,a4,a2
+;   srl t3,a0,a2
+;   sll t0,a1,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   or a3,t3,t2
+;   srl a4,a1,a2
+;   sll a5,a0,a6
+;   select_reg a7,zero,a5##condition=(a2 eq zero)
+;   or t4,a4,a7
+;   li t1,64
+;   select_reg a0,t4,a3##condition=(a2 uge t1)
+;   select_reg a1,a3,t4##condition=(a2 uge t1)
 ;   ret
 
 function %f0(i64, i64) -> i64 {
@@ -36,15 +36,15 @@ block0(v0: i64, v1: i64):
 }
 
 ; block0:
-;   mv a6,a0
+;   mv a7,a0
 ;   andi a0,a1,63
 ;   li a2,64
-;   sub a2,a2,a0
-;   mv t4,a6
-;   srl a5,t4,a0
-;   sll a7,t4,a2
-;   select_reg t4,zero,a7##condition=(a0 eq zero)
-;   or a0,a5,t4
+;   sub a4,a2,a0
+;   mv t0,a7
+;   srl a6,t0,a0
+;   sll t3,t0,a4
+;   select_reg t0,zero,t3##condition=(a0 eq zero)
+;   or a0,a6,t0
 ;   ret
 
 function %f1(i32, i32) -> i32 {
@@ -57,11 +57,11 @@ block0(v0: i32, v1: i32):
 ;   uext.w a0,a0
 ;   andi a2,a1,31
 ;   li a4,32
-;   sub a4,a4,a2
-;   srl a7,a0,a2
-;   sll t4,a0,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   or a0,a7,t1
+;   sub a6,a4,a2
+;   srl t3,a0,a2
+;   sll t0,a0,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   or a0,t3,t2
 ;   ret
 
 function %f2(i16, i16) -> i16 {
@@ -74,11 +74,11 @@ block0(v0: i16, v1: i16):
 ;   uext.h a0,a0
 ;   andi a2,a1,15
 ;   li a4,16
-;   sub a4,a4,a2
-;   srl a7,a0,a2
-;   sll t4,a0,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   or a0,a7,t1
+;   sub a6,a4,a2
+;   srl t3,a0,a2
+;   sll t0,a0,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   or a0,t3,t2
 ;   ret
 
 function %f3(i8, i8) -> i8 {
@@ -91,11 +91,11 @@ block0(v0: i8, v1: i8):
 ;   uext.b a0,a0
 ;   andi a2,a1,7
 ;   li a4,8
-;   sub a4,a4,a2
-;   srl a7,a0,a2
-;   sll t4,a0,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   or a0,a7,t1
+;   sub a6,a4,a2
+;   srl t3,a0,a2
+;   sll t0,a0,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   or a0,t3,t2
 ;   ret
 
 function %i128_rotl(i128, i128) -> i128 {
@@ -107,18 +107,18 @@ block0(v0: i128, v1: i128):
 ; block0:
 ;   andi a2,a2,127
 ;   li a4,128
-;   sub a4,a4,a2
-;   sll a7,a0,a2
-;   srl t4,a1,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   or a3,a7,t1
-;   sll a5,a1,a2
-;   srl a4,a0,a4
-;   select_reg a6,zero,a4##condition=(a2 eq zero)
-;   or t3,a5,a6
-;   li t0,64
-;   select_reg a0,t3,a3##condition=(a2 uge t0)
-;   select_reg a1,a3,t3##condition=(a2 uge t0)
+;   sub a6,a4,a2
+;   sll t3,a0,a2
+;   srl t0,a1,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   or a3,t3,t2
+;   sll a4,a1,a2
+;   srl a5,a0,a6
+;   select_reg a7,zero,a5##condition=(a2 eq zero)
+;   or t4,a4,a7
+;   li t1,64
+;   select_reg a0,t4,a3##condition=(a2 uge t1)
+;   select_reg a1,a3,t4##condition=(a2 uge t1)
 ;   ret
 
 function %f4(i64, i64) -> i64 {
@@ -128,15 +128,15 @@ block0(v0: i64, v1: i64):
 }
 
 ; block0:
-;   mv a6,a0
+;   mv a7,a0
 ;   andi a0,a1,63
 ;   li a2,64
-;   sub a2,a2,a0
-;   mv t4,a6
-;   sll a5,t4,a0
-;   srl a7,t4,a2
-;   select_reg t4,zero,a7##condition=(a0 eq zero)
-;   or a0,a5,t4
+;   sub a4,a2,a0
+;   mv t0,a7
+;   sll a6,t0,a0
+;   srl t3,t0,a4
+;   select_reg t0,zero,t3##condition=(a0 eq zero)
+;   or a0,a6,t0
 ;   ret
 
