From e021995323f0b43f5649d3b7f90ea0f56b9de8fd Mon Sep 17 00:00:00 2001
From: Afonso Bordado <afonsobordado@az8.co>
Date: Sat, 8 May 2021 17:11:16 +0100
Subject: [PATCH] Allow i128 amount operands on shift instructions in the x64
 backend

Fixes #2727.
---
 cranelift/codegen/src/isa/x64/lower.rs        |   2 +-
 .../filetests/filetests/isa/x64/i128.clif     | 216 ++++++++++++++++++
 .../filetests/isa/x64/shift-i128-run.clif     |  93 ++++++++
 3 files changed, 310 insertions(+), 1 deletion(-)

diff --git a/cranelift/codegen/src/isa/x64/lower.rs b/cranelift/codegen/src/isa/x64/lower.rs
index 6f675b9232..a1969d5642 100644
--- a/cranelift/codegen/src/isa/x64/lower.rs
+++ b/cranelift/codegen/src/isa/x64/lower.rs
@@ -2046,7 +2046,7 @@ fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
                 }
                 ctx.emit(Inst::shift_r(size, shift_kind, count, dst));
             } else if dst_ty == types::I128 {
-                let amt_src = put_input_in_reg(ctx, inputs[1]);
+                let amt_src = put_input_in_regs(ctx, inputs[1]).regs()[0];
                 let src = put_input_in_regs(ctx, inputs[0]);
                 let dst = get_output_reg(ctx, outputs[0]);
 
diff --git a/cranelift/filetests/filetests/isa/x64/i128.clif b/cranelift/filetests/filetests/isa/x64/i128.clif
index 1b28abfca2..c480b857bb 100644
--- a/cranelift/filetests/filetests/isa/x64/i128.clif
+++ b/cranelift/filetests/filetests/isa/x64/i128.clif
@@ -890,3 +890,219 @@ block0(v0: i8, v1: i128):
 ; nextln: movq    %rbp, %rsp
 ; nextln: popq    %rbp
 ; nextln: ret
+
+function %f30(i128, i128) -> i128 {
+block0(v0: i128, v1: i128):
+    v2 = ishl v0, v1
+    return v2
+}
+
+; check:  pushq   %rbp
+; nextln: movq    %rsp, %rbp
+; nextln: movq    %rsi, %rax
+; nextln: movq    %rdi, %rsi
+; nextln: movq    %rdx, %rcx
+; nextln: shlq    %cl, %rsi
+; nextln: movq    %rdx, %rcx
+; nextln: shlq    %cl, %rax
+; nextln: movl    $$64, %ecx
+; nextln: subq    %rdx, %rcx
+; nextln: shrq    %cl, %rdi
+; nextln: xorq    %rcx, %rcx
+; nextln: testq   $$127, %rdx
+; nextln: cmovzq  %rcx, %rdi
+; nextln: orq     %rax, %rdi
+; nextln: xorq    %rax, %rax
+; nextln: andq    $$64, %rdx
+; nextln: cmovzq  %rdi, %rax
+; nextln: cmovzq  %rsi, %rcx
+; nextln: cmovnzq %rsi, %rax
+; nextln: movq    %rax, %rdx
+; nextln: movq    %rcx, %rax
+; nextln: movq    %rbp, %rsp
+; nextln: popq    %rbp
+; nextln: ret
+
+function %f31(i128, i128) -> i128 {
+block0(v0: i128, v1: i128):
+    v2 = ushr v0, v1
+    return v2
+}
+
+; check:  pushq   %rbp
+; nextln: movq    %rsp, %rbp
+; nextln: movq    %rdi, %rax
+; nextln: movq    %rsi, %rdi
+; nextln: movq    %rdi, %rsi
+; nextln: movq    %rdx, %rcx
+; nextln: shrq    %cl, %rsi
+; nextln: movq    %rdx, %rcx
+; nextln: shrq    %cl, %rax
+; nextln: movl    $$64, %ecx
+; nextln: subq    %rdx, %rcx
+; nextln: shlq    %cl, %rdi
+; nextln: xorq    %rcx, %rcx
+; nextln: testq   $$127, %rdx
+; nextln: cmovzq  %rcx, %rdi
+; nextln: orq     %rax, %rdi
+; nextln: xorq    %rax, %rax
+; nextln: xorq    %rcx, %rcx
+; nextln: andq    $$64, %rdx
+; nextln: cmovzq  %rsi, %rax
+; nextln: cmovzq  %rdi, %rcx
+; nextln: cmovnzq %rsi, %rcx
+; nextln: movq    %rax, %rdx
