From 1bc36a0827f852994f7bf6813d29de26883801d4 Mon Sep 17 00:00:00 2001
From: Nick Fitzgerald <fitzgen@gmail.com>
Date: Tue, 26 Sep 2023 09:51:57 -0700
Subject: [PATCH] cranelift(x64): Make xmm{8..15} registers non-preferred

These registers require an additional byte to reference when encoded in certain
AVX instruction formats (and maybe other situations as well?) so prefer
xmm{0..7} when they are available and only fall back to xmm{8..15} when register
pressure is higher.
---
 cranelift/codegen/src/isa/x64/abi.rs          |  25 +--
 .../filetests/isa/x64/call-conv.clif          |  24 +--
 .../filetests/isa/x64/cmp-mem-bug.clif        |  16 +-
 .../filetests/filetests/isa/x64/fastcall.clif | 168 +++++++++---------
 .../filetests/isa/x64/fcopysign.clif          |  16 +-
 .../filetests/filetests/isa/x64/fcvt.clif     |  94 +++++-----
 .../filetests/isa/x64/float-avx.clif          |  36 ++--
 .../filetests/filetests/isa/x64/fma-call.clif |  32 ++--
 .../filetests/isa/x64/simd-arith-avx.clif     |  50 +++---
 .../filetests/isa/x64/simd-bitselect.clif     |  11 +-
 .../filetests/isa/x64/simd-bitwise-avx.clif   |   9 +-
 .../isa/x64/simd-bitwise-compile.clif         |  99 ++++++-----
 .../filetests/isa/x64/simd-cmp-avx.clif       |  72 ++++----
 .../isa/x64/simd-comparison-legalize.clif     |  10 +-
 .../filetests/isa/x64/simd-splat-avx.clif     |   8 +-
 .../filetests/isa/x64/simd-widen-mul.clif     |  36 ++--
 .../wasm/x64-relaxed-simd-deterministic.wat   |  76 ++++----
 .../filetests/wasm/x64-relaxed-simd.wat       |  34 ++--
 18 files changed, 416 insertions(+), 400 deletions(-)

diff --git a/cranelift/codegen/src/isa/x64/abi.rs b/cranelift/codegen/src/isa/x64/abi.rs
index 51758010162a..be005e102243 100644
--- a/cranelift/codegen/src/isa/x64/abi.rs
+++ b/cranelift/codegen/src/isa/x64/abi.rs
@@ -1243,7 +1243,8 @@ fn create_reg_env_systemv(enable_pinned_reg: bool) -> MachineEnv {
                 preg(regs::r10()),
                 preg(regs::r11()),
             ],
-            // Preferred XMMs: all of them.
+            // Preferred XMMs: the first 8, which can have smaller encodings
+            // with AVX instructions.
             vec![
                 preg(regs::xmm0()),
                 preg(regs::xmm1()),
@@ -1253,14 +1254,6 @@ fn create_reg_env_systemv(enable_pinned_reg: bool) -> MachineEnv {
                 preg(regs::xmm5()),
                 preg(regs::xmm6()),
                 preg(regs::xmm7()),
-                preg(regs::xmm8()),
-                preg(regs::xmm9()),
-                preg(regs::xmm10()),
-                preg(regs::xmm11()),
-                preg(regs::xmm12()),
-                preg(regs::xmm13()),
-                preg(regs::xmm14()),
-                preg(regs::xmm15()),
             ],
             // The Vector Regclass is unused
             vec![],
@@ -1273,8 +1266,18 @@ fn create_reg_env_systemv(enable_pinned_reg: bool) -> MachineEnv {
                 preg(regs::r13()),
                 preg(regs::r14()),
             ],
-            // Non-preferred XMMs: none.
-            vec![],
+            // Non-preferred XMMs: the last 8 registers, which can have larger
+            // encodings with AVX instructions.
+            vec![
+                preg(regs::xmm8()),
+                preg(regs::xmm9()),
+                preg(regs::xmm10()),
+                preg(regs::xmm11()),
+                preg(regs::xmm12()),
+                preg(regs::xmm13()),
+                preg(regs::xmm14()),
+                preg(regs::xmm15()),
+            ],
             // The Vector Regclass is unused
             vec![],
         ],
diff --git a/cranelift/filetests/filetests/isa/x64/call-conv.clif b/cranelift/filetests/filetests/isa/x64/call-conv.clif
index a4e383e817f0..1e0cede45e3b 100644
--- a/cranelift/filetests/filetests/isa/x64/call-conv.clif
+++ b/cranelift/filetests/filetests/isa/x64/call-conv.clif
@@ -202,8 +202,8 @@ block0(
 ;   movq    %rax, %rdi
 ;   movq    16(%rbp), %r10
 ;   movq    24(%rbp), %r11
-;   movss   32(%rbp), %xmm9
-;   movsd   40(%rbp), %xmm8
+;   movss   32(%rbp), %xmm11
+;   movsd   40(%rbp), %xmm13
 ;   subq    %rsp, $144, %rsp
 ;   virtual_sp_offset_adjust 144
 ;   movq    %r8, 32(%rsp)
@@ -218,8 +218,8 @@ block0(
 ;   movsd   %xmm7, 104(%rsp)
 ;   movq    %r10, 112(%rsp)
 ;   movl    %r11d, 120(%rsp)
-;   movss   %xmm9, 128(%rsp)
-;   movsd   %xmm8, 136(%rsp)
+;   movss   %xmm11, 128(%rsp)
+;   movsd   %xmm13, 136(%rsp)
 ;   movq    %rdi, %r9
 ;   movq    %rcx, %r8
 ;   movq    %rsi, %rcx
@@ -242,8 +242,8 @@ block0(
 ;   movq %rax, %rdi
 ;   movq 0x10(%rbp), %r10
 ;   movq 0x18(%rbp), %r11
-;   movss 0x20(%rbp), %xmm9
-;   movsd 0x28(%rbp), %xmm8
+;   movss 0x20(%rbp), %xmm11
+;   movsd 0x28(%rbp), %xmm13
 ;   subq $0x90, %rsp
 ;   movq %r8, 0x20(%rsp)
 ;   movq %r9, 0x28(%rsp)
@@ -257,8 +257,8 @@ block0(
 ;   movsd %xmm7, 0x68(%rsp)
 ;   movq %r10, 0x70(%rsp)
 ;   movl %r11d, 0x78(%rsp)
-;   movss %xmm9, 0x80(%rsp)
-;   movsd %xmm8, 0x88(%rsp)
+;   movss %xmm11, 0x80(%rsp)
+;   movsd %xmm13, 0x88(%rsp)
 ;   movq %rdi, %r9
 ;   movq %rcx, %r8
 ;   movq %rsi, %rcx
@@ -330,7 +330,7 @@ block0(v0: i32, v1: f32, v2: i64, v3: f64, v4: i32, v5: i32, v6: i32, v7: f32, v
 ; block0:
 ;   movq    %rsi, %r9
 ;   movq    %rdi, %rsi
-;   movdqa  %xmm1, %xmm12
+;   movdqa  %xmm1, %xmm6
 ;   movdqa  %xmm0, %xmm1
 ;   subq    %rsp, $96, %rsp
 ;   virtual_sp_offset_adjust 96
@@ -343,7 +343,7 @@ block0(v0: i32, v1: f32, v2: i64, v3: f64, v4: i32, v5: i32, v6: i32, v7: f32, v
 ;   movsd   %xmm5, 80(%rsp)
 ;   movq    %rsi, %rcx
 ;   movq    %r9, %r8
-;   movdqa  %xmm12, %xmm3
+;   movdqa  %xmm6, %xmm3
 ;   call    *%rcx
 ;   addq    %rsp, $96, %rsp
 ;   virtual_sp_offset_adjust -96
@@ -358,7 +358,7 @@ block0(v0: i32, v1: f32, v2: i64, v3: f64, v4: i32, v5: i32, v6: i32, v7: f32, v
 ; block1: ; offset 0x4
 ;   movq %rsi, %r9
 ;   movq %rdi, %rsi
-;   movdqa %xmm1, %xmm12
+;   movdqa %xmm1, %xmm6
 ;   movdqa %xmm0, %xmm1
 ;   subq $0x60, %rsp
 ;   movl %edx, 0x20(%rsp)
@@ -370,7 +370,7 @@ block0(v0: i32, v1: f32, v2: i64, v3: f64, v4: i32, v5: i32, v6: i32, v7: f32, v
 ;   movsd %xmm5, 0x50(%rsp)
 ;   movq %rsi, %rcx
 ;   movq %r9, %r8
-;   movdqa %xmm12, %xmm3
+;   movdqa %xmm6, %xmm3
 ;   callq *%rcx
 ;   addq $0x60, %rsp
 ;   movq %rbp, %rsp