 function %f5(i32, i32) -> i32 {
@@ -149,11 +149,11 @@ block0(v0: i32, v1: i32):
 ;   uext.w a0,a0
 ;   andi a2,a1,31
 ;   li a4,32
-;   sub a4,a4,a2
-;   sll a7,a0,a2
-;   srl t4,a0,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   or a0,a7,t1
+;   sub a6,a4,a2
+;   sll t3,a0,a2
+;   srl t0,a0,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   or a0,t3,t2
 ;   ret
 
 function %f6(i16, i16) -> i16 {
@@ -166,11 +166,11 @@ block0(v0: i16, v1: i16):
 ;   uext.h a0,a0
 ;   andi a2,a1,15
 ;   li a4,16
-;   sub a4,a4,a2
-;   sll a7,a0,a2
-;   srl t4,a0,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   or a0,a7,t1
+;   sub a6,a4,a2
+;   sll t3,a0,a2
+;   srl t0,a0,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   or a0,t3,t2
 ;   ret
 
 function %f7(i8, i8) -> i8 {
@@ -183,11 +183,11 @@ block0(v0: i8, v1: i8):
 ;   uext.b a0,a0
 ;   andi a2,a1,7
 ;   li a4,8
-;   sub a4,a4,a2
-;   sll a7,a0,a2
-;   srl t4,a0,a4
-;   select_reg t1,zero,t4##condition=(a2 eq zero)
-;   or a0,a7,t1
+;   sub a6,a4,a2
+;   sll t3,a0,a2
+;   srl t0,a0,a6
+;   select_reg t2,zero,t0##condition=(a2 eq zero)
+;   or a0,t3,t2
 ;   ret
 
 function %f8(i64, i64) -> i64 {
@@ -331,11 +331,11 @@ block0(v0: i64):
 ;   li t2,17
 ;   andi a1,t2,63
 ;   li a3,64
-;   sub a3,a3,a1
-;   srl a6,a0,a1
-;   sll t3,a0,a3
-;   select_reg t0,zero,t3##condition=(a1 eq zero)
-;   or a0,a6,t0
+;   sub a5,a3,a1
+;   srl a7,a0,a1
+;   sll t4,a0,a5
+;   select_reg t1,zero,t4##condition=(a1 eq zero)
+;   or a0,a7,t1
 ;   ret
 
 function %f21(i64) -> i64 {
@@ -349,11 +349,11 @@ block0(v0: i64):
 ;   li t2,17
 ;   andi a1,t2,63
 ;   li a3,64
-;   sub a3,a3,a1
-;   sll a6,a0,a1
-;   srl t3,a0,a3
-;   select_reg t0,zero,t3##condition=(a1 eq zero)
-;   or a0,a6,t0
+;   sub a5,a3,a1
+;   sll a7,a0,a1
+;   srl t4,a0,a5
+;   select_reg t1,zero,t4##condition=(a1 eq zero)
+;   or a0,a7,t1
 ;   ret
 
 function %f22(i32) -> i32 {
@@ -368,11 +368,11 @@ block0(v0: i32):
 ;   uext.w a1,a0
 ;   andi a3,t2,31
 ;   li a5,32
-;   sub a5,a5,a3
-;   sll t3,a1,a3
-;   srl t0,a1,a5
-;   select_reg t2,zero,t0##condition=(a3 eq zero)
-;   or a0,t3,t2
+;   sub a7,a5,a3
+;   sll t4,a1,a3
+;   srl t1,a1,a7
+;   select_reg a0,zero,t1##condition=(a3 eq zero)
+;   or a0,t4,a0
 ;   ret
 
 function %f23(i16) -> i16 {
@@ -387,11 +387,11 @@ block0(v0: i16):
 ;   uext.h a1,a0
 ;   andi a3,t2,15
 ;   li a5,16
-;   sub a5,a5,a3
-;   sll t3,a1,a3
-;   srl t0,a1,a5
-;   select_reg t2,zero,t0##condition=(a3 eq zero)
-;   or a0,t3,t2
+;   sub a7,a5,a3
+;   sll t4,a1,a3
+;   srl t1,a1,a7
+;   select_reg a0,zero,t1##condition=(a3 eq zero)
+;   or a0,t4,a0
 ;   ret
 
 function %f24(i8) -> i8 {
@@ -406,11 +406,11 @@ block0(v0: i8):
 ;   uext.b a1,a0
 ;   andi a3,t2,7
 ;   li a5,8
-;   sub a5,a5,a3
-;   sll t3,a1,a3
-;   srl t0,a1,a5
-;   select_reg t2,zero,t0##condition=(a3 eq zero)
-;   or a0,t3,t2
+;   sub a7,a5,a3
+;   sll t4,a1,a3
+;   srl t1,a1,a7
+;   select_reg a0,zero,t1##condition=(a3 eq zero)
+;   or a0,t4,a0
 ;   ret
 
 function %f25(i64) -> i64 {