+; nextln: movq    %rcx, %rax
+; nextln: movq    %rbp, %rsp
+; nextln: popq    %rbp
+; nextln: ret
+
+function %f32(i128, i128) -> i128 {
+block0(v0: i128, v1: i128):
+    v2 = sshr v0, v1
+    return v2
+}
+
+; check:  pushq   %rbp
+; nextln: movq    %rsp, %rbp
+; nextln: movq    %rdi, %r8
+; nextln: movq    %rsi, %rdi
+; nextln: movq    %rdi, %rsi
+; nextln: movq    %rdx, %rcx
+; nextln: sarq    %cl, %rsi
+; nextln: movq    %rdx, %rcx
+; nextln: shrq    %cl, %r8
+; nextln: movl    $$64, %ecx
+; nextln: subq    %rdx, %rcx
+; nextln: movq    %rdi, %rax
+; nextln: shlq    %cl, %rax
+; nextln: xorq    %rcx, %rcx
+; nextln: testq   $$127, %rdx
+; nextln: cmovzq  %rcx, %rax
+; nextln: orq     %r8, %rax
+; nextln: sarq    $$63, %rdi
+; nextln: xorq    %rcx, %rcx
+; nextln: andq    $$64, %rdx
+; nextln: cmovzq  %rsi, %rdi
+; nextln: cmovzq  %rax, %rcx
+; nextln: cmovnzq %rsi, %rcx
+; nextln: movq    %rcx, %rax
+; nextln: movq    %rdi, %rdx
+; nextln: movq    %rbp, %rsp
+; nextln: popq    %rbp
+; nextln: ret
+
+function %f33(i128, i128) -> i128 {
+block0(v0: i128, v1: i128):
+    v2 = rotl v0, v1
+    return v2
+}
+
+; check: pushq   %rbp
+; nextln: movq    %rsp, %rbp
+; nextln: movq    %rdi, %r8
+; nextln: movq    %r8, %r9
+; nextln: movq    %rdx, %rcx
+; nextln: shlq    %cl, %r9
+; nextln: movq    %rsi, %rax
+; nextln: movq    %rdx, %rcx
+; nextln: shlq    %cl, %rax
+; nextln: movl    $$64, %ecx
+; nextln: subq    %rdx, %rcx
+; nextln: movq    %r8, %r10
+; nextln: shrq    %cl, %r10
+; nextln: xorq    %rdi, %rdi
+; nextln: testq   $$127, %rdx
+; nextln: cmovzq  %rdi, %r10
+; nextln: orq     %rax, %r10
+; nextln: xorq    %rax, %rax
+; nextln: movq    %rdx, %rcx
+; nextln: andq    $$64, %rcx
+; nextln: cmovzq  %r10, %rax
+; nextln: cmovzq  %r9, %rdi
+; nextln: cmovnzq %r9, %rax
+; nextln: movl    $$128, %r9d
+; nextln: subq    %rdx, %r9
+; nextln: movq    %rsi, %rdx
+; nextln: movq    %r9, %rcx
+; nextln: shrq    %cl, %rdx
+; nextln: movq    %r9, %rcx
+; nextln: shrq    %cl, %r8
+; nextln: movl    $$64, %ecx
+; nextln: subq    %r9, %rcx
+; nextln: shlq    %cl, %rsi
+; nextln: xorq    %rcx, %rcx
+; nextln: testq   $$127, %r9
+; nextln: cmovzq  %rcx, %rsi
+; nextln: orq     %r8, %rsi
+; nextln: xorq    %rcx, %rcx
+; nextln: xorq    %r8, %r8
+; nextln: andq    $$64, %r9
+; nextln: cmovzq  %rdx, %rcx
+; nextln: cmovzq  %rsi, %r8
+; nextln: cmovnzq %rdx, %r8
+; nextln: orq     %rdi, %r8
+; nextln: orq     %rax, %rcx
+; nextln: movq    %r8, %rax
+; nextln: movq    %rcx, %rdx
+; nextln: movq    %rbp, %rsp
+; nextln: popq    %rbp
+; nextln: ret
+
+
+function %f34(i128, i128) -> i128 {
+block0(v0: i128, v1: i128):
+    v2 = rotr v0, v1
+    return v2
+}
+
+; check:  pushq   %rbp
+; nextln: movq    %rsp, %rbp
+; nextln: movq    %rsi, %r9
+; nextln: movq    %rdx, %rcx
+; nextln: shrq    %cl, %r9
+; nextln: movq    %rdi, %rax
+; nextln: movq    %rdx, %rcx
+; nextln: shrq    %cl, %rax
+; nextln: movl    $$64, %ecx
+; nextln: subq    %rdx, %rcx
+; nextln: movq    %rsi, %r10
+; nextln: shlq    %cl, %r10
+; nextln: xorq    %rcx, %rcx
+; nextln: testq   $$127, %rdx
+; nextln: cmovzq  %rcx, %r10
+; nextln: orq     %rax, %r10
+; nextln: xorq    %rax, %rax
+; nextln: xorq    %r8, %r8