diff --git a/cranelift/filetests/filetests/isa/x64/cmp-mem-bug.clif b/cranelift/filetests/filetests/isa/x64/cmp-mem-bug.clif
index 331239f54827..2574ddfb3de5 100644
--- a/cranelift/filetests/filetests/isa/x64/cmp-mem-bug.clif
+++ b/cranelift/filetests/filetests/isa/x64/cmp-mem-bug.clif
@@ -54,13 +54,13 @@ block0(v0: f64, v1: i64):
 ;   pushq   %rbp
 ;   movq    %rsp, %rbp
 ; block0:
-;   movsd   0(%rdi), %xmm9
-;   ucomisd %xmm9, %xmm0
+;   movsd   0(%rdi), %xmm1
+;   ucomisd %xmm1, %xmm0
 ;   setnp   %dil
 ;   setz    %al
 ;   andl    %edi, %eax, %edi
 ;   movzbq  %dil, %rax
-;   ucomisd %xmm9, %xmm0
+;   ucomisd %xmm1, %xmm0
 ;   movdqa  %xmm0, %xmm2
 ;   movsd %xmm0, %xmm0; jnp $next; movsd %xmm2, %xmm0; $next:
 ;   movsd %xmm0, %xmm0; jz $next; movsd %xmm2, %xmm0; $next:
@@ -73,17 +73,17 @@ block0(v0: f64, v1: i64):
 ;   pushq %rbp
 ;   movq %rsp, %rbp
 ; block1: ; offset 0x4
-;   movsd (%rdi), %xmm9 ; trap: heap_oob
-;   ucomisd %xmm9, %xmm0
+;   movsd (%rdi), %xmm1 ; trap: heap_oob
+;   ucomisd %xmm1, %xmm0
 ;   setnp %dil
 ;   sete %al
 ;   andl %eax, %edi
 ;   movzbq %dil, %rax
-;   ucomisd %xmm9, %xmm0
+;   ucomisd %xmm1, %xmm0
 ;   movdqa %xmm0, %xmm2
-;   jnp 0x2f
+;   jnp 0x2c
 ;   movsd %xmm2, %xmm0
-;   je 0x39
+;   je 0x36
 ;   movsd %xmm2, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
diff --git a/cranelift/filetests/filetests/isa/x64/fastcall.clif b/cranelift/filetests/filetests/isa/x64/fastcall.clif
index 0bf874402fdb..86e865b61493 100644
--- a/cranelift/filetests/filetests/isa/x64/fastcall.clif
+++ b/cranelift/filetests/filetests/isa/x64/fastcall.clif
@@ -356,51 +356,53 @@ block0(v0: i64):
 ;   unwind SaveReg { clobber_offset: 144, reg: p15f }
 ; block0:
 ;   movsd   0(%rcx), %xmm0
-;   movsd   8(%rcx), %xmm10
-;   movdqu  %xmm10, rsp(48 + virtual offset)
-;   movsd   16(%rcx), %xmm5
-;   movsd   24(%rcx), %xmm14
-;   movdqu  %xmm14, rsp(32 + virtual offset)
-;   movsd   32(%rcx), %xmm13
-;   movsd   40(%rcx), %xmm15
-;   movdqu  %xmm15, rsp(16 + virtual offset)
+;   movsd   8(%rcx), %xmm8
+;   movdqu  %xmm8, rsp(48 + virtual offset)
+;   movsd   16(%rcx), %xmm10
+;   movdqu  %xmm10, rsp(0 + virtual offset)
+;   movsd   24(%rcx), %xmm9
+;   movdqa  %xmm9, %xmm10
+;   movsd   32(%rcx), %xmm5
+;   movsd   40(%rcx), %xmm6
+;   movdqu  %xmm6, rsp(32 + virtual offset)
 ;   movsd   48(%rcx), %xmm7
-;   movsd   56(%rcx), %xmm8
-;   movdqu  %xmm8, rsp(0 + virtual offset)
-;   movsd   64(%rcx), %xmm12
-;   movsd   72(%rcx), %xmm2
-;   movsd   80(%rcx), %xmm9
-;   movsd   88(%rcx), %xmm4
+;   movsd   56(%rcx), %xmm12
+;   movdqu  %xmm12, rsp(16 + virtual offset)
+;   movsd   64(%rcx), %xmm4
+;   movsd   72(%rcx), %xmm12
+;   movsd   80(%rcx), %xmm1
+;   movsd   88(%rcx), %xmm14
 ;   movsd   96(%rcx), %xmm3
-;   movsd   104(%rcx), %xmm8
+;   movsd   104(%rcx), %xmm15
 ;   movsd   112(%rcx), %xmm11
-;   movsd   120(%rcx), %xmm10
-;   movsd   128(%rcx), %xmm6
-;   movsd   136(%rcx), %xmm14
-;   movsd   144(%rcx), %xmm1
-;   movdqu  rsp(48 + virtual offset), %xmm15
-;   addsd   %xmm0, %xmm15, %xmm0
-;   movdqu  rsp(32 + virtual offset), %xmm15
-;   addsd   %xmm5, %xmm15, %xmm5
-;   movdqu  rsp(16 + virtual offset), %xmm15
-;   addsd   %xmm13, %xmm15, %xmm13
-;   movdqu  rsp(0 + virtual offset), %xmm15
-;   addsd   %xmm7, %xmm15, %xmm7
-;   addsd   %xmm12, %xmm2, %xmm12
-;   addsd   %xmm9, %xmm4, %xmm9
-;   addsd   %xmm3, %xmm8, %xmm3
-;   addsd   %xmm11, %xmm10, %xmm11
-;   addsd   %xmm6, %xmm14, %xmm6
-;   addsd   %xmm1, 152(%rcx), %xmm1
-;   addsd   %xmm0, %xmm5, %xmm0
-;   addsd   %xmm13, %xmm7, %xmm13
-;   addsd   %xmm12, %xmm9, %xmm12
-;   addsd   %xmm3, %xmm11, %xmm3
-;   addsd   %xmm6, %xmm1, %xmm6
-;   addsd   %xmm0, %xmm13, %xmm0
-;   addsd   %xmm12, %xmm3, %xmm12
-;   addsd   %xmm0, %xmm12, %xmm0
+;   movsd   120(%rcx), %xmm8
+;   movsd   128(%rcx), %xmm2
+;   movsd   136(%rcx), %xmm9
+;   movsd   144(%rcx), %xmm13
+;   movdqu  rsp(48 + virtual offset), %xmm6
 ;   addsd   %xmm0, %xmm6, %xmm0
+;   movdqa  %xmm10, %xmm6
+;   movdqu  rsp(0 + virtual offset), %xmm10
+;   addsd   %xmm10, %xmm6, %xmm10
+;   movdqu  rsp(32 + virtual offset), %xmm6
+;   addsd   %xmm5, %xmm6, %xmm5
+;   movdqu  rsp(16 + virtual offset), %xmm6
+;   addsd   %xmm7, %xmm6, %xmm7
+;   addsd   %xmm4, %xmm12, %xmm4
+;   addsd   %xmm1, %xmm14, %xmm1
+;   addsd   %xmm3, %xmm15, %xmm3
+;   addsd   %xmm11, %xmm8, %xmm11
+;   addsd   %xmm2, %xmm9, %xmm2
+;   addsd   %xmm13, 152(%rcx), %xmm13
+;   addsd   %xmm0, %xmm10, %xmm0
+;   addsd   %xmm5, %xmm7, %xmm5
+;   addsd   %xmm4, %xmm1, %xmm4
+;   addsd   %xmm3, %xmm11, %xmm3
+;   addsd   %xmm2, %xmm13, %xmm2
+;   addsd   %xmm0, %xmm5, %xmm0
+;   addsd   %xmm4, %xmm3, %xmm4
+;   addsd   %xmm0, %xmm4, %xmm0
+;   addsd   %xmm0, %xmm2, %xmm0
 ;   movdqu  64(%rsp), %xmm6
 ;   movdqu  80(%rsp), %xmm7
 ;   movdqu  96(%rsp), %xmm8
@@ -433,51 +435,53 @@ block0(v0: i64):
 ;   movdqu %xmm15, 0xd0(%rsp)
 ; block1: ; offset 0x61
 ;   movsd (%rcx), %xmm0 ; trap: heap_oob
-;   movsd 8(%rcx), %xmm10 ; trap: heap_oob
-;   movdqu %xmm10, 0x30(%rsp)
-;   movsd 0x10(%rcx), %xmm5 ; trap: heap_oob
-;   movsd 0x18(%rcx), %xmm14 ; trap: heap_oob
-;   movdqu %xmm14, 0x20(%rsp)
-;   movsd 0x20(%rcx), %xmm13 ; trap: heap_oob
-;   movsd 0x28(%rcx), %xmm15 ; trap: heap_oob
-;   movdqu %xmm15, 0x10(%rsp)
+;   movsd 8(%rcx), %xmm8 ; trap: heap_oob
+;   movdqu %xmm8, 0x30(%rsp)
+;   movsd 0x10(%rcx), %xmm10 ; trap: heap_oob
+;   movdqu %xmm10, (%rsp)
+;   movsd 0x18(%rcx), %xmm9 ; trap: heap_oob
+;   movdqa %xmm9, %xmm10
+;   movsd 0x20(%rcx), %xmm5 ; trap: heap_oob
+;   movsd 0x28(%rcx), %xmm6 ; trap: heap_oob
+;   movdqu %xmm6, 0x20(%rsp)
 ;   movsd 0x30(%rcx), %xmm7 ; trap: heap_oob
-;   movsd 0x38(%rcx), %xmm8 ; trap: heap_oob
-;   movdqu %xmm8, (%rsp)
-;   movsd 0x40(%rcx), %xmm12 ; trap: heap_oob
-;   movsd 0x48(%rcx), %xmm2 ; trap: heap_oob
-;   movsd 0x50(%rcx), %xmm9 ; trap: heap_oob
-;   movsd 0x58(%rcx), %xmm4 ; trap: heap_oob
+;   movsd 0x38(%rcx), %xmm12 ; trap: heap_oob
+;   movdqu %xmm12, 0x10(%rsp)
+;   movsd 0x40(%rcx), %xmm4 ; trap: heap_oob
+;   movsd 0x48(%rcx), %xmm12 ; trap: heap_oob
+;   movsd 0x50(%rcx), %xmm1 ; trap: heap_oob
+;   movsd 0x58(%rcx), %xmm14 ; trap: heap_oob
 ;   movsd 0x60(%rcx), %xmm3 ; trap: heap_oob
-;   movsd 0x68(%rcx), %xmm8 ; trap: heap_oob
