Improve code quality for push_new_empty_item.

[jrmuizel]

LLVM's ability to eliminate memcpy's across basic blocks is bad.
Because of this, we want to make sure that we avoid putting branches
in code where we want the memcpy elimination to happen.
rust-lang/rust#56172 has a reduced test case
of this happening.

This change lifts the branch caused by unwrap() above the creation of
the SpecificDisplayItem. It ends up saving a memcpy of 127 bytes along
with reducing pop_reference_frame by 18 instructions.

---

This change is 

[gw3583]

@bors-servo r+

[bors-servo]

📌 Commit 97a7db1 has been approved by gw3583

[jrmuizel]

This is sort of an ugly change so I'm interested to know how people feel about it.

Here's the before and after asm:

after:

webrender_api::display_list::DisplayListBuilder::pop_reference_frame (src/display_list.rs:1281):
 push    rbp
 mov     rbp, rsp
 sub     rsp, 368
 mov     rax, qword, ptr, [rdi, +, 40]
 test    rax, rax
 je      LBB514_2
 mov     rcx, qword, ptr, [rdi, +, 24]
 lea     rax, [rax, +, 2*rax]
 shl     rax, 4
 mov     rdx, qword, ptr, [rcx, +, rax, -, 8]
 mov     qword, ptr, [rbp, -, 8], rdx
 mov     rdx, qword, ptr, [rcx, +, rax, -, 16]
 mov     qword, ptr, [rbp, -, 16], rdx
 mov     rdx, qword, ptr, [rcx, +, rax, -, 24]
 mov     qword, ptr, [rbp, -, 24], rdx
 mov     rdx, qword, ptr, [rcx, +, rax, -, 32]
 mov     qword, ptr, [rbp, -, 32], rdx
 mov     rdx, qword, ptr, [rcx, +, rax, -, 48]
 mov     rax, qword, ptr, [rcx, +, rax, -, 40]
 mov     qword, ptr, [rbp, -, 40], rax
 mov     qword, ptr, [rbp, -, 48], rdx
 mov     qword, ptr, [rbp, -, 56], 0
 mov     qword, ptr, [rbp, -, 64], 0
 mov     qword, ptr, [rbp, -, 368], 18
 mov     rax, qword, ptr, [rbp, -, 48]
 mov     rcx, qword, ptr, [rbp, -, 40]
 mov     qword, ptr, [rbp, -, 192], rax
 mov     qword, ptr, [rbp, -, 184], rcx
 mov     rax, qword, ptr, [rbp, -, 32]
 mov     qword, ptr, [rbp, -, 176], rax
 mov     rax, qword, ptr, [rbp, -, 24]
 mov     qword, ptr, [rbp, -, 168], rax
 mov     rax, qword, ptr, [rbp, -, 16]
 mov     qword, ptr, [rbp, -, 160], rax
 mov     rax, qword, ptr, [rbp, -, 8]
 mov     qword, ptr, [rbp, -, 152], rax
 mov     qword, ptr, [rbp, -, 144], 0
 mov     rax, qword, ptr, [rbp, -, 80]
 mov     rcx, qword, ptr, [rbp, -, 72]
 mov     qword, ptr, [rbp, -, 136], rax
 mov     qword, ptr, [rbp, -, 128], rcx
 mov     rax, qword, ptr, [rbp, -, 64]
 mov     qword, ptr, [rbp, -, 120], rax
 mov     rax, qword, ptr, [rbp, -, 56]
 mov     qword, ptr, [rbp, -, 112], rax
 mov     qword, ptr, [rbp, -, 96], 0
 mov     qword, ptr, [rbp, -, 104], 0
 mov     byte, ptr, [rbp, -, 88], 1
 lea     rsi, [rbp, -, 368]
 call    webrender_api::display_list::serialize_fast
 add     rsp, 368
 pop     rbp
 ret
LBB514_2:
 lea     rdi, [rip, +, l_anon.1f7bba913a8d8ed49380105111efd8e9.29]
 call    core::panicking::panic

before:

webrender_api::display_list::DisplayListBuilder::pop_reference_frame (src/display_list.rs:1281):
 push    rbp
 mov     rbp, rsp
 push    rbx
 sub     rsp, 568
 xorps   xmm0, xmm0
 movaps  xmmword, ptr, [rbp, -, 80], xmm0
 mov     qword, ptr, [rbp, -, 88], 0
 mov     qword, ptr, [rbp, -, 96], 0
 mov     rax, qword, ptr, [rdi, +, 40]
 test    rax, rax
 je      LBB514_2
 mov     rbx, rdi
 mov     rcx, qword, ptr, [rdi, +, 24]
 lea     rax, [rax, +, 2*rax]
 shl     rax, 4
 mov     rdx, qword, ptr, [rcx, +, rax, -, 8]
 mov     qword, ptr, [rbp, -, 24], rdx
 mov     rdx, qword, ptr, [rcx, +, rax, -, 16]
 mov     qword, ptr, [rbp, -, 32], rdx
 mov     rdx, qword, ptr, [rcx, +, rax, -, 24]
 mov     qword, ptr, [rbp, -, 40], rdx
 mov     rdx, qword, ptr, [rcx, +, rax, -, 32]
 mov     qword, ptr, [rbp, -, 48], rdx
 mov     rdx, qword, ptr, [rcx, +, rax, -, 48]
 mov     rax, qword, ptr, [rcx, +, rax, -, 40]
 mov     qword, ptr, [rbp, -, 56], rax
 mov     qword, ptr, [rbp, -, 64], rdx
 mov     qword, ptr, [rbp, -, 400], 18
 lea     rdi, [rbp, -, 392]
 lea     rsi, [rbp, -, 568]
 mov     edx, 168
 call    _memcpy
 mov     rax, qword, ptr, [rbp, -, 64]
 mov     rcx, qword, ptr, [rbp, -, 56]
 mov     qword, ptr, [rbp, -, 224], rax
 mov     qword, ptr, [rbp, -, 216], rcx
 mov     rax, qword, ptr, [rbp, -, 48]
 mov     qword, ptr, [rbp, -, 208], rax
 mov     rax, qword, ptr, [rbp, -, 40]
 mov     qword, ptr, [rbp, -, 200], rax
 mov     rax, qword, ptr, [rbp, -, 32]
 mov     qword, ptr, [rbp, -, 192], rax
 mov     rax, qword, ptr, [rbp, -, 24]
 mov     qword, ptr, [rbp, -, 184], rax
 mov     qword, ptr, [rbp, -, 176], 0
 mov     rax, qword, ptr, [rbp, -, 112]
 mov     rcx, qword, ptr, [rbp, -, 104]
 mov     qword, ptr, [rbp, -, 168], rax
 mov     qword, ptr, [rbp, -, 160], rcx
 mov     rax, qword, ptr, [rbp, -, 96]
 mov     qword, ptr, [rbp, -, 152], rax
 mov     rax, qword, ptr, [rbp, -, 88]
 mov     qword, ptr, [rbp, -, 144], rax
 mov     rax, qword, ptr, [rbp, -, 80]
 mov     rcx, qword, ptr, [rbp, -, 72]
 mov     qword, ptr, [rbp, -, 128], rcx
 mov     qword, ptr, [rbp, -, 136], rax
 mov     byte, ptr, [rbp, -, 120], 1
 mov     al, byte, ptr, [rbp, -, 9]
 mov     byte, ptr, [rbp, -, 113], al
 movzx   eax, word, ptr, [rbp, -, 11]
 mov     word, ptr, [rbp, -, 115], ax
 mov     eax, dword, ptr, [rbp, -, 15]
 mov     dword, ptr, [rbp, -, 119], eax
 lea     rsi, [rbp, -, 400]
 mov     rdi, rbx
 call    webrender_api::display_list::serialize_fast
 add     rsp, 568
 pop     rbx
 pop     rbp
 ret
LBB514_2:
 lea     rdi, [rip, +, l_anon.cb11e9af7d531ec043b1c95d23ca69be.29]
 call    core::panicking::panic

[jrmuizel]

Ha. That was a more aggressive r+ than I was expecting :)

[bors-servo]

⌛ Testing commit 97a7db1 with merge b4f7e43...

[bors-servo]

☀️ Test successful - status-appveyor, status-taskcluster
Approved by: gw3583
Pushing b4f7e43 to master...