+; nextln: movq    %rdx, %rcx
+; nextln: andq    $$64, %rcx
+; nextln: cmovzq  %r9, %rax
+; nextln: cmovzq  %r10, %r8
+; nextln: cmovnzq %r9, %r8
+; nextln: movl    $$128, %r9d
+; nextln: subq    %rdx, %r9
+; nextln: movq    %rdi, %rdx
+; nextln: movq    %r9, %rcx
+; nextln: shlq    %cl, %rdx
+; nextln: movq    %rsi, %r10
+; nextln: movq    %r9, %rcx
+; nextln: shlq    %cl, %r10
+; nextln: movl    $$64, %ecx
+; nextln: subq    %r9, %rcx
+; nextln: shrq    %cl, %rdi
+; nextln: xorq    %rsi, %rsi
+; nextln: testq   $$127, %r9
+; nextln: cmovzq  %rsi, %rdi
+; nextln: orq     %r10, %rdi
+; nextln: xorq    %rcx, %rcx
+; nextln: andq    $$64, %r9
+; nextln: cmovzq  %rdi, %rcx
+; nextln: cmovzq  %rdx, %rsi
+; nextln: cmovnzq %rdx, %rcx
+; nextln: orq     %r8, %rsi
+; nextln: orq     %rax, %rcx
+; nextln: movq    %rsi, %rax
+; nextln: movq    %rcx, %rdx
+; nextln: movq    %rbp, %rsp
+; nextln: popq    %rbp
+; nextln: ret
\ No newline at end of file
diff --git a/cranelift/filetests/filetests/isa/x64/shift-i128-run.clif b/cranelift/filetests/filetests/isa/x64/shift-i128-run.clif
index 3a076ed536..029444eea0 100644
--- a/cranelift/filetests/filetests/isa/x64/shift-i128-run.clif
+++ b/cranelift/filetests/filetests/isa/x64/shift-i128-run.clif
@@ -71,3 +71,96 @@ block0(v0: i64, v1: i64, v2: i8):
 ; run: %rotr(0x01010101_01010101, 0x01010101_01010101, 73) == [0x80808080_80808080, 0x80808080_80808080]
 ; run: %rotr(0x01010101_01010101, 0x02020202_02020202, 0) == [0x01010101_01010101, 0x02020202_02020202]
 ; run: %rotr(0x01010101_01010101, 0x03030303_03030303, 128) == [0x01010101_01010101, 0x03030303_03030303]
+
+
+; i128 amount operand tests
+function %ishl_amt_i128(i64, i64, i8) -> i64, i64 {
+block0(v0: i64, v1: i64, v2: i8):
+    v3 = uextend.i64 v2
+    v4 = iconcat v3, v3
+
+    v5 = iconcat v0, v1
+
+    v6 = ishl.i128 v5, v4
+    v7, v8 = isplit v6
+    return v7, v8
+}
+; run: %ishl_amt_i128(0x01010101_01010101, 0x01010101_01010101, 2) == [0x04040404_04040404, 0x04040404_04040404]
+; run: %ishl_amt_i128(0x01010101_01010101, 0x01010101_01010101, 9) == [0x02020202_02020200, 0x02020202_02020202]
+; run: %ishl_amt_i128(0x01010101_01010101, 0xffffffff_ffffffff, 66) == [0x00000000_00000000, 0x04040404_04040404]
+; run: %ishl_amt_i128(0x01010101_01010101, 0x01010101_01010101, 0) == [0x01010101_01010101, 0x01010101_01010101]
+; run: %ishl_amt_i128(0x01010101_01010101, 0x01010101_01010101, 128) == [0x01010101_01010101, 0x01010101_01010101]
+; run: %ishl_amt_i128(0x00000000_00000001, 0x00000000_00000000, 0) == [0x00000000_00000001, 0x00000000_00000000]
+; run: %ishl_amt_i128(0x00000000_00000000, 0x00000000_00000001, 0) == [0x00000000_00000000, 0x00000000_00000001]
+; run: %ishl_amt_i128(0x12340000_00000000, 0x56780000_00000000, 0) == [0x12340000_00000000, 0x56780000_00000000]
+; run: %ishl_amt_i128(0x12340000_00000000, 0x56780000_00000000, 64) == [0x00000000_00000000, 0x12340000_00000000]
+; run: %ishl_amt_i128(0x12340000_00000000, 0x56780000_00000000, 32) == [0x00000000_00000000, 0x00000000_12340000]
+
+function %ushr_amt_i128(i64, i64, i8) -> i64, i64 {
+block0(v0: i64, v1: i64, v2: i8):
+    v3 = uextend.i64 v2
+    v4 = iconcat v3, v3
+
+    v5 = iconcat v0, v1
+