+;   movsd 0x68(%rcx), %xmm15 ; trap: heap_oob
 ;   movsd 0x70(%rcx), %xmm11 ; trap: heap_oob
-;   movsd 0x78(%rcx), %xmm10 ; trap: heap_oob
-;   movsd 0x80(%rcx), %xmm6 ; trap: heap_oob
-;   movsd 0x88(%rcx), %xmm14 ; trap: heap_oob
-;   movsd 0x90(%rcx), %xmm1 ; trap: heap_oob
-;   movdqu 0x30(%rsp), %xmm15
-;   addsd %xmm15, %xmm0
-;   movdqu 0x20(%rsp), %xmm15
-;   addsd %xmm15, %xmm5
-;   movdqu 0x10(%rsp), %xmm15
-;   addsd %xmm15, %xmm13
-;   movdqu (%rsp), %xmm15
-;   addsd %xmm15, %xmm7
-;   addsd %xmm2, %xmm12
-;   addsd %xmm4, %xmm9
-;   addsd %xmm8, %xmm3
-;   addsd %xmm10, %xmm11
-;   addsd %xmm14, %xmm6
-;   addsd 0x98(%rcx), %xmm1 ; trap: heap_oob
-;   addsd %xmm5, %xmm0
-;   addsd %xmm7, %xmm13
-;   addsd %xmm9, %xmm12
-;   addsd %xmm11, %xmm3
-;   addsd %xmm1, %xmm6
-;   addsd %xmm13, %xmm0
-;   addsd %xmm3, %xmm12
-;   addsd %xmm12, %xmm0
+;   movsd 0x78(%rcx), %xmm8 ; trap: heap_oob
+;   movsd 0x80(%rcx), %xmm2 ; trap: heap_oob
+;   movsd 0x88(%rcx), %xmm9 ; trap: heap_oob
+;   movsd 0x90(%rcx), %xmm13 ; trap: heap_oob
+;   movdqu 0x30(%rsp), %xmm6
 ;   addsd %xmm6, %xmm0
+;   movdqa %xmm10, %xmm6
+;   movdqu (%rsp), %xmm10
+;   addsd %xmm6, %xmm10
+;   movdqu 0x20(%rsp), %xmm6
+;   addsd %xmm6, %xmm5
+;   movdqu 0x10(%rsp), %xmm6
+;   addsd %xmm6, %xmm7
+;   addsd %xmm12, %xmm4
+;   addsd %xmm14, %xmm1
+;   addsd %xmm15, %xmm3
+;   addsd %xmm8, %xmm11
+;   addsd %xmm9, %xmm2
+;   addsd 0x98(%rcx), %xmm13 ; trap: heap_oob
+;   addsd %xmm10, %xmm0
+;   addsd %xmm7, %xmm5
+;   addsd %xmm1, %xmm4
+;   addsd %xmm11, %xmm3
+;   addsd %xmm13, %xmm2
+;   addsd %xmm5, %xmm0
+;   addsd %xmm3, %xmm4
+;   addsd %xmm4, %xmm0
+;   addsd %xmm2, %xmm0
 ;   movdqu 0x40(%rsp), %xmm6
 ;   movdqu 0x50(%rsp), %xmm7
 ;   movdqu 0x60(%rsp), %xmm8
diff --git a/cranelift/filetests/filetests/isa/x64/fcopysign.clif b/cranelift/filetests/filetests/isa/x64/fcopysign.clif
index a6140711c829..c2f39049ac96 100644
--- a/cranelift/filetests/filetests/isa/x64/fcopysign.clif
+++ b/cranelift/filetests/filetests/isa/x64/fcopysign.clif
@@ -13,9 +13,9 @@ block0(v0: f32, v1: f32):
 ; block0:
 ;   movl    $-2147483648, %ecx
 ;   movd    %ecx, %xmm7
-;   movdqa  %xmm0, %xmm10
+;   movdqa  %xmm0, %xmm2
 ;   movdqa  %xmm7, %xmm0
-;   andnps  %xmm0, %xmm10, %xmm0
+;   andnps  %xmm0, %xmm2, %xmm0
 ;   andps   %xmm7, %xmm1, %xmm7
 ;   orps    %xmm0, %xmm7, %xmm0
 ;   movq    %rbp, %rsp
@@ -29,9 +29,9 @@ block0(v0: f32, v1: f32):
 ; block1: ; offset 0x4
 ;   movl $0x80000000, %ecx
 ;   movd %ecx, %xmm7
-;   movdqa %xmm0, %xmm10
+;   movdqa %xmm0, %xmm2
 ;   movdqa %xmm7, %xmm0
-;   andnps %xmm10, %xmm0
+;   andnps %xmm2, %xmm0
 ;   andps %xmm1, %xmm7
 ;   orps %xmm7, %xmm0
 ;   movq %rbp, %rsp
@@ -50,9 +50,9 @@ block0(v0: f64, v1: f64):
 ; block0:
 ;   movabsq $-9223372036854775808, %rcx
 ;   movq    %rcx, %xmm7
-;   movdqa  %xmm0, %xmm10
+;   movdqa  %xmm0, %xmm2
 ;   movdqa  %xmm7, %xmm0
-;   andnpd  %xmm0, %xmm10, %xmm0
+;   andnpd  %xmm0, %xmm2, %xmm0
 ;   andpd   %xmm7, %xmm1, %xmm7
 ;   orpd    %xmm0, %xmm7, %xmm0
 ;   movq    %rbp, %rsp
@@ -66,9 +66,9 @@ block0(v0: f64, v1: f64):
 ; block1: ; offset 0x4
 ;   movabsq $9223372036854775808, %rcx
 ;   movq %rcx, %xmm7
-;   movdqa %xmm0, %xmm10
+;   movdqa %xmm0, %xmm2
 ;   movdqa %xmm7, %xmm0
-;   andnpd %xmm10, %xmm0
+;   andnpd %xmm2, %xmm0
 ;   andpd %xmm1, %xmm7
 ;   orpd %xmm7, %xmm0
 ;   movq %rbp, %rsp
diff --git a/cranelift/filetests/filetests/isa/x64/fcvt.clif b/cranelift/filetests/filetests/isa/x64/fcvt.clif
index cbe440745cf4..0dfa8637e0cd 100644
--- a/cranelift/filetests/filetests/isa/x64/fcvt.clif
+++ b/cranelift/filetests/filetests/isa/x64/fcvt.clif
@@ -257,10 +257,10 @@ block0(v0: i8, v1: i16, v2: i32, v3: i64):
 ;   cvtsi2ss %r9, %xmm1
 ;   movl    %edx, %r9d
 ;   cvtsi2ss %r9, %xmm2
-;   u64_to_f32_seq %rcx, %xmm14, %r9, %r10
+;   u64_to_f32_seq %rcx, %xmm6, %r9, %r10
 ;   addss   %xmm0, %xmm1, %xmm0
 ;   addss   %xmm0, %xmm2, %xmm0
-;   addss   %xmm0, %xmm14, %xmm0
+;   addss   %xmm0, %xmm6, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -278,18 +278,18 @@ block0(v0: i8, v1: i16, v2: i32, v3: i64):
 ;   cvtsi2ssq %r9, %xmm2
 ;   cmpq $0, %rcx
 ;   jl 0x32
-;   cvtsi2ssq %rcx, %xmm14
-;   jmp 0x4d
+;   cvtsi2ssq %rcx, %xmm6
+;   jmp 0x4c
 ;   movq %rcx, %r9
 ;   shrq $1, %r9
 ;   movq %rcx, %r10
 ;   andq $1, %r10
 ;   orq %r9, %r10
-;   cvtsi2ssq %r10, %xmm14
-;   addss %xmm14, %xmm14
+;   cvtsi2ssq %r10, %xmm6
+;   addss %xmm6, %xmm6
 ;   addss %xmm1, %xmm0
 ;   addss %xmm2, %xmm0
-;   addss %xmm14, %xmm0
+;   addss %xmm6, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -353,13 +353,12 @@ block0(v0: i32x4):
 ;   movdqa  %xmm0, %xmm3
 ;   pslld   %xmm3, $16, %xmm3
 ;   psrld   %xmm3, $16, %xmm3
-;   movdqa  %xmm0, %xmm9
-;   psubd   %xmm9, %xmm3, %xmm9
-;   cvtdq2ps %xmm3, %xmm8
-;   psrld   %xmm9, $1, %xmm9
-;   cvtdq2ps %xmm9, %xmm0
+;   psubd   %xmm0, %xmm3, %xmm0
+;   cvtdq2ps %xmm3, %xmm1
+;   psrld   %xmm0, $1, %xmm0
+;   cvtdq2ps %xmm0, %xmm0
 ;   addps   %xmm0, %xmm0, %xmm0
-;   addps   %xmm0, %xmm8, %xmm0
+;   addps   %xmm0, %xmm1, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -372,13 +371,12 @@ block0(v0: i32x4):
 ;   movdqa %xmm0, %xmm3
 ;   pslld $0x10, %xmm3
 ;   psrld $0x10, %xmm3
-;   movdqa %xmm0, %xmm9
-;   psubd %xmm3, %xmm9
-;   cvtdq2ps %xmm3, %xmm8
-;   psrld $1, %xmm9
-;   cvtdq2ps %xmm9, %xmm0
+;   psubd %xmm3, %xmm0
+;   cvtdq2ps %xmm3, %xmm1
+;   psrld $1, %xmm0
+;   cvtdq2ps %xmm0, %xmm0
 ;   addps %xmm0, %xmm0
-;   addps %xmm8, %xmm0
+;   addps %xmm1, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -1035,20 +1033,20 @@ block0(v0: f32x4):
 ; block0:
 ;   uninit  %xmm6
 ;   xorps   %xmm6, %xmm6, %xmm6
-;   movdqa  %xmm0, %xmm11
-;   maxps   %xmm11, %xmm6, %xmm11
+;   movdqa  %xmm0, %xmm3
+;   maxps   %xmm3, %xmm6, %xmm3
 ;   pcmpeqd %xmm6, %xmm6, %xmm6
 ;   psrld   %xmm6, $1, %xmm6
-;   cvtdq2ps %xmm6, %xmm15
-;   cvttps2dq %xmm11, %xmm14
-;   subps   %xmm11, %xmm15, %xmm11
-;   cmpps   $2, %xmm15, %xmm11, %xmm15
-;   cvttps2dq %xmm11, %xmm0
-;   pxor    %xmm0, %xmm15, %xmm0
-;   uninit  %xmm9
-;   pxor    %xmm9, %xmm9, %xmm9
-;   pmaxsd  %xmm0, %xmm9, %xmm0
-;   paddd   %xmm0, %xmm14, %xmm0
+;   cvtdq2ps %xmm6, %xmm7
+;   cvttps2dq %xmm3, %xmm6
+;   subps   %xmm3, %xmm7, %xmm3