+    v6 = ushr.i128 v5, v4
+    v7, v8 = isplit v6
+    return v7, v8
+}
+; run: %ushr_amt_i128(0x01010101_01010101, 0x01010101_01010101, 2) == [0x40404040_40404040, 0x00404040_40404040]
+; run: %ushr_amt_i128(0x01010101_01010101, 0x01010101_01010101, 66) == [0x00404040_40404040, 0x00000000_00000000]
+; run: %ushr_amt_i128(0x01010101_01010101, 0x01010101_01010101, 0) == [0x01010101_01010101, 0x01010101_01010101]
+; run: %ushr_amt_i128(0x01010101_01010101, 0x01010101_01010101, 128) == [0x01010101_01010101, 0x01010101_01010101]
+; run: %ushr_amt_i128(0x00000000_00000001, 0x00000000_00000000, 0) == [0x00000000_00000001, 0x00000000_00000000]
+; run: %ushr_amt_i128(0x00000000_00000000, 0x00000000_00000001, 0) == [0x00000000_00000000, 0x00000000_00000001]
+; run: %ushr_amt_i128(0x12340000_00000000, 0x56780000_00000000, 0) == [0x12340000_00000000, 0x56780000_00000000]
+; run: %ushr_amt_i128(0x12340000_00000000, 0x56780000_00000000, 64) == [0x56780000_00000000, 0x00000000_00000000]
+; run: %ushr_amt_i128(0x12340000_00000000, 0x56780000_00000000, 32) == [0x00000000_12340000, 0x00000000_56780000]
+
+function %sshr_amt_i128(i64, i64, i8) -> i64, i64 {
+block0(v0: i64, v1: i64, v2: i8):
+    v3 = uextend.i64 v2
+    v4 = iconcat v3, v3
+
+    v5 = iconcat v0, v1
+
+    v6 = sshr.i128 v5, v4
+    v7, v8 = isplit v6
+    return v7, v8
+}
+; run: %sshr_amt_i128(0x01010101_01010101, 0x81010101_01010101, 2) == [0x40404040_40404040, 0xe0404040_40404040]
+; run: %sshr_amt_i128(0x12345678_9abcdef0, 0x80101010_10101010, 66) == [0xe0040404_04040404, 0xffffffff_ffffffff]
+; run: %sshr_amt_i128(0x12345678_9abcdef0, 0x80101010_10101010, 0) == [0x12345678_9abcdef0, 0x80101010_10101010]
+; run: %sshr_amt_i128(0x12345678_9abcdef0, 0x80101010_10101010, 128) == [0x12345678_9abcdef0, 0x80101010_10101010]
+
+function %rotl_amt_i128(i64, i64, i8) -> i64, i64 {
+block0(v0: i64, v1: i64, v2: i8):
+    v3 = uextend.i64 v2
+    v4 = iconcat v3, v3
+
+    v5 = iconcat v0, v1
+
+    v6 = rotl.i128 v5, v4
+    v7, v8 = isplit v6
+    return v7, v8
+}
+; run: %rotl_amt_i128(0x01010101_01010101, 0x01010101_01010101, 9) == [0x02020202_02020202, 0x02020202_02020202]
+; run: %rotl_amt_i128(0x01010101_01010101, 0x01010101_01010101, 73) == [0x02020202_02020202, 0x02020202_02020202]
+; run: %rotl_amt_i128(0x01010101_01010101, 0x02020202_02020202, 0) == [0x01010101_01010101, 0x02020202_02020202]
+; run: %rotl_amt_i128(0x01010101_01010101, 0x03030303_03030303, 128) == [0x01010101_01010101, 0x03030303_03030303]
+
+function %rotr_amt_i128(i64, i64, i8) -> i64, i64 {
+block0(v0: i64, v1: i64, v2: i8):
+    v3 = uextend.i64 v2
+    v4 = iconcat v3, v3
+
+    v5 = iconcat v0, v1
+
+    v6 = rotr.i128 v5, v4
+    v7, v8 = isplit v6
+    return v7, v8
+}
+; run: %rotr_amt_i128(0x01010101_01010101, 0x01010101_01010101, 9) == [0x80808080_80808080, 0x80808080_80808080]
+; run: %rotr_amt_i128(0x01010101_01010101, 0x01010101_01010101, 73) == [0x80808080_80808080, 0x80808080_80808080]
+; run: %rotr_amt_i128(0x01010101_01010101, 0x02020202_02020202, 0) == [0x01010101_01010101, 0x02020202_02020202]
+; run: %rotr_amt_i128(0x01010101_01010101, 0x03030303_03030303, 128) == [0x01010101_01010101, 0x03030303_03030303]