+;   cmpps   $2, %xmm7, %xmm3, %xmm7
+;   cvttps2dq %xmm3, %xmm0
+;   pxor    %xmm0, %xmm7, %xmm0
+;   uninit  %xmm1
+;   pxor    %xmm1, %xmm1, %xmm1
+;   pmaxsd  %xmm0, %xmm1, %xmm0
+;   paddd   %xmm0, %xmm6, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -1059,19 +1057,19 @@ block0(v0: f32x4):
 ;   movq %rsp, %rbp
 ; block1: ; offset 0x4
 ;   xorps %xmm6, %xmm6
-;   movdqa %xmm0, %xmm11
-;   maxps %xmm6, %xmm11
+;   movdqa %xmm0, %xmm3
+;   maxps %xmm6, %xmm3
 ;   pcmpeqd %xmm6, %xmm6
 ;   psrld $1, %xmm6
-;   cvtdq2ps %xmm6, %xmm15
-;   cvttps2dq %xmm11, %xmm14
-;   subps %xmm15, %xmm11
-;   cmpleps %xmm11, %xmm15
-;   cvttps2dq %xmm11, %xmm0
-;   pxor %xmm15, %xmm0
-;   pxor %xmm9, %xmm9
-;   pmaxsd %xmm9, %xmm0
-;   paddd %xmm14, %xmm0
+;   cvtdq2ps %xmm6, %xmm7
+;   cvttps2dq %xmm3, %xmm6
+;   subps %xmm7, %xmm3
+;   cmpleps %xmm3, %xmm7
+;   cvttps2dq %xmm3, %xmm0
+;   pxor %xmm7, %xmm0
+;   pxor %xmm1, %xmm1
+;   pmaxsd %xmm1, %xmm0
+;   paddd %xmm6, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -1091,11 +1089,11 @@ block0(v0: f32x4):
 ;   movdqa  %xmm0, %xmm5
 ;   andps   %xmm5, %xmm4, %xmm5
 ;   pxor    %xmm4, %xmm5, %xmm4
-;   cvttps2dq %xmm5, %xmm8
-;   movdqa  %xmm8, %xmm0
+;   cvttps2dq %xmm5, %xmm1
+;   movdqa  %xmm1, %xmm0
 ;   pand    %xmm0, %xmm4, %xmm0
 ;   psrad   %xmm0, $31, %xmm0
-;   pxor    %xmm0, %xmm8, %xmm0
+;   pxor    %xmm0, %xmm1, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -1110,11 +1108,11 @@ block0(v0: f32x4):
 ;   movdqa %xmm0, %xmm5
 ;   andps %xmm4, %xmm5
 ;   pxor %xmm5, %xmm4
-;   cvttps2dq %xmm5, %xmm8
-;   movdqa %xmm8, %xmm0
+;   cvttps2dq %xmm5, %xmm1
+;   movdqa %xmm1, %xmm0
 ;   pand %xmm4, %xmm0
 ;   psrad $0x1f, %xmm0
-;   pxor %xmm8, %xmm0
+;   pxor %xmm1, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
diff --git a/cranelift/filetests/filetests/isa/x64/float-avx.clif b/cranelift/filetests/filetests/isa/x64/float-avx.clif
index 948056a03d14..e7ddfa2fb7e4 100644
--- a/cranelift/filetests/filetests/isa/x64/float-avx.clif
+++ b/cranelift/filetests/filetests/isa/x64/float-avx.clif
@@ -444,11 +444,11 @@ block0(v0: i32x4):
 ;   vpslld  %xmm0, $16, %xmm2
 ;   vpsrld  %xmm2, $16, %xmm4
 ;   vpsubd  %xmm0, %xmm4, %xmm6
-;   vcvtdq2ps %xmm4, %xmm8
-;   vpsrld  %xmm6, $1, %xmm10
-;   vcvtdq2ps %xmm10, %xmm12
-;   vaddps  %xmm12, %xmm12, %xmm14
-;   vaddps  %xmm14, %xmm8, %xmm0
+;   vcvtdq2ps %xmm4, %xmm0
+;   vpsrld  %xmm6, $1, %xmm2
+;   vcvtdq2ps %xmm2, %xmm4
+;   vaddps  %xmm4, %xmm4, %xmm6
+;   vaddps  %xmm6, %xmm0, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -461,11 +461,11 @@ block0(v0: i32x4):
 ;   vpslld $0x10, %xmm0, %xmm2
 ;   vpsrld $0x10, %xmm2, %xmm4
 ;   vpsubd %xmm4, %xmm0, %xmm6
-;   vcvtdq2ps %xmm4, %xmm8
-;   vpsrld $1, %xmm6, %xmm10
-;   vcvtdq2ps %xmm10, %xmm12
-;   vaddps %xmm12, %xmm12, %xmm14
-;   vaddps %xmm8, %xmm14, %xmm0
+;   vcvtdq2ps %xmm4, %xmm0
+;   vpsrld $1, %xmm6, %xmm2
+;   vcvtdq2ps %xmm2, %xmm4
+;   vaddps %xmm4, %xmm4, %xmm6
+;   vaddps %xmm0, %xmm6, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -533,10 +533,10 @@ block0(v0: f32x4):
 ;   vcmpps  $0, %xmm0, %xmm0, %xmm2
 ;   vandps  %xmm0, %xmm2, %xmm4
 ;   vpxor   %xmm2, %xmm4, %xmm6
-;   vcvttps2dq %xmm4, %xmm8
-;   vpand   %xmm8, %xmm6, %xmm10
-;   vpsrad  %xmm10, $31, %xmm12
-;   vpxor   %xmm12, %xmm8, %xmm0
+;   vcvttps2dq %xmm4, %xmm0
+;   vpand   %xmm0, %xmm6, %xmm2
+;   vpsrad  %xmm2, $31, %xmm4
+;   vpxor   %xmm4, %xmm0, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -549,10 +549,10 @@ block0(v0: f32x4):
 ;   vcmpeqps %xmm0, %xmm0, %xmm2
 ;   vandps %xmm2, %xmm0, %xmm4
 ;   vpxor %xmm4, %xmm2, %xmm6
-;   vcvttps2dq %xmm4, %xmm8
-;   vpand %xmm6, %xmm8, %xmm10
-;   vpsrad $0x1f, %xmm10, %xmm12
-;   vpxor %xmm8, %xmm12, %xmm0
+;   vcvttps2dq %xmm4, %xmm0
+;   vpand %xmm6, %xmm0, %xmm2
+;   vpsrad $0x1f, %xmm2, %xmm4
+;   vpxor %xmm0, %xmm4, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
diff --git a/cranelift/filetests/filetests/isa/x64/fma-call.clif b/cranelift/filetests/filetests/isa/x64/fma-call.clif
index 0b92c3f44e79..3bb595a4fe26 100644
--- a/cranelift/filetests/filetests/isa/x64/fma-call.clif
+++ b/cranelift/filetests/filetests/isa/x64/fma-call.clif
@@ -77,17 +77,17 @@ block0(v0: f32x4, v1: f32x4, v2: f32x4):
 ;   movdqu  %xmm0, rsp(48 + virtual offset)
 ;   movdqu  rsp(0 + virtual offset), %xmm4
 ;   pshufd  $1, %xmm4, %xmm0
-;   movdqu  rsp(16 + virtual offset), %xmm8
-;   pshufd  $1, %xmm8, %xmm1
-;   movdqu  rsp(32 + virtual offset), %xmm12
-;   pshufd  $1, %xmm12, %xmm2
+;   movdqu  rsp(16 + virtual offset), %xmm2
+;   pshufd  $1, %xmm2, %xmm1
+;   movdqu  rsp(32 + virtual offset), %xmm4
+;   pshufd  $1, %xmm4, %xmm2
 ;   load_ext_name %FmaF32+0, %r9
 ;   call    *%r9
 ;   movdqu  %xmm0, rsp(64 + virtual offset)
 ;   movdqu  rsp(0 + virtual offset), %xmm4
 ;   pshufd  $2, %xmm4, %xmm0
-;   movdqu  rsp(16 + virtual offset), %xmm14
-;   pshufd  $2, %xmm14, %xmm1
+;   movdqu  rsp(16 + virtual offset), %xmm6
+;   pshufd  $2, %xmm6, %xmm1
 ;   movdqu  rsp(32 + virtual offset), %xmm3
 ;   pshufd  $2, %xmm3, %xmm2
 ;   load_ext_name %FmaF32+0, %r10
@@ -131,17 +131,17 @@ block0(v0: f32x4, v1: f32x4, v2: f32x4):
 ;   movdqu %xmm0, 0x30(%rsp)
 ;   movdqu (%rsp), %xmm4
 ;   pshufd $1, %xmm4, %xmm0
-;   movdqu 0x10(%rsp), %xmm8
-;   pshufd $1, %xmm8, %xmm1
-;   movdqu 0x20(%rsp), %xmm12
-;   pshufd $1, %xmm12, %xmm2
+;   movdqu 0x10(%rsp), %xmm2
+;   pshufd $1, %xmm2, %xmm1
+;   movdqu 0x20(%rsp), %xmm4
+;   pshufd $1, %xmm4, %xmm2
 ;   movabsq $0, %r9 ; reloc_external Abs8 %FmaF32 0
 ;   callq *%r9
 ;   movdqu %xmm0, 0x40(%rsp)
 ;   movdqu (%rsp), %xmm4
 ;   pshufd $2, %xmm4, %xmm0
-;   movdqu 0x10(%rsp), %xmm14
-;   pshufd $2, %xmm14, %xmm1
+;   movdqu 0x10(%rsp), %xmm6
+;   pshufd $2, %xmm6, %xmm1
 ;   movdqu 0x20(%rsp), %xmm3
 ;   pshufd $2, %xmm3, %xmm2
 ;   movabsq $0, %r10 ; reloc_external Abs8 %FmaF32 0
@@ -196,9 +196,9 @@ block0(v0: f64x2, v1: f64x2, v2: f64x2):
 ;   pshufd  $238, %xmm2, %xmm2
 ;   load_ext_name %FmaF64+0, %r9
 ;   call    *%r9
-;   movdqa  %xmm0, %xmm14
+;   movdqa  %xmm0, %xmm6
 ;   movdqu  rsp(48 + virtual offset), %xmm0
-;   movlhps %xmm0, %xmm14, %xmm0
+;   movlhps %xmm0, %xmm6, %xmm0
 ;   addq    %rsp, $64, %rsp
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
@@ -227,9 +227,9 @@ block0(v0: f64x2, v1: f64x2, v2: f64x2):
 ;   pshufd $0xee, %xmm2, %xmm2
 ;   movabsq $0, %r9 ; reloc_external Abs8 %FmaF64 0
 ;   callq *%r9
-;   movdqa %xmm0, %xmm14
+;   movdqa %xmm0, %xmm6
 ;   movdqu 0x30(%rsp), %xmm0
-;   movlhps %xmm14, %xmm0
+;   movlhps %xmm6, %xmm0
 ;   addq $0x40, %rsp
 ;   movq %rbp, %rsp
 ;   popq %rbp
diff --git a/cranelift/filetests/filetests/isa/x64/simd-arith-avx.clif b/cranelift/filetests/filetests/isa/x64/simd-arith-avx.clif
index f8c390ba8a17..8cd260a3f536 100644
--- a/cranelift/filetests/filetests/isa/x64/simd-arith-avx.clif
+++ b/cranelift/filetests/filetests/isa/x64/simd-arith-avx.clif
@@ -919,10 +919,10 @@ block0(v0: i8x16, v1: i32):
 ;   vpunpcklbw %xmm0, %xmm0, %xmm5
 ;   vpunpckhbw %xmm0, %xmm0, %xmm7
 ;   addl    %r9d, $8, %r9d
-;   vmovd   %r9d, %xmm11
-;   vpsraw  %xmm5, %xmm11, %xmm13
-;   vpsraw  %xmm7, %xmm11, %xmm15
-;   vpacksswb %xmm13, %xmm15, %xmm0
+;   vmovd   %r9d, %xmm3
+;   vpsraw  %xmm5, %xmm3, %xmm5
+;   vpsraw  %xmm7, %xmm3, %xmm7
+;   vpacksswb %xmm5, %xmm7, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -937,10 +937,10 @@ block0(v0: i8x16, v1: i32):
 ;   vpunpcklbw %xmm0, %xmm0, %xmm5
 ;   vpunpckhbw %xmm0, %xmm0, %xmm7
 ;   addl $8, %r9d
-;   vmovd %r9d, %xmm11
-;   vpsraw %xmm11, %xmm5, %xmm13
-;   vpsraw %xmm11, %xmm7, %xmm15
-;   vpacksswb %xmm15, %xmm13, %xmm0
+;   vmovd %r9d, %xmm3
+;   vpsraw %xmm3, %xmm5, %xmm5
+;   vpsraw %xmm3, %xmm7, %xmm7
+;   vpacksswb %xmm7, %xmm5, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -959,8 +959,8 @@ block0(v0: i8x16):
 ;   vpunpcklbw %xmm0, %xmm0, %xmm2
 ;   vpunpckhbw %xmm0, %xmm0, %xmm4
 ;   vpsraw  %xmm2, $11, %xmm6
-;   vpsraw  %xmm4, $11, %xmm8
-;   vpacksswb %xmm6, %xmm8, %xmm0
+;   vpsraw  %xmm4, $11, %xmm0
+;   vpacksswb %xmm6, %xmm0, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -973,8 +973,8 @@ block0(v0: i8x16):
 ;   vpunpcklbw %xmm0, %xmm0, %xmm2
 ;   vpunpckhbw %xmm0, %xmm0, %xmm4
 ;   vpsraw $0xb, %xmm2, %xmm6
-;   vpsraw $0xb, %xmm4, %xmm8
-;   vpacksswb %xmm8, %xmm6, %xmm0
+;   vpsraw $0xb, %xmm4, %xmm0
+;   vpacksswb %xmm0, %xmm6, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -1350,10 +1350,10 @@ block0(v0: f64x2):
 ;   uninit  %xmm2
 ;   vxorpd  %xmm2, %xmm2, %xmm4
 ;   vmaxpd  %xmm0, %xmm4, %xmm6
-;   vminpd  %xmm6, const(0), %xmm8
-;   vroundpd $3, %xmm8, %xmm10
-;   vaddpd  %xmm10, const(1), %xmm12
-;   vshufps $136, %xmm12, %xmm4, %xmm0
+;   vminpd  %xmm6, const(0), %xmm0
+;   vroundpd $3, %xmm0, %xmm2
+;   vaddpd  %xmm2, const(1), %xmm5
+;   vshufps $136, %xmm5, %xmm4, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -1365,10 +1365,10 @@ block0(v0: f64x2):
 ; block1: ; offset 0x4
 ;   vxorpd %xmm2, %xmm2, %xmm4
 ;   vmaxpd %xmm4, %xmm0, %xmm6
-;   vminpd 0x1c(%rip), %xmm6, %xmm8
-;   vroundpd $3, %xmm8, %xmm10
-;   vaddpd 0x1e(%rip), %xmm10, %xmm12
-;   vshufps $0x88, %xmm4, %xmm12, %xmm0
+;   vminpd 0x1c(%rip), %xmm6, %xmm0
+;   vroundpd $3, %xmm0, %xmm2
+;   vaddpd 0x1e(%rip), %xmm2, %xmm5
+;   vshufps $0x88, %xmm4, %xmm5, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -1393,8 +1393,8 @@ block0(v0: i8x16, v1: i32):
 ;   vpsllw  %xmm0, %xmm5, %xmm7
 ;   lea     const(0), %rsi
 ;   shlq    $4, %r10, %r10
-;   vmovdqu 0(%rsi,%r10,1), %xmm13
-;   vpand   %xmm7, %xmm13, %xmm0
+;   vmovdqu 0(%rsi,%r10,1), %xmm5
+;   vpand   %xmm7, %xmm5, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -1410,12 +1410,12 @@ block0(v0: i8x16, v1: i32):
 ;   vpsllw %xmm5, %xmm0, %xmm7
 ;   leaq 0x15(%rip), %rsi
 ;   shlq $4, %r10
-;   vmovdqu (%rsi, %r10), %xmm13
-;   vpand %xmm13, %xmm7, %xmm0
+;   vmovdqu (%rsi, %r10), %xmm5
+;   vpand %xmm5, %xmm7, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
-;   addb %bh, %bh
+;   addb %al, (%rax)
 
 function %i8x16_shl_imm(i8x16) -> i8x16 {
 block0(v0: i8x16):
diff --git a/cranelift/filetests/filetests/isa/x64/simd-bitselect.clif b/cranelift/filetests/filetests/isa/x64/simd-bitselect.clif
index 50eb59680e30..6ffc922ee764 100644
--- a/cranelift/filetests/filetests/isa/x64/simd-bitselect.clif
+++ b/cranelift/filetests/filetests/isa/x64/simd-bitselect.clif
@@ -211,9 +211,9 @@ block0(v0: i8x16, v1: i8x16):
 ;   pushq   %rbp
 ;   movq    %rsp, %rbp
 ; block0:
-;   movdqa  %xmm0, %xmm8
+;   movdqa  %xmm0, %xmm2
 ;   movdqu  const(0), %xmm0
-;   movdqa  %xmm8, %xmm4
+;   movdqa  %xmm2, %xmm4
 ;   pand    %xmm4, %xmm0, %xmm4
 ;   pandn   %xmm0, %xmm1, %xmm0
 ;   por     %xmm0, %xmm4, %xmm0
@@ -226,9 +226,9 @@ block0(v0: i8x16, v1: i8x16):
 ;   pushq %rbp
 ;   movq %rsp, %rbp
 ; block1: ; offset 0x4
-;   movdqa %xmm0, %xmm8
-;   movdqu 0x1f(%rip), %xmm0
-;   movdqa %xmm8, %xmm4
+;   movdqa %xmm0, %xmm2
+;   movdqu 0x20(%rip), %xmm0
+;   movdqa %xmm2, %xmm4
 ;   pand %xmm0, %xmm4
 ;   pandn %xmm1, %xmm0
 ;   por %xmm4, %xmm0
@@ -240,6 +240,7 @@ block0(v0: i8x16, v1: i8x16):
 ;   addb %al, (%rax)
 ;   addb %al, (%rax)
 ;   addb %al, (%rax)
+;   addb %al, (%rax)
 ;   addb %dh, %al
 ;   addb %al, (%rax)
 ;   incl (%rax)
diff --git a/cranelift/filetests/filetests/isa/x64/simd-bitwise-avx.clif b/cranelift/filetests/filetests/isa/x64/simd-bitwise-avx.clif
index 8b540989c2b1..0de407a7abc7 100644
--- a/cranelift/filetests/filetests/isa/x64/simd-bitwise-avx.clif
+++ b/cranelift/filetests/filetests/isa/x64/simd-bitwise-avx.clif
@@ -42,8 +42,8 @@ block0(v0: i64):
 ;   movl    $-2147483648, %eax
 ;   vmovd   %eax, %xmm4
 ;   vandnps %xmm4, const(0), %xmm6
-;   vandps  %xmm4, 0(%rdi), %xmm8
-;   vorps   %xmm6, %xmm8, %xmm0
+;   vandps  %xmm4, 0(%rdi), %xmm0
+;   vorps   %xmm6, %xmm0, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -56,8 +56,8 @@ block0(v0: i64):
 ;   movl $0x80000000, %eax
 ;   vmovd %eax, %xmm4
 ;   vandnps 0x1b(%rip), %xmm4, %xmm6
-;   vandps (%rdi), %xmm4, %xmm8
-;   vorps %xmm8, %xmm6, %xmm0
+;   vandps (%rdi), %xmm4, %xmm0
+;   vorps %xmm0, %xmm6, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -75,6 +75,7 @@ block0(v0: i64):
 ;   addb %al, (%rax)
 ;   addb %al, (%rax)
 ;   addb %al, (%rax)
+;   addb %al, (%rax)
 
 function %bor_f32x4(f32x4, f32x4) -> f32x4 {
 block0(v0: f32x4, v1: f32x4):
diff --git a/cranelift/filetests/filetests/isa/x64/simd-bitwise-compile.clif b/cranelift/filetests/filetests/isa/x64/simd-bitwise-compile.clif
index 817598710f4e..b1ed1466069f 100644
--- a/cranelift/filetests/filetests/isa/x64/simd-bitwise-compile.clif
+++ b/cranelift/filetests/filetests/isa/x64/simd-bitwise-compile.clif
@@ -337,8 +337,8 @@ block0(v0: i32):
 ;   psllw   %xmm0, %xmm5, %xmm0
 ;   lea     const(0), %rsi
 ;   shlq    $4, %r10, %r10
-;   movdqu  0(%rsi,%r10,1), %xmm13
-;   pand    %xmm0, %xmm13, %xmm0
+;   movdqu  0(%rsi,%r10,1), %xmm5
+;   pand    %xmm0, %xmm5, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -355,8 +355,8 @@ block0(v0: i32):
 ;   psllw %xmm5, %xmm0
 ;   leaq 0x2d(%rip), %rsi
 ;   shlq $4, %r10
-;   movdqu (%rsi, %r10), %xmm13
-;   pand %xmm13, %xmm0
+;   movdqu (%rsi, %r10), %xmm5
+;   pand %xmm5, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -365,10 +365,9 @@ block0(v0: i32):
 ;   addb %al, (%rax)
 ;   addb %al, (%rax)
 ;   addb %al, (%rax)
-;   addl %eax, (%rdx)
-;   addl 0x9080706(, %rax), %eax
-;   orb (%rbx), %cl
-;   orb $0xd, %al
+;   addb %al, (%rcx)
+;   addb (%rbx), %al
+;   addb $5, %al
 
 function %ishl_i8x16_imm(i8x16) -> i8x16 {
 block0(v0: i8x16):
@@ -605,17 +604,17 @@ block0(v0: i32):
 ;   pushq   %rbp
 ;   movq    %rsp, %rbp
 ; block0:
-;   movdqu  const(0), %xmm8
+;   movdqu  const(0), %xmm1
 ;   movq    %rdi, %r9
 ;   andq    %r9, $7, %r9
-;   movdqa  %xmm8, %xmm0
-;   punpcklbw %xmm0, %xmm8, %xmm0
-;   punpckhbw %xmm8, %xmm8, %xmm8
+;   movdqa  %xmm1, %xmm0
+;   punpcklbw %xmm0, %xmm1, %xmm0
+;   punpckhbw %xmm1, %xmm1, %xmm1
 ;   addl    %r9d, $8, %r9d
-;   movd    %r9d, %xmm11
-;   psraw   %xmm0, %xmm11, %xmm0
-;   psraw   %xmm8, %xmm11, %xmm8
-;   packsswb %xmm0, %xmm8, %xmm0
+;   movd    %r9d, %xmm3
+;   psraw   %xmm0, %xmm3, %xmm0
+;   psraw   %xmm1, %xmm3, %xmm1
+;   packsswb %xmm0, %xmm1, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -625,23 +624,28 @@ block0(v0: i32):
 ;   pushq %rbp
 ;   movq %rsp, %rbp
 ; block1: ; offset 0x4
-;   movdqu 0x33(%rip), %xmm8
+;   movdqu 0x34(%rip), %xmm1
 ;   movq %rdi, %r9
 ;   andq $7, %r9
-;   movdqa %xmm8, %xmm0
-;   punpcklbw %xmm8, %xmm0
-;   punpckhbw %xmm8, %xmm8
+;   movdqa %xmm1, %xmm0
+;   punpcklbw %xmm1, %xmm0
+;   punpckhbw %xmm1, %xmm1
 ;   addl $8, %r9d
-;   movd %r9d, %xmm11
-;   psraw %xmm11, %xmm0
-;   psraw %xmm11, %xmm8
-;   packsswb %xmm8, %xmm0
+;   movd %r9d, %xmm3
+;   psraw %xmm3, %xmm0
+;   psraw %xmm3, %xmm1
+;   packsswb %xmm1, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
-;   addb %al, (%rcx)
-;   addb (%rbx), %al
-;   addb $5, %al
+;   addb %al, (%rax)
+;   addb %al, (%rax)
+;   addb %al, (%rax)
+;   addb %al, (%rax)
+;   addl %eax, (%rdx)
+;   addl 0x9080706(, %rax), %eax
+;   orb (%rbx), %cl
+;   orb $0xd, %al
 
 function %sshr_i8x16_imm(i8x16, i32) -> i8x16 {
 block0(v0: i8x16, v1: i32):
@@ -655,10 +659,10 @@ block0(v0: i8x16, v1: i32):
 ; block0:
 ;   movdqa  %xmm0, %xmm7
 ;   punpcklbw %xmm7, %xmm0, %xmm7
-;   movdqa  %xmm7, %xmm8
+;   movdqa  %xmm7, %xmm1
 ;   movdqa  %xmm0, %xmm7
 ;   punpckhbw %xmm7, %xmm0, %xmm7
-;   movdqa  %xmm8, %xmm0
+;   movdqa  %xmm1, %xmm0
 ;   psraw   %xmm0, $11, %xmm0
 ;   psraw   %xmm7, $11, %xmm7
 ;   packsswb %xmm0, %xmm7, %xmm0
@@ -673,10 +677,10 @@ block0(v0: i8x16, v1: i32):
 ; block1: ; offset 0x4
 ;   movdqa %xmm0, %xmm7
 ;   punpcklbw %xmm0, %xmm7
-;   movdqa %xmm7, %xmm8
+;   movdqa %xmm7, %xmm1
 ;   movdqa %xmm0, %xmm7
 ;   punpckhbw %xmm0, %xmm7
-;   movdqa %xmm8, %xmm0
+;   movdqa %xmm1, %xmm0
 ;   psraw $0xb, %xmm0
 ;   psraw $0xb, %xmm7
 ;   packsswb %xmm7, %xmm0
@@ -899,13 +903,13 @@ block0(v0: i64x2, v1: i32):
 ;   movq    %rdi, %rcx
 ;   andq    %rcx, $63, %rcx
 ;   movq    %rcx, %xmm5
-;   movdqu  const(0), %xmm8
-;   psrlq   %xmm8, %xmm5, %xmm8
-;   movdqa  %xmm0, %xmm11
-;   psrlq   %xmm11, %xmm5, %xmm11
-;   movdqa  %xmm8, %xmm0
-;   pxor    %xmm0, %xmm11, %xmm0
-;   psubq   %xmm0, %xmm8, %xmm0
+;   movdqu  const(0), %xmm1
+;   psrlq   %xmm1, %xmm5, %xmm1
+;   movdqa  %xmm0, %xmm3
+;   psrlq   %xmm3, %xmm5, %xmm3
+;   movdqa  %xmm1, %xmm0
+;   pxor    %xmm0, %xmm3, %xmm0
+;   psubq   %xmm0, %xmm1, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -918,13 +922,13 @@ block0(v0: i64x2, v1: i32):
 ;   movq %rdi, %rcx
 ;   andq $0x3f, %rcx
 ;   movq %rcx, %xmm5
-;   movdqu 0x27(%rip), %xmm8
-;   psrlq %xmm5, %xmm8
-;   movdqa %xmm0, %xmm11
-;   psrlq %xmm5, %xmm11
-;   movdqa %xmm8, %xmm0
-;   pxor %xmm11, %xmm0
-;   psubq %xmm8, %xmm0
+;   movdqu 0x28(%rip), %xmm1
+;   psrlq %xmm5, %xmm1
+;   movdqa %xmm0, %xmm3
+;   psrlq %xmm5, %xmm3
+;   movdqa %xmm1, %xmm0
+;   pxor %xmm3, %xmm0
+;   psubq %xmm1, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -935,4 +939,9 @@ block0(v0: i64x2, v1: i32):
 ;   addb %al, (%rax)
 ;   addb %al, (%rax)
 ;   addb %al, (%rax)
+;   addb %al, (%rax)
+;   addb %al, (%rax)
+;   addb $0, (%rax)
+;   addb %al, (%rax)
+;   addb %al, (%rax)
 
diff --git a/cranelift/filetests/filetests/isa/x64/simd-cmp-avx.clif b/cranelift/filetests/filetests/isa/x64/simd-cmp-avx.clif
index bccc8657cb9a..60bd126554ad 100644
--- a/cranelift/filetests/filetests/isa/x64/simd-cmp-avx.clif
+++ b/cranelift/filetests/filetests/isa/x64/simd-cmp-avx.clif
@@ -214,10 +214,10 @@ block0(v0: f32x4, v1: f32x4):
 ;   vminps  %xmm0, %xmm1, %xmm3
 ;   vminps  %xmm1, %xmm0, %xmm5
 ;   vorps   %xmm3, %xmm5, %xmm7
-;   vcmpps  $3, %xmm7, %xmm5, %xmm9
-;   vorps   %xmm7, %xmm9, %xmm11
-;   vpsrld  %xmm9, $10, %xmm13
-;   vandnps %xmm13, %xmm11, %xmm0
+;   vcmpps  $3, %xmm7, %xmm5, %xmm1
+;   vorps   %xmm7, %xmm1, %xmm3
+;   vpsrld  %xmm1, $10, %xmm5
+;   vandnps %xmm5, %xmm3, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -230,10 +230,10 @@ block0(v0: f32x4, v1: f32x4):
 ;   vminps %xmm1, %xmm0, %xmm3
 ;   vminps %xmm0, %xmm1, %xmm5
 ;   vorps %xmm5, %xmm3, %xmm7
-;   vcmpunordps %xmm5, %xmm7, %xmm9
-;   vorps %xmm9, %xmm7, %xmm11
-;   vpsrld $0xa, %xmm9, %xmm13
-;   vandnps %xmm11, %xmm13, %xmm0
+;   vcmpunordps %xmm5, %xmm7, %xmm1
+;   vorps %xmm1, %xmm7, %xmm3
+;   vpsrld $0xa, %xmm1, %xmm5
+;   vandnps %xmm3, %xmm5, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -251,10 +251,10 @@ block0(v0: f64x2, v1: f64x2):
 ;   vminpd  %xmm0, %xmm1, %xmm3
 ;   vminpd  %xmm1, %xmm0, %xmm5
 ;   vorpd   %xmm3, %xmm5, %xmm7
-;   vcmppd  $3, %xmm3, %xmm5, %xmm9
-;   vorpd   %xmm7, %xmm9, %xmm11
-;   vpsrlq  %xmm9, $13, %xmm13
-;   vandnpd %xmm13, %xmm11, %xmm0
+;   vcmppd  $3, %xmm3, %xmm5, %xmm1
+;   vorpd   %xmm7, %xmm1, %xmm3
+;   vpsrlq  %xmm1, $13, %xmm5
+;   vandnpd %xmm5, %xmm3, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -267,10 +267,10 @@ block0(v0: f64x2, v1: f64x2):
 ;   vminpd %xmm1, %xmm0, %xmm3
 ;   vminpd %xmm0, %xmm1, %xmm5
 ;   vorpd %xmm5, %xmm3, %xmm7
-;   vcmpunordpd %xmm5, %xmm3, %xmm9
-;   vorpd %xmm9, %xmm7, %xmm11
-;   vpsrlq $0xd, %xmm9, %xmm13
-;   vandnpd %xmm11, %xmm13, %xmm0
+;   vcmpunordpd %xmm5, %xmm3, %xmm1
+;   vorpd %xmm1, %xmm7, %xmm3
+;   vpsrlq $0xd, %xmm1, %xmm5
+;   vandnpd %xmm3, %xmm5, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -288,11 +288,11 @@ block0(v0: f32x4, v1: f32x4):
 ;   vmaxps  %xmm0, %xmm1, %xmm3
 ;   vmaxps  %xmm1, %xmm0, %xmm5
 ;   vxorps  %xmm3, %xmm5, %xmm7
-;   vorps   %xmm3, %xmm7, %xmm9
-;   vsubps  %xmm9, %xmm7, %xmm11
-;   vcmpps  $3, %xmm9, %xmm9, %xmm13
-;   vpsrld  %xmm13, $10, %xmm15
-;   vandnps %xmm15, %xmm11, %xmm0
+;   vorps   %xmm3, %xmm7, %xmm1
+;   vsubps  %xmm1, %xmm7, %xmm3
+;   vcmpps  $3, %xmm1, %xmm1, %xmm5
+;   vpsrld  %xmm5, $10, %xmm7
+;   vandnps %xmm7, %xmm3, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -305,11 +305,11 @@ block0(v0: f32x4, v1: f32x4):
 ;   vmaxps %xmm1, %xmm0, %xmm3
 ;   vmaxps %xmm0, %xmm1, %xmm5
 ;   vxorps %xmm5, %xmm3, %xmm7
-;   vorps %xmm7, %xmm3, %xmm9
-;   vsubps %xmm7, %xmm9, %xmm11
-;   vcmpunordps %xmm9, %xmm9, %xmm13
-;   vpsrld $0xa, %xmm13, %xmm15
-;   vandnps %xmm11, %xmm15, %xmm0
+;   vorps %xmm7, %xmm3, %xmm1
+;   vsubps %xmm7, %xmm1, %xmm3
+;   vcmpunordps %xmm1, %xmm1, %xmm5
+;   vpsrld $0xa, %xmm5, %xmm7
+;   vandnps %xmm3, %xmm7, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -327,11 +327,11 @@ block0(v0: f64x2, v1: f64x2):
 ;   vmaxpd  %xmm0, %xmm1, %xmm3
 ;   vmaxpd  %xmm1, %xmm0, %xmm5
 ;   vxorpd  %xmm3, %xmm5, %xmm7
-;   vorpd   %xmm3, %xmm7, %xmm9
-;   vsubpd  %xmm9, %xmm7, %xmm11
-;   vcmppd  $3, %xmm9, %xmm9, %xmm13
-;   vpsrlq  %xmm13, $13, %xmm15
-;   vandnpd %xmm15, %xmm11, %xmm0
+;   vorpd   %xmm3, %xmm7, %xmm1
+;   vsubpd  %xmm1, %xmm7, %xmm3
+;   vcmppd  $3, %xmm1, %xmm1, %xmm5
+;   vpsrlq  %xmm5, $13, %xmm7
+;   vandnpd %xmm7, %xmm3, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -344,11 +344,11 @@ block0(v0: f64x2, v1: f64x2):
 ;   vmaxpd %xmm1, %xmm0, %xmm3
 ;   vmaxpd %xmm0, %xmm1, %xmm5
 ;   vxorpd %xmm5, %xmm3, %xmm7
-;   vorpd %xmm7, %xmm3, %xmm9
-;   vsubpd %xmm7, %xmm9, %xmm11
-;   vcmpunordpd %xmm9, %xmm9, %xmm13
-;   vpsrlq $0xd, %xmm13, %xmm15
-;   vandnpd %xmm11, %xmm15, %xmm0
+;   vorpd %xmm7, %xmm3, %xmm1
+;   vsubpd %xmm7, %xmm1, %xmm3
+;   vcmpunordpd %xmm1, %xmm1, %xmm5
+;   vpsrlq $0xd, %xmm5, %xmm7
+;   vandnpd %xmm3, %xmm7, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
diff --git a/cranelift/filetests/filetests/isa/x64/simd-comparison-legalize.clif b/cranelift/filetests/filetests/isa/x64/simd-comparison-legalize.clif
index 805258582eb4..26f10fdcca1a 100644
--- a/cranelift/filetests/filetests/isa/x64/simd-comparison-legalize.clif
+++ b/cranelift/filetests/filetests/isa/x64/simd-comparison-legalize.clif
@@ -43,9 +43,9 @@ block0(v0: i32x4, v1: i32x4):
 ; block0:
 ;   pmaxud  %xmm0, %xmm1, %xmm0
 ;   pcmpeqd %xmm0, %xmm1, %xmm0
-;   uninit  %xmm8
-;   pcmpeqd %xmm8, %xmm8, %xmm8
-;   pxor    %xmm0, %xmm8, %xmm0
+;   uninit  %xmm1
+;   pcmpeqd %xmm1, %xmm1, %xmm1
+;   pxor    %xmm0, %xmm1, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -57,8 +57,8 @@ block0(v0: i32x4, v1: i32x4):
 ; block1: ; offset 0x4
 ;   pmaxud %xmm1, %xmm0
 ;   pcmpeqd %xmm1, %xmm0
-;   pcmpeqd %xmm8, %xmm8
-;   pxor %xmm8, %xmm0
+;   pcmpeqd %xmm1, %xmm1
+;   pxor %xmm1, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
diff --git a/cranelift/filetests/filetests/isa/x64/simd-splat-avx.clif b/cranelift/filetests/filetests/isa/x64/simd-splat-avx.clif
index c7a3a92af894..9e994d56593a 100644
--- a/cranelift/filetests/filetests/isa/x64/simd-splat-avx.clif
+++ b/cranelift/filetests/filetests/isa/x64/simd-splat-avx.clif
@@ -178,8 +178,8 @@ block0(v0: i64):
 ;   uninit  %xmm2
 ;   vpinsrb $0, %xmm2, 0(%rdi), %xmm4
 ;   uninit  %xmm6
-;   vpxor   %xmm6, %xmm6, %xmm8
-;   vpshufb %xmm4, %xmm8, %xmm0
+;   vpxor   %xmm6, %xmm6, %xmm0
+;   vpshufb %xmm4, %xmm0, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -190,8 +190,8 @@ block0(v0: i64):
 ;   movq %rsp, %rbp
 ; block1: ; offset 0x4
 ;   vpinsrb $0, (%rdi), %xmm2, %xmm4 ; trap: heap_oob
-;   vpxor %xmm6, %xmm6, %xmm8
-;   vpshufb %xmm8, %xmm4, %xmm0
+;   vpxor %xmm6, %xmm6, %xmm0
+;   vpshufb %xmm0, %xmm4, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
diff --git a/cranelift/filetests/filetests/isa/x64/simd-widen-mul.clif b/cranelift/filetests/filetests/isa/x64/simd-widen-mul.clif
index 4941dda84a70..4890d8bc4f75 100644
--- a/cranelift/filetests/filetests/isa/x64/simd-widen-mul.clif
+++ b/cranelift/filetests/filetests/isa/x64/simd-widen-mul.clif
@@ -18,8 +18,8 @@ block0(v0: i8x16, v1: i8x16):
 ;   pmovsxbw %xmm6, %xmm0
 ;   movdqa  %xmm1, %xmm6
 ;   palignr $8, %xmm6, %xmm1, %xmm6
-;   pmovsxbw %xmm6, %xmm8
-;   pmullw  %xmm0, %xmm8, %xmm0
+;   pmovsxbw %xmm6, %xmm1
+;   pmullw  %xmm0, %xmm1, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -34,8 +34,8 @@ block0(v0: i8x16, v1: i8x16):
 ;   pmovsxbw %xmm6, %xmm0
 ;   movdqa %xmm1, %xmm6
 ;   palignr $8, %xmm1, %xmm6
-;   pmovsxbw %xmm6, %xmm8
-;   pmullw %xmm8, %xmm0
+;   pmovsxbw %xmm6, %xmm1
+;   pmullw %xmm1, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
@@ -223,14 +223,14 @@ block0(v0: i8x16, v1: i8x16):
 ;   pushq   %rbp
 ;   movq    %rsp, %rbp
 ; block0:
-;   uninit  %xmm8
-;   pxor    %xmm8, %xmm8, %xmm8
-;   punpckhbw %xmm0, %xmm8, %xmm0
-;   uninit  %xmm8
-;   pxor    %xmm8, %xmm8, %xmm8
-;   movdqa  %xmm1, %xmm11
-;   punpckhbw %xmm11, %xmm8, %xmm11
-;   pmullw  %xmm0, %xmm11, %xmm0
+;   uninit  %xmm2
+;   pxor    %xmm2, %xmm2, %xmm2
+;   punpckhbw %xmm0, %xmm2, %xmm0
+;   uninit  %xmm2
+;   pxor    %xmm2, %xmm2, %xmm2
+;   movdqa  %xmm1, %xmm3
+;   punpckhbw %xmm3, %xmm2, %xmm3
+;   pmullw  %xmm0, %xmm3, %xmm0
 ;   movq    %rbp, %rsp
 ;   popq    %rbp
 ;   ret
@@ -240,12 +240,12 @@ block0(v0: i8x16, v1: i8x16):
 ;   pushq %rbp
 ;   movq %rsp, %rbp
 ; block1: ; offset 0x4
-;   pxor %xmm8, %xmm8
-;   punpckhbw %xmm8, %xmm0
-;   pxor %xmm8, %xmm8
-;   movdqa %xmm1, %xmm11
-;   punpckhbw %xmm8, %xmm11
-;   pmullw %xmm11, %xmm0
+;   pxor %xmm2, %xmm2
+;   punpckhbw %xmm2, %xmm0
+;   pxor %xmm2, %xmm2
+;   movdqa %xmm1, %xmm3
+;   punpckhbw %xmm2, %xmm3
+;   pmullw %xmm3, %xmm0
 ;   movq %rbp, %rsp
 ;   popq %rbp
 ;   retq
diff --git a/cranelift/filetests/filetests/wasm/x64-relaxed-simd-deterministic.wat b/cranelift/filetests/filetests/wasm/x64-relaxed-simd-deterministic.wat
index 3d5a9fb3363b..381a74fe34c8 100644
--- a/cranelift/filetests/filetests/wasm/x64-relaxed-simd-deterministic.wat
+++ b/cranelift/filetests/filetests/wasm/x64-relaxed-simd-deterministic.wat
@@ -47,10 +47,10 @@
 ;;   vcmpps  $0, %xmm0, %xmm0, %xmm3
 ;;   vandps  %xmm0, %xmm3, %xmm5
 ;;   vpxor   %xmm3, %xmm5, %xmm7
-;;   vcvttps2dq %xmm5, %xmm9
-;;   vpand   %xmm9, %xmm7, %xmm11
-;;   vpsrad  %xmm11, $31, %xmm13
-;;   vpxor   %xmm13, %xmm9, %xmm0
+;;   vcvttps2dq %xmm5, %xmm1
+;;   vpand   %xmm1, %xmm7, %xmm3
+;;   vpsrad  %xmm3, $31, %xmm5
+;;   vpxor   %xmm5, %xmm1, %xmm0
 ;;   jmp     label1
 ;; block1:
 ;;   movq    %rbp, %rsp
@@ -65,19 +65,19 @@
 ;; block0:
 ;;   uninit  %xmm3
 ;;   vxorps  %xmm3, %xmm3, %xmm5
-;;   vmaxps  %xmm0, %xmm5, %xmm7
-;;   vpcmpeqd %xmm5, %xmm5, %xmm9
-;;   vpsrld  %xmm9, $1, %xmm11
-;;   vcvtdq2ps %xmm11, %xmm13
-;;   vcvttps2dq %xmm7, %xmm15
-;;   vsubps  %xmm7, %xmm13, %xmm1
-;;   vcmpps  $2, %xmm13, %xmm1, %xmm3
+;;   vmaxps  %xmm0, %xmm5, %xmm0
+;;   vpcmpeqd %xmm5, %xmm5, %xmm1
+;;   vpsrld  %xmm1, $1, %xmm3
+;;   vcvtdq2ps %xmm3, %xmm5
+;;   vcvttps2dq %xmm0, %xmm7
+;;   vsubps  %xmm0, %xmm5, %xmm1
+;;   vcmpps  $2, %xmm5, %xmm1, %xmm3
 ;;   vcvttps2dq %xmm1, %xmm5
-;;   vpxor   %xmm5, %xmm3, %xmm7
-;;   uninit  %xmm9
-;;   vpxor   %xmm9, %xmm9, %xmm11
-;;   vpmaxsd %xmm7, %xmm11, %xmm13
-;;   vpaddd  %xmm13, %xmm15, %xmm0
+;;   vpxor   %xmm5, %xmm3, %xmm0
+;;   uninit  %xmm1
+;;   vpxor   %xmm1, %xmm1, %xmm3
+;;   vpmaxsd %xmm0, %xmm3, %xmm5
+;;   vpaddd  %xmm5, %xmm7, %xmm0
 ;;   jmp     label1
 ;; block1:
 ;;   movq    %rbp, %rsp
@@ -109,10 +109,10 @@
 ;;   uninit  %xmm3
 ;;   vxorpd  %xmm3, %xmm3, %xmm5
 ;;   vmaxpd  %xmm0, %xmm5, %xmm7
-;;   vminpd  %xmm7, const(0), %xmm9
-;;   vroundpd $3, %xmm9, %xmm11
-;;   vaddpd  %xmm11, const(1), %xmm13
-;;   vshufps $136, %xmm13, %xmm5, %xmm0
+;;   vminpd  %xmm7, const(0), %xmm1
+;;   vroundpd $3, %xmm1, %xmm3
+;;   vaddpd  %xmm3, const(1), %xmm6
+;;   vshufps $136, %xmm6, %xmm5, %xmm0
 ;;   jmp     label1
 ;; block1:
 ;;   movq    %rbp, %rsp
@@ -125,15 +125,15 @@
 ;;   movq    %rsp, %rbp
 ;;   unwind DefineNewFrame { offset_upward_to_caller_sp: 16, offset_downward_to_clobbers: 0 }
 ;; block0:
-;;   vpmovsxbw %xmm0, %xmm12
-;;   vpmovsxbw %xmm1, %xmm13
-;;   vpmullw %xmm12, %xmm13, %xmm12
-;;   vpalignr $8, %xmm0, %xmm0, %xmm11
-;;   vpmovsxbw %xmm11, %xmm13
-;;   vpalignr $8, %xmm1, %xmm1, %xmm11
-;;   vpmovsxbw %xmm11, %xmm14
-;;   vpmullw %xmm13, %xmm14, %xmm13
-;;   vphaddw %xmm12, %xmm13, %xmm0
+;;   vpmovsxbw %xmm0, %xmm4
+;;   vpmovsxbw %xmm1, %xmm5
+;;   vpmullw %xmm4, %xmm5, %xmm4
+;;   vpalignr $8, %xmm0, %xmm0, %xmm3
+;;   vpmovsxbw %xmm3, %xmm5
+;;   vpalignr $8, %xmm1, %xmm1, %xmm3
+;;   vpmovsxbw %xmm3, %xmm6
+;;   vpmullw %xmm5, %xmm6, %xmm5
+;;   vphaddw %xmm4, %xmm5, %xmm0
 ;;   jmp     label1
 ;; block1:
 ;;   movq    %rbp, %rsp
@@ -146,17 +146,17 @@
 ;;   movq    %rsp, %rbp
 ;;   unwind DefineNewFrame { offset_upward_to_caller_sp: 16, offset_downward_to_clobbers: 0 }
 ;; block0:
-;;   vpmovsxbw %xmm0, %xmm15
+;;   vpmovsxbw %xmm0, %xmm7
 ;;   vpmovsxbw %xmm1, %xmm3
-;;   vpmullw %xmm15, %xmm3, %xmm15
-;;   vpalignr $8, %xmm0, %xmm0, %xmm14
-;;   vpmovsxbw %xmm14, %xmm0
-;;   vpalignr $8, %xmm1, %xmm1, %xmm14
-;;   vpmovsxbw %xmm14, %xmm1
+;;   vpmullw %xmm7, %xmm3, %xmm7
+;;   vpalignr $8, %xmm0, %xmm0, %xmm6
+;;   vpmovsxbw %xmm6, %xmm0
+;;   vpalignr $8, %xmm1, %xmm1, %xmm6
+;;   vpmovsxbw %xmm6, %xmm1
 ;;   vpmullw %xmm0, %xmm1, %xmm0
-;;   vphaddw %xmm15, %xmm0, %xmm15
-;;   vpmaddwd %xmm15, const(0), %xmm15
-;;   vpaddd  %xmm15, %xmm2, %xmm0
+;;   vphaddw %xmm7, %xmm0, %xmm7
+;;   vpmaddwd %xmm7, const(0), %xmm7
+;;   vpaddd  %xmm7, %xmm2, %xmm0
 ;;   jmp     label1
 ;; block1:
 ;;   movq    %rbp, %rsp
diff --git a/cranelift/filetests/filetests/wasm/x64-relaxed-simd.wat b/cranelift/filetests/filetests/wasm/x64-relaxed-simd.wat
index 39ebc2324b95..d3bbfd4441e2 100644
--- a/cranelift/filetests/filetests/wasm/x64-relaxed-simd.wat
+++ b/cranelift/filetests/filetests/wasm/x64-relaxed-simd.wat
@@ -58,20 +58,20 @@
 ;; block0:
 ;;   uninit  %xmm7
 ;;   xorps   %xmm7, %xmm7, %xmm7
-;;   movdqa  %xmm0, %xmm12
-;;   maxps   %xmm12, %xmm7, %xmm12
+;;   movdqa  %xmm0, %xmm4
+;;   maxps   %xmm4, %xmm7, %xmm4
 ;;   pcmpeqd %xmm7, %xmm7, %xmm7
 ;;   psrld   %xmm7, $1, %xmm7
 ;;   cvtdq2ps %xmm7, %xmm1
-;;   cvttps2dq %xmm12, %xmm15
-;;   subps   %xmm12, %xmm1, %xmm12
-;;   cmpps   $2, %xmm1, %xmm12, %xmm1
-;;   cvttps2dq %xmm12, %xmm0
+;;   cvttps2dq %xmm4, %xmm7
+;;   subps   %xmm4, %xmm1, %xmm4
+;;   cmpps   $2, %xmm1, %xmm4, %xmm1
+;;   cvttps2dq %xmm4, %xmm0
 ;;   pxor    %xmm0, %xmm1, %xmm0
-;;   uninit  %xmm10
-;;   pxor    %xmm10, %xmm10, %xmm10
-;;   pmaxsd  %xmm0, %xmm10, %xmm0
-;;   paddd   %xmm0, %xmm15, %xmm0
+;;   uninit  %xmm2
+;;   pxor    %xmm2, %xmm2, %xmm2
+;;   pmaxsd  %xmm0, %xmm2, %xmm0
+;;   paddd   %xmm0, %xmm7, %xmm0
 ;;   jmp     label1
 ;; block1:
 ;;   movq    %rbp, %rsp
@@ -99,10 +99,9 @@
 ;; block0:
 ;;   uninit  %xmm4
 ;;   xorpd   %xmm4, %xmm4, %xmm4
-;;   movdqa  %xmm0, %xmm8
-;;   maxpd   %xmm8, %xmm4, %xmm8
-;;   minpd   %xmm8, const(0), %xmm8
-;;   roundpd $3, %xmm8, %xmm0
+;;   maxpd   %xmm0, %xmm4, %xmm0
+;;   minpd   %xmm0, const(0), %xmm0
+;;   roundpd $3, %xmm0, %xmm0
 ;;   addpd   %xmm0, const(1), %xmm0
 ;;   shufps  $136, %xmm0, %xmm4, %xmm0
 ;;   jmp     label1
@@ -132,9 +131,10 @@
 ;;   movq    %rsp, %rbp
 ;;   unwind DefineNewFrame { offset_upward_to_caller_sp: 16, offset_downward_to_clobbers: 0 }
 ;; block0:
-;;   movdqa  %xmm0, %xmm8
-;;   movdqa  %xmm1, %xmm0
-;;   pmaddubsw %xmm0, %xmm8, %xmm0
+;;   movdqa  %xmm1, %xmm3
+;;   movdqa  %xmm0, %xmm1
+;;   movdqa  %xmm3, %xmm0
+;;   pmaddubsw %xmm0, %xmm1, %xmm0
 ;;   pmaddwd %xmm0, const(0), %xmm0
 ;;   paddd   %xmm0, %xmm2, %xmm0
 ;;   jmp     label1