diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/mul-i512.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/mul-i512.ll | 2176 |
1 files changed, 892 insertions, 1284 deletions
diff --git a/llvm/test/CodeGen/X86/mul-i512.ll b/llvm/test/CodeGen/X86/mul-i512.ll index a0ed13517bb..40f6b09288e 100644 --- a/llvm/test/CodeGen/X86/mul-i512.ll +++ b/llvm/test/CodeGen/X86/mul-i512.ll @@ -9,453 +9,263 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X32-NEXT: pushl %ebx ; X32-NEXT: pushl %edi ; X32-NEXT: pushl %esi -; X32-NEXT: subl $244, %esp -; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X32-NEXT: movl 20(%ecx), %edi -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl 16(%ecx), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ecx, %ebp -; X32-NEXT: xorl %ebx, %ebx -; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: movl %eax, %esi -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ebx -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl %ecx, %edi -; X32-NEXT: movl %edx, %ebx -; X32-NEXT: adcl $0, %ebx -; X32-NEXT: addl %esi, %edi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ecx, %ebx -; X32-NEXT: movl %ecx, %edi +; X32-NEXT: subl $180, %esp +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: movl {{[0-9]+}}(%esp), %edx +; X32-NEXT: movl 28(%edx), %ecx ; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: setb %cl -; X32-NEXT: addl %eax, %ebx -; X32-NEXT: movzbl %cl, %ecx -; X32-NEXT: adcl %edx, %ecx -; X32-NEXT: movl 24(%ebp), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: xorl %edx, %edx -; X32-NEXT: mull %edx -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: addl %eax, %esi -; X32-NEXT: adcl %edx, %edi -; X32-NEXT: addl %ebx, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ecx, %edi +; X32-NEXT: movl 24(%edx), %ebp +; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl (%eax), %edi ; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X32-NEXT: movl (%ecx), %eax +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, %esi ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: xorl %ebx, %ebx -; X32-NEXT: mull %ebx -; X32-NEXT: movl %eax, %ebp +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, %edi -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl 4(%ecx), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ecx, %esi -; X32-NEXT: mull %ebx ; X32-NEXT: movl %eax, %ebx -; X32-NEXT: addl %edi, %ebx -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: addl %ebp, %ebx -; X32-NEXT: movl %ebx, (%esp) # 4-byte Spill -; X32-NEXT: adcl %edi, %ecx -; X32-NEXT: setb %bl -; X32-NEXT: addl %eax, %ecx -; X32-NEXT: movzbl %bl, %ebx -; X32-NEXT: adcl %edx, %ebx -; X32-NEXT: movl 8(%esi), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: xorl %edx, %edx -; X32-NEXT: mull %edx -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebp, %esi -; X32-NEXT: addl %eax, %esi -; X32-NEXT: movl %edi, %eax -; X32-NEXT: adcl %edx, %eax -; X32-NEXT: addl %ecx, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ebx, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: addl %ebp, %eax -; X32-NEXT: movl %ebp, %esi -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %edi, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X32-NEXT: movl (%ecx), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: xorl %ebp, %ebp -; X32-NEXT: mull %ebp -; X32-NEXT: movl %edx, %ebx -; X32-NEXT: movl %eax, %ecx -; X32-NEXT: movl %eax, %edx -; X32-NEXT: addl %esi, %edx -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: adcl %edi, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: addl %esi, %ebx +; X32-NEXT: adcl $0, %edi ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 16(%eax), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: mull %ebp -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ecx, %edi +; X32-NEXT: movl 4(%eax), %ecx +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: mull %ecx ; X32-NEXT: movl %ecx, %ebp -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: addl %eax, %edi -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: adcl %edx, %eax +; X32-NEXT: movl %edx, %esi +; X32-NEXT: addl %ebx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl %edi, %esi +; X32-NEXT: setb %cl ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: mull %ebp ; X32-NEXT: addl %esi, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl (%esp), %eax # 4-byte Folded Reload -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 4(%eax), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: xorl %edi, %edi -; X32-NEXT: mull %edi -; X32-NEXT: movl %eax, %esi -; X32-NEXT: addl %ebx, %esi -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: addl %ebp, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ebx, %ecx -; X32-NEXT: movl %ebx, %esi -; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: setb %bl -; X32-NEXT: addl %eax, %ecx -; X32-NEXT: movzbl %bl, %ebx -; X32-NEXT: adcl %edx, %ebx -; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 8(%eax), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: mull %edi -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movzbl %cl, %eax +; X32-NEXT: adcl %eax, %edx ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebp, %edi -; X32-NEXT: addl %eax, %edi -; X32-NEXT: adcl %edx, %esi -; X32-NEXT: addl %ecx, %edi -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ebx, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %eax, (%esp) # 4-byte Folded Spill -; X32-NEXT: adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[0-9]+}}(%esp), %esi -; X32-NEXT: movl 20(%esi), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: xorl %ecx, %ecx -; X32-NEXT: mull %ecx -; X32-NEXT: movl %eax, %ebx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: addl %edi, %ebx -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: addl %ebp, %ebx +; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X32-NEXT: movl 16(%ecx), %ebx +; X32-NEXT: movl %ebx, %eax ; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edi, %ecx -; X32-NEXT: setb %bl -; X32-NEXT: addl %eax, %ecx -; X32-NEXT: movzbl %bl, %ebx -; X32-NEXT: adcl %edx, %ebx -; X32-NEXT: movl 24(%esi), %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: xorl %edx, %edx -; X32-NEXT: mull %edx -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebp, %esi -; X32-NEXT: addl %eax, %esi -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: adcl %edx, %edi -; X32-NEXT: addl %ecx, %esi -; X32-NEXT: adcl %ebx, %edi -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: addl %ebp, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebp, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: adcl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: addl %edx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: addl %edx, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload ; X32-NEXT: mull %edi ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edx, %ebp -; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 28(%eax), %esi -; X32-NEXT: movl %esi, %eax -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl 20(%ecx), %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, %edi -; X32-NEXT: movl %eax, %ebx -; X32-NEXT: addl %ebp, %ebx +; X32-NEXT: movl %eax, %ecx +; X32-NEXT: addl %esi, %ecx ; X32-NEXT: adcl $0, %edi -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: mull %ecx +; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl %ebp, %esi +; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: mull %ebp ; X32-NEXT: movl %edx, %ebp -; X32-NEXT: addl %ebx, %eax +; X32-NEXT: addl %ecx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl %edi, %ebp -; X32-NEXT: setb %bl -; X32-NEXT: movl %esi, %eax -; X32-NEXT: mull %ecx -; X32-NEXT: addl %ebp, %eax -; X32-NEXT: movzbl %bl, %ecx -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: addl %eax, %ecx -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edx, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: setb %cl +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload +; X32-NEXT: movl %ebx, %eax ; X32-NEXT: mull %esi +; X32-NEXT: movl %eax, %esi +; X32-NEXT: addl %ebp, %esi +; X32-NEXT: movzbl %cl, %eax +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: movl 8(%eax), %ebp +; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload ; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %ebx -; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: adcl $0, %ebx -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi +; X32-NEXT: mull %ebp ; X32-NEXT: movl %edx, %ecx -; X32-NEXT: addl %ebp, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ebx, %ecx -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl %edi, %ebp -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %edi +; X32-NEXT: movl %ebx, %eax +; X32-NEXT: mull %ebp +; X32-NEXT: movl %edx, %ebp ; X32-NEXT: movl %eax, %ebx ; X32-NEXT: addl %ecx, %ebx -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %edi -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl %ecx, %ebp -; X32-NEXT: adcl $0, %esi +; X32-NEXT: adcl $0, %ebp ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax ; X32-NEXT: movl 12(%eax), %ecx -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: movl %edi, %eax ; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: addl %ebp, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %esi, %ecx +; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl %edx, %edi +; X32-NEXT: addl %ebx, %eax +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: adcl %ebp, %edi ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload -; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %ebp -; X32-NEXT: adcl %edx, %ecx -; X32-NEXT: addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, %ebp +; X32-NEXT: mull %ecx +; X32-NEXT: movl %eax, %ecx +; X32-NEXT: addl %edi, %ecx +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl $0, %ecx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload +; X32-NEXT: adcl $0, %edx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: movl %ebx, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi +; X32-NEXT: movl %esi, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, %ebp +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: adcl $0, %ebp +; X32-NEXT: movl %esi, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, %esi +; X32-NEXT: addl %ebx, %eax +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: adcl %ebp, %esi +; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: mull %edi +; X32-NEXT: addl %esi, %eax +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload +; X32-NEXT: adcl %esi, %edx +; X32-NEXT: addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload +; X32-NEXT: adcl %ecx, %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl $0, %edx ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl {{[0-9]+}}(%esp), %esi +; X32-NEXT: movl %esi, %ecx +; X32-NEXT: movl 8(%esi), %ebx ; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %ebx -; X32-NEXT: addl %edi, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: movl %edx, %ebp +; X32-NEXT: movl 12(%ecx), %ecx +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: mull %esi -; X32-NEXT: addl %ebx, %eax -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload -; X32-NEXT: adcl %esi, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %eax, %edi +; X32-NEXT: addl %ebp, %edi +; X32-NEXT: adcl $0, %esi +; X32-NEXT: movl %ebx, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %edi -; X32-NEXT: adcl %edx, %ebx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload -; X32-NEXT: addl %ebp, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: adcl %ecx, %esi +; X32-NEXT: mull %ebx +; X32-NEXT: movl %edx, %ebp +; X32-NEXT: addl %edi, %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl %esi, %ebp +; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: mull %ebx +; X32-NEXT: addl %ebp, %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %edi -; X32-NEXT: adcl $0, %ebx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: adcl %eax, %edx ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload +; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X32-NEXT: movl (%ecx), %esi +; X32-NEXT: movl %esi, %eax ; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload ; X32-NEXT: mull %edi ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 12(%eax), %eax +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl 4(%ecx), %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, %edi ; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl %esi, %ebp +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload ; X32-NEXT: adcl $0, %edi -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload +; X32-NEXT: movl %esi, %eax ; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %ebx, %esi +; X32-NEXT: movl %edx, %ecx ; X32-NEXT: addl %ebp, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edi, %esi -; X32-NEXT: setb %cl -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %ebx -; X32-NEXT: addl %esi, %eax -; X32-NEXT: movzbl %cl, %ecx -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: addl %eax, %ecx -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edx, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: movl %esi, %eax +; X32-NEXT: adcl %edi, %ecx +; X32-NEXT: setb %bl +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: mull %esi +; X32-NEXT: movl %eax, %esi +; X32-NEXT: addl %ecx, %esi +; X32-NEXT: movzbl %bl, %eax +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: mull %ecx +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, %ebp +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: adcl $0, %ebp +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: mull %ecx ; X32-NEXT: movl %edx, %edi -; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: adcl $0, %edi -; X32-NEXT: movl %esi, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: addl %ebp, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edi, %ecx +; X32-NEXT: addl %ebx, %eax +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: adcl %ebp, %edi ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl %ebx, %edi -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: mull %esi +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: mull %ecx ; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl %ecx, %ebp +; X32-NEXT: addl %edi, %ebp ; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload ; X32-NEXT: adcl %eax, %edx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, %ebp +; X32-NEXT: adcl $0, %edx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: movl %esi, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %ecx +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: mull %ecx +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ebx +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: mull %ecx ; X32-NEXT: movl %edx, %edi ; X32-NEXT: movl %eax, %ebx -; X32-NEXT: addl %ecx, %ebx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload ; X32-NEXT: adcl $0, %edi ; X32-NEXT: movl %esi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload @@ -469,112 +279,32 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X32-NEXT: mull %ecx ; X32-NEXT: addl %esi, %eax ; X32-NEXT: movzbl %bl, %ecx -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: movl %edx, %esi +; X32-NEXT: adcl %ecx, %esi ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %ebx -; X32-NEXT: adcl %edx, %ecx -; X32-NEXT: addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, %ebx -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: adcl $0, %edx -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %ebp -; X32-NEXT: addl %edi, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: addl %ebp, %eax -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload -; X32-NEXT: adcl %esi, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: addl %eax, %ebp -; X32-NEXT: adcl %edx, %edi +; X32-NEXT: addl %ebp, %ecx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload -; X32-NEXT: addl %ebx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: adcl %ecx, %esi -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %ebp -; X32-NEXT: adcl $0, %edi -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload -; X32-NEXT: adcl (%esp), %esi # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 1-byte Folded Reload +; X32-NEXT: adcl %edi, %eax +; X32-NEXT: adcl $0, %esi +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload +; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload ; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, (%esp) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %edi -; X32-NEXT: movl %eax, %ecx -; X32-NEXT: addl %esi, %ecx -; X32-NEXT: adcl $0, %edi -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %esi -; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edi, %esi -; X32-NEXT: setb %cl -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: mull %ebx -; X32-NEXT: addl %esi, %eax -; X32-NEXT: movzbl %cl, %ecx -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: addl %eax, %ecx -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edx, %esi +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: movl 16(%eax), %esi ; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload ; X32-NEXT: movl %ebx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: mull %esi ; X32-NEXT: movl %edx, %ecx ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill @@ -585,9 +315,12 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X32-NEXT: movl %eax, %edi ; X32-NEXT: addl %ecx, %edi ; X32-NEXT: adcl $0, %esi +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: movl 20(%eax), %ecx ; X32-NEXT: movl %ebx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: mull %ebx +; X32-NEXT: mull %ecx +; X32-NEXT: movl %ecx, %ebx +; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %edx, %ecx ; X32-NEXT: addl %edi, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill @@ -595,272 +328,267 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill ; X32-NEXT: movl %ebp, %eax ; X32-NEXT: mull %ebx -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl %ecx, %edi +; X32-NEXT: addl %ecx, %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload ; X32-NEXT: adcl %eax, %edx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload -; X32-NEXT: addl (%esp), %edi # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, %ecx +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: movl %edi, %eax +; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: addl %ecx, %ebx +; X32-NEXT: adcl $0, %esi +; X32-NEXT: movl %ebp, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, (%esp) # 4-byte Spill +; X32-NEXT: movl %edx, %ebp +; X32-NEXT: addl %ebx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebp, %eax +; X32-NEXT: adcl %esi, %ebp +; X32-NEXT: setb %bl +; X32-NEXT: movl %edi, %eax ; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %esi ; X32-NEXT: movl %eax, %ecx -; X32-NEXT: addl (%esp), %ecx # 4-byte Folded Reload -; X32-NEXT: adcl $0, %esi +; X32-NEXT: addl %ebp, %ecx +; X32-NEXT: movzbl %bl, %eax +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 28(%eax), %ebp -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: mull %ebp -; X32-NEXT: movl %ebp, (%esp) # 4-byte Spill -; X32-NEXT: movl %edx, %ebx -; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %esi, %ebx -; X32-NEXT: setb %cl -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %ebp -; X32-NEXT: addl %ebx, %eax -; X32-NEXT: movzbl %cl, %ecx -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %ebp -; X32-NEXT: adcl %edx, %ecx -; X32-NEXT: addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, %ebp -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl 24(%eax), %ebx +; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: movl %esi, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: mull %edi +; X32-NEXT: mull %ebx ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl %edi, %eax +; X32-NEXT: mull %ebx +; X32-NEXT: movl %edx, %ebp +; X32-NEXT: movl %eax, %edi +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: adcl $0, %ebp +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: movl 28(%eax), %ebx +; X32-NEXT: movl %esi, %eax +; X32-NEXT: mull %ebx +; X32-NEXT: movl %edx, %esi +; X32-NEXT: addl %edi, %eax +; X32-NEXT: movl %eax, %edi +; X32-NEXT: adcl %ebp, %esi +; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %edi -; X32-NEXT: movl %edx, %edi +; X32-NEXT: mull %ebx +; X32-NEXT: movl %ebx, %ebp ; X32-NEXT: movl %eax, %ebx +; X32-NEXT: addl %esi, %ebx +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, %ebx +; X32-NEXT: adcl $0, %edx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: adcl $0, %edi -; X32-NEXT: movl %esi, %eax -; X32-NEXT: movl (%esp), %esi # 4-byte Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: movl %edx, (%esp) # 4-byte Spill +; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: mull %esi -; X32-NEXT: addl %ebx, %eax +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edi, %edx -; X32-NEXT: movl %edx, %edi -; X32-NEXT: setb %bl ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %eax, %edi +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: adcl $0, %esi +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl %ebp, %ecx +; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: mull %ebp +; X32-NEXT: movl %edx, %ebp ; X32-NEXT: addl %edi, %eax -; X32-NEXT: movzbl %bl, %esi -; X32-NEXT: adcl %esi, %edx +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl %esi, %ebp +; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: mull %ecx +; X32-NEXT: addl %ebp, %eax +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload +; X32-NEXT: adcl %ecx, %edx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: addl %ebx, %edi ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %edi -; X32-NEXT: adcl %edx, %ebx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload -; X32-NEXT: addl %ebp, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: adcl %ecx, %esi -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %edi -; X32-NEXT: adcl $0, %ebx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl (%esp), %ebx # 4-byte Folded Reload +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload +; X32-NEXT: adcl %ecx, %eax +; X32-NEXT: adcl $0, %edx +; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: movl %edi, %edx ; X32-NEXT: adcl $0, %edx +; X32-NEXT: movl %ebx, %ecx +; X32-NEXT: adcl $0, %ecx +; X32-NEXT: adcl $0, %eax ; X32-NEXT: adcl $0, %esi -; X32-NEXT: adcl $0, %edi -; X32-NEXT: adcl $0, %ebx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload +; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload ; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: movl %esi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: mull %edi -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: mull %edi -; X32-NEXT: movl %edx, %ebx -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl %ecx, %edi -; X32-NEXT: adcl $0, %ebx -; X32-NEXT: movl %esi, %eax +; X32-NEXT: movl %edi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: mull %esi ; X32-NEXT: movl %edx, %ecx -; X32-NEXT: addl %edi, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ebx, %ecx -; X32-NEXT: setb %bl +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload ; X32-NEXT: movl %ebp, %eax ; X32-NEXT: mull %esi -; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movzbl %bl, %ecx -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edx, %ecx -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: addl %ecx, %ebx +; X32-NEXT: adcl $0, %esi +; X32-NEXT: movl %edi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload ; X32-NEXT: mull %edi -; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %edx, %ecx +; X32-NEXT: addl %ebx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload +; X32-NEXT: adcl %esi, %ecx +; X32-NEXT: setb %bl ; X32-NEXT: movl %ebp, %eax ; X32-NEXT: mull %edi -; X32-NEXT: movl %edx, %edi -; X32-NEXT: movl %eax, %ebx -; X32-NEXT: addl %esi, %ebx -; X32-NEXT: adcl $0, %edi -; X32-NEXT: movl %ecx, %eax +; X32-NEXT: addl %ecx, %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movzbl %bl, %eax +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: movl %edi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: mull %esi ; X32-NEXT: movl %edx, %ecx -; X32-NEXT: addl %ebx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edi, %ecx -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload ; X32-NEXT: movl %ebp, %eax -; X32-NEXT: movl %ebp, %edi ; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, %esi ; X32-NEXT: movl %eax, %ebx ; X32-NEXT: addl %ecx, %ebx -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload +; X32-NEXT: adcl $0, %esi +; X32-NEXT: movl %edi, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: mull %ecx +; X32-NEXT: movl %edx, %edi +; X32-NEXT: addl %ebx, %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl %esi, %edi +; X32-NEXT: setb %bl +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: mull %ecx +; X32-NEXT: movl %eax, %esi +; X32-NEXT: addl %edi, %esi +; X32-NEXT: movzbl %bl, %eax ; X32-NEXT: adcl %eax, %edx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: movl %ebp, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, (%esp) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %edi -; X32-NEXT: movl %eax, %ecx -; X32-NEXT: addl %esi, %ecx -; X32-NEXT: adcl $0, %edi ; X32-NEXT: movl %ebp, %eax -; X32-NEXT: movl (%esp), %esi # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %ebp -; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edi, %ebp -; X32-NEXT: setb %cl -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %esi +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, %ebx +; X32-NEXT: movl %eax, %ebp +; X32-NEXT: addl (%esp), %ebp # 4-byte Folded Reload +; X32-NEXT: adcl $0, %ebx +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, %ecx ; X32-NEXT: addl %ebp, %eax -; X32-NEXT: movzbl %cl, %ecx -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %esi -; X32-NEXT: adcl %edx, %ecx -; X32-NEXT: addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: movl %eax, %ebp +; X32-NEXT: adcl %ebx, %ecx +; X32-NEXT: setb (%esp) # 1-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, %esi -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: mull %edi +; X32-NEXT: movl %edi, %ebx +; X32-NEXT: movl %eax, %edi +; X32-NEXT: addl %ecx, %edi +; X32-NEXT: movzbl (%esp), %eax # 1-byte Folded Reload +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload +; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, %edi +; X32-NEXT: adcl $0, %edx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: movl %edx, (%esp) # 4-byte Spill ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: movl %ebp, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %ecx +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: mull %esi ; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl %ecx, %edi +; X32-NEXT: movl %eax, %ebp +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload ; X32-NEXT: adcl $0, %esi -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: movl (%esp), %ebx # 4-byte Reload +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl %ebx, %ecx ; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %ecx +; X32-NEXT: movl %edx, %ebx +; X32-NEXT: addl %ebp, %eax ; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl %edi, %ebp -; X32-NEXT: adcl %esi, %ecx +; X32-NEXT: adcl %esi, %ebx ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %ebx -; X32-NEXT: addl %ecx, %eax +; X32-NEXT: mull %ecx +; X32-NEXT: addl %ebx, %eax ; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload ; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %esi -; X32-NEXT: adcl %edx, %ecx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %esi -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload +; X32-NEXT: addl %edi, %ecx +; X32-NEXT: adcl (%esp), %ebp # 4-byte Folded Reload +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload +; X32-NEXT: adcl %esi, %eax +; X32-NEXT: movl %eax, %esi +; X32-NEXT: adcl $0, %edx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload @@ -870,232 +598,193 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %edx -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl %eax, %ecx +; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl $0, %ebp ; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl $0, %esi ; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X32-NEXT: movl 32(%ecx), %edi +; X32-NEXT: adcl $0, %edx +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: movl 32(%eax), %edi ; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, %eax ; X32-NEXT: mull %edi -; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %edx, %ecx ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload ; X32-NEXT: movl %ebp, %eax ; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, %edi ; X32-NEXT: movl %eax, %ebx -; X32-NEXT: addl %esi, %ebx +; X32-NEXT: addl %ecx, %ebx ; X32-NEXT: adcl $0, %edi ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 36(%eax), %esi -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: mull %esi -; X32-NEXT: movl %esi, %ecx -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl 36(%eax), %ecx +; X32-NEXT: movl %esi, %eax +; X32-NEXT: movl %ecx, %esi +; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: mull %ecx +; X32-NEXT: movl %edx, %ecx ; X32-NEXT: addl %ebx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %edi, %esi +; X32-NEXT: adcl %edi, %ecx ; X32-NEXT: setb %bl ; X32-NEXT: movl %ebp, %eax -; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %ecx -; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl %esi, %ebp -; X32-NEXT: movzbl %bl, %eax -; X32-NEXT: adcl %eax, %ecx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: xorl %edx, %edx -; X32-NEXT: mull %edx +; X32-NEXT: mull %esi +; X32-NEXT: addl %ecx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movzbl %bl, %eax +; X32-NEXT: adcl %eax, %edx ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: addl %eax, %edi ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: adcl %edx, %esi -; X32-NEXT: addl %ebp, %edi -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ecx, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: movl %ebx, %esi -; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %edi -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl %edi, %ebp -; X32-NEXT: adcl $0, %esi -; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl %esi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload ; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, %ecx -; X32-NEXT: addl %ebp, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %esi, %ecx -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload ; X32-NEXT: movl %ebx, %eax ; X32-NEXT: mull %edi -; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: movl %edx, %edi +; X32-NEXT: movl %eax, %ebp +; X32-NEXT: addl %ecx, %ebp +; X32-NEXT: adcl $0, %edi +; X32-NEXT: movl %esi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %edi -; X32-NEXT: adcl %edx, %ecx +; X32-NEXT: mull %ecx +; X32-NEXT: movl %edx, %esi +; X32-NEXT: addl %ebp, %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl %edi, %esi +; X32-NEXT: setb (%esp) # 1-byte Folded Spill +; X32-NEXT: movl %ebx, %eax +; X32-NEXT: mull %ecx +; X32-NEXT: movl %eax, %edi +; X32-NEXT: addl %esi, %edi +; X32-NEXT: movzbl (%esp), %eax # 1-byte Folded Reload +; X32-NEXT: adcl %eax, %edx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 40(%eax), %ebp -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ebp -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl 40(%eax), %esi +; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, (%esp) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %ebx, %eax -; X32-NEXT: mull %ebp -; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, %ecx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: adcl $0, %esi +; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, %ebp +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: addl (%esp), %ebx # 4-byte Folded Reload +; X32-NEXT: adcl $0, %ebp ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax -; X32-NEXT: movl 44(%eax), %ebx -; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %edi -; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %esi, %edi +; X32-NEXT: movl 44(%eax), %esi +; X32-NEXT: movl %esi, (%esp) # 4-byte Spill +; X32-NEXT: movl %ecx, %eax +; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, %ecx +; X32-NEXT: addl %ebx, %eax +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: adcl %ebp, %ecx ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %ecx +; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, %ebp ; X32-NEXT: movl %eax, %esi -; X32-NEXT: addl %edi, %esi +; X32-NEXT: addl %ecx, %esi ; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %ecx -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: xorl %edx, %edx -; X32-NEXT: mull %edx -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: addl %eax, %edi -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %edx, %eax -; X32-NEXT: addl %esi, %edi -; X32-NEXT: adcl %ecx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, %edi -; X32-NEXT: adcl $0, %eax -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl %eax, %ebp +; X32-NEXT: addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, %esi +; X32-NEXT: adcl $0, %ebp +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: movl %ecx, %eax -; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: mull %ebp +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ebp -; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, %edi ; X32-NEXT: movl %eax, %ebx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: adcl $0, %esi +; X32-NEXT: adcl $0, %edi ; X32-NEXT: movl %ecx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: mull %ebp +; X32-NEXT: mull (%esp) # 4-byte Folded Reload ; X32-NEXT: movl %edx, %ecx ; X32-NEXT: addl %ebx, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %esi, %ecx -; X32-NEXT: setb %bl -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ebp +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: adcl %edi, %ecx +; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: movl (%esp), %edi # 4-byte Reload +; X32-NEXT: mull %edi ; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movzbl %bl, %ecx +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload ; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %esi -; X32-NEXT: adcl %edx, %ecx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl %ebp, %ebx +; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload +; X32-NEXT: adcl %ecx, %eax +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, %edx +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %esi -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl $0, %ecx +; X32-NEXT: movl %edi, %esi +; X32-NEXT: imull %eax, %esi +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: mull %ecx +; X32-NEXT: movl %eax, (%esp) # 4-byte Spill +; X32-NEXT: addl %esi, %edx +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload +; X32-NEXT: imull %ebx, %ecx +; X32-NEXT: addl %edx, %ecx ; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: imull %eax, %ebp -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: addl %ebp, %edx -; X32-NEXT: imull {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: addl %edx, %esi -; X32-NEXT: movl %esi, %ebp -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: movl %eax, %esi -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: imull %ebx, %esi ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: mull %edi +; X32-NEXT: imull %edi, %esi +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload +; X32-NEXT: mull %ebp ; X32-NEXT: movl %eax, %ecx ; X32-NEXT: addl %esi, %edx -; X32-NEXT: movl (%esp), %esi # 4-byte Reload -; X32-NEXT: movl %edi, %eax -; X32-NEXT: imull %edi, %esi +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: imull %ebp, %esi ; X32-NEXT: addl %edx, %esi -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: movl %ecx, %edi -; X32-NEXT: adcl %ebp, %esi -; X32-NEXT: movl %esi, (%esp) # 4-byte Spill +; X32-NEXT: addl (%esp), %ecx # 4-byte Folded Reload +; X32-NEXT: movl %ecx, (%esp) # 4-byte Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload +; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: mull %ecx ; X32-NEXT: movl %edx, %ebp ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl %edi, %eax ; X32-NEXT: mull %ecx ; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, %ebx -; X32-NEXT: addl %ebp, %ebx +; X32-NEXT: movl %eax, %edi +; X32-NEXT: addl %ebp, %edi ; X32-NEXT: adcl $0, %esi ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: mull %ebp +; X32-NEXT: mull %ebx +; X32-NEXT: movl %ebx, %ebp ; X32-NEXT: movl %edx, %ecx -; X32-NEXT: addl %ebx, %eax +; X32-NEXT: addl %edi, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl %esi, %ecx ; X32-NEXT: setb %bl @@ -1104,351 +793,312 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X32-NEXT: addl %ecx, %eax ; X32-NEXT: movzbl %bl, %ecx ; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: addl %edi, %eax +; X32-NEXT: addl (%esp), %eax # 4-byte Folded Reload ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl (%esp), %edx # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[0-9]+}}(%esp), %edi ; X32-NEXT: movl 60(%edi), %ecx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: imull %eax, %ecx ; X32-NEXT: movl 56(%edi), %esi +; X32-NEXT: movl %edi, %ebx ; X32-NEXT: mull %esi ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: addl %ecx, %edx ; X32-NEXT: imull {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload ; X32-NEXT: addl %edx, %esi -; X32-NEXT: movl 48(%edi), %ebx -; X32-NEXT: movl 52(%edi), %ebp +; X32-NEXT: movl 48(%edi), %edi +; X32-NEXT: movl 52(%ebx), %ebp ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: movl %eax, %edi -; X32-NEXT: imull %ebp, %edi +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: imull %ebp, %ebx ; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: mull %ebx -; X32-NEXT: addl %edi, %edx +; X32-NEXT: mull %edi +; X32-NEXT: addl %ebx, %edx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: imull %ebx, %ecx +; X32-NEXT: imull %edi, %ecx ; X32-NEXT: addl %edx, %ecx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl %esi, %ecx ; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl %edi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: mull %ecx ; X32-NEXT: movl %edx, %esi ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %ebp, %eax ; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %ebp -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl %esi, %edi -; X32-NEXT: adcl $0, %ebp -; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl %edx, %ebx +; X32-NEXT: movl %eax, %ebp +; X32-NEXT: addl %esi, %ebp +; X32-NEXT: adcl $0, %ebx +; X32-NEXT: movl %edi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: mull %esi ; X32-NEXT: movl %edx, %ecx -; X32-NEXT: movl %eax, %ebx -; X32-NEXT: addl %edi, %ebx -; X32-NEXT: adcl %ebp, %ecx -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill +; X32-NEXT: movl %eax, %edi +; X32-NEXT: addl %ebp, %edi +; X32-NEXT: adcl %ebx, %ecx +; X32-NEXT: setb %bl ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: mull %esi ; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload +; X32-NEXT: movzbl %bl, %ecx ; X32-NEXT: adcl %ecx, %edx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload ; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[0-9]+}}(%esp), %esi -; X32-NEXT: movl %esi, %ecx -; X32-NEXT: movl 40(%esi), %ebx -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi +; X32-NEXT: movl {{[0-9]+}}(%esp), %edi +; X32-NEXT: movl 40(%edi), %ebp +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: mull %ecx ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %edx, %ebp -; X32-NEXT: movl 44(%ecx), %ecx -; X32-NEXT: movl %ecx, %eax -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: mull %esi ; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl %ebp, %edi -; X32-NEXT: adcl $0, %esi -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: mull %ebp +; X32-NEXT: movl 44(%edi), %edi +; X32-NEXT: movl %edi, %eax +; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: mull %ecx ; X32-NEXT: movl %edx, %ebx -; X32-NEXT: addl %edi, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %esi, %ebx -; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl %eax, %ecx +; X32-NEXT: addl %esi, %ecx +; X32-NEXT: adcl $0, %ebx +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload ; X32-NEXT: mull %ebp ; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl %ebx, %edi -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %esi -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: xorl %ecx, %ecx -; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload -; X32-NEXT: addl %edi, %eax -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %esi, %edx -; X32-NEXT: movl %edx, (%esp) # 4-byte Spill -; X32-NEXT: movl {{[0-9]+}}(%esp), %esi -; X32-NEXT: movl 32(%esi), %edi +; X32-NEXT: addl %ecx, %eax +; X32-NEXT: movl %eax, (%esp) # 4-byte Spill +; X32-NEXT: adcl %ebx, %esi +; X32-NEXT: setb %cl ; X32-NEXT: movl %edi, %eax -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: mull %ecx +; X32-NEXT: mull %ebp +; X32-NEXT: addl %esi, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movzbl %cl, %eax +; X32-NEXT: adcl %eax, %edx ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl 36(%esi), %esi +; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X32-NEXT: movl 32(%ecx), %esi ; X32-NEXT: movl %esi, %eax ; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: mull %ecx +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload +; X32-NEXT: mull %edi +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl 36(%ecx), %ebp +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, %ebx -; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload +; X32-NEXT: movl %eax, %ecx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload ; X32-NEXT: adcl $0, %ebx -; X32-NEXT: movl %edi, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: mull %ecx +; X32-NEXT: movl %esi, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: mull %esi ; X32-NEXT: movl %edx, %edi -; X32-NEXT: addl %ebp, %eax +; X32-NEXT: addl %ecx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl %ebx, %edi -; X32-NEXT: setb %bl -; X32-NEXT: movl %esi, %eax -; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %ebp -; X32-NEXT: movl %eax, %esi -; X32-NEXT: addl %edi, %esi -; X32-NEXT: movzbl %bl, %eax -; X32-NEXT: adcl %eax, %ebp -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, %eax -; X32-NEXT: xorl %ecx, %ecx -; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: addl %eax, %ecx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: adcl %edx, %eax -; X32-NEXT: addl %esi, %ecx -; X32-NEXT: adcl %ebp, %eax -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: setb %cl +; X32-NEXT: movl %ebp, %eax +; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: mull %esi +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: addl %edi, %ebx +; X32-NEXT: movzbl %cl, %eax +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload +; X32-NEXT: adcl (%esp), %edx # 4-byte Folded Reload +; X32-NEXT: movl %edx, (%esp) # 4-byte Spill ; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl $0, (%esp) # 4-byte Folded Spill -; X32-NEXT: movl %edi, %eax +; X32-NEXT: adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload +; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %ecx +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl %ebp, %eax ; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %edx, %edi ; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl %ecx, %ebp -; X32-NEXT: adcl $0, %esi -; X32-NEXT: movl %edi, %eax +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload +; X32-NEXT: adcl $0, %edi +; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: mull %ecx -; X32-NEXT: movl %edx, %edi +; X32-NEXT: movl %edx, %esi ; X32-NEXT: addl %ebp, %eax ; X32-NEXT: movl %eax, %ebp -; X32-NEXT: adcl %esi, %edi +; X32-NEXT: adcl %edi, %esi ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: mull %ecx -; X32-NEXT: movl %ecx, %ebx -; X32-NEXT: addl %edi, %eax -; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %esi -; X32-NEXT: adcl %edx, %ecx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload -; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl $0, %esi +; X32-NEXT: mull %ecx +; X32-NEXT: movl %edx, %ecx +; X32-NEXT: movl %eax, %edi +; X32-NEXT: addl %esi, %edi +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload +; X32-NEXT: adcl %eax, %ecx +; X32-NEXT: addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl (%esp), %ebp # 4-byte Folded Reload +; X32-NEXT: movl %ebp, (%esp) # 4-byte Spill +; X32-NEXT: adcl $0, %edi ; X32-NEXT: adcl $0, %ecx -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload -; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl (%esp), %ecx # 4-byte Folded Reload -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload ; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, (%esp) # 4-byte Spill +; X32-NEXT: movl %esi, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload +; X32-NEXT: mull %ebx +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %esi -; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, %ebp -; X32-NEXT: addl (%esp), %ebp # 4-byte Folded Reload -; X32-NEXT: adcl $0, %esi -; X32-NEXT: movl %ecx, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %ecx +; X32-NEXT: movl %edx, %ebx +; X32-NEXT: movl %eax, %ebp +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload +; X32-NEXT: adcl $0, %ebx +; X32-NEXT: movl %esi, %eax +; X32-NEXT: mull {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload +; X32-NEXT: movl %edx, %esi ; X32-NEXT: addl %ebp, %eax ; X32-NEXT: movl %eax, %ebp -; X32-NEXT: adcl %esi, %ecx -; X32-NEXT: setb (%esp) # 1-byte Folded Spill -; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ebx -; X32-NEXT: addl %ecx, %eax -; X32-NEXT: movzbl (%esp), %ecx # 1-byte Folded Reload -; X32-NEXT: adcl %ecx, %edx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: addl %eax, %edi -; X32-NEXT: adcl %edx, %ecx +; X32-NEXT: adcl %ebx, %esi +; X32-NEXT: setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload -; X32-NEXT: addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload +; X32-NEXT: mull %ebx +; X32-NEXT: addl %esi, %eax +; X32-NEXT: movl %eax, %esi +; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload +; X32-NEXT: adcl %eax, %edx +; X32-NEXT: addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill +; X32-NEXT: adcl %ecx, %ebp ; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload -; X32-NEXT: adcl %eax, %edi -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl $0, %ecx -; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl %eax, %esi +; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl $0, %edx +; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X32-NEXT: movl 48(%ecx), %ebp -; X32-NEXT: imull %ebp, %ebx -; X32-NEXT: movl %ebp, %eax -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: mull %edi +; X32-NEXT: movl 48(%ecx), %edi +; X32-NEXT: imull %edi, %ebx +; X32-NEXT: movl %edi, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: mull %esi ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: addl %ebx, %edx ; X32-NEXT: movl 52(%ecx), %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: imull %eax, %edi -; X32-NEXT: addl %edx, %edi -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: imull %eax, %esi +; X32-NEXT: addl %edx, %esi +; X32-NEXT: movl %esi, %ebp ; X32-NEXT: movl 56(%ecx), %eax -; X32-NEXT: movl %ecx, %ebx -; X32-NEXT: movl %eax, %edi -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload -; X32-NEXT: imull %esi, %edi +; X32-NEXT: movl %eax, %esi +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload +; X32-NEXT: imull %ebx, %esi ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: mull %ecx +; X32-NEXT: addl %esi, %edx +; X32-NEXT: movl {{[0-9]+}}(%esp), %esi +; X32-NEXT: movl 60(%esi), %esi +; X32-NEXT: imull %ecx, %esi +; X32-NEXT: addl %edx, %esi +; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: addl %edi, %edx -; X32-NEXT: movl 60(%ebx), %ebx +; X32-NEXT: adcl %ebp, %esi ; X32-NEXT: movl %ecx, %eax -; X32-NEXT: imull %ecx, %ebx -; X32-NEXT: addl %edx, %ebx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload -; X32-NEXT: mull %ebp -; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: mull %edi +; X32-NEXT: movl %edx, %ebp ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %esi, %eax -; X32-NEXT: mull %ebp +; X32-NEXT: movl %ebx, %eax +; X32-NEXT: mull %edi ; X32-NEXT: movl %edx, %ecx -; X32-NEXT: movl %eax, %edi -; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload +; X32-NEXT: movl %eax, %ebx +; X32-NEXT: addl %ebp, %ebx ; X32-NEXT: adcl $0, %ecx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload ; X32-NEXT: mull %ebp -; X32-NEXT: movl %edx, %esi -; X32-NEXT: addl %edi, %eax +; X32-NEXT: movl %edx, %edi +; X32-NEXT: addl %ebx, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ecx, %esi +; X32-NEXT: adcl %ecx, %edi ; X32-NEXT: setb %cl ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: mull %ebp -; X32-NEXT: addl %esi, %eax +; X32-NEXT: addl %edi, %eax ; X32-NEXT: movzbl %cl, %ecx ; X32-NEXT: adcl %ecx, %edx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl %ebx, %edx +; X32-NEXT: adcl %esi, %edx ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload -; X32-NEXT: imull %ebp, %edi -; X32-NEXT: movl %ebp, %eax +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: imull %esi, %edi +; X32-NEXT: movl %esi, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: mull %ecx -; X32-NEXT: movl %eax, %esi +; X32-NEXT: movl %eax, %ebp ; X32-NEXT: addl %edi, %edx ; X32-NEXT: imull {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload ; X32-NEXT: addl %edx, %ecx ; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: movl %eax, %ecx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload -; X32-NEXT: imull %ebx, %ecx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: mull %edi +; X32-NEXT: imull %edi, %ecx +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload +; X32-NEXT: mull %ebx ; X32-NEXT: addl %ecx, %edx ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload -; X32-NEXT: imull %edi, %ecx +; X32-NEXT: imull %ebx, %ecx ; X32-NEXT: addl %edx, %ecx -; X32-NEXT: addl %esi, %eax +; X32-NEXT: addl %ebp, %eax ; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload ; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: movl %ebx, %eax +; X32-NEXT: movl %ebx, %ebp +; X32-NEXT: mull %esi +; X32-NEXT: movl %edx, %ebx +; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: movl %edi, %eax -; X32-NEXT: mull %ebp +; X32-NEXT: mull %esi ; X32-NEXT: movl %edx, %esi -; X32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: movl %ebx, %eax -; X32-NEXT: mull %ebp -; X32-NEXT: movl %edx, %edi ; X32-NEXT: movl %eax, %ecx -; X32-NEXT: addl %esi, %ecx -; X32-NEXT: adcl $0, %edi -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload +; X32-NEXT: addl %ebx, %ecx +; X32-NEXT: adcl $0, %esi +; X32-NEXT: movl %ebp, %eax ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload ; X32-NEXT: mull %ebx -; X32-NEXT: movl %edx, %esi +; X32-NEXT: movl %edx, %edi ; X32-NEXT: movl %eax, %ebp ; X32-NEXT: addl %ecx, %ebp -; X32-NEXT: adcl %edi, %esi +; X32-NEXT: adcl %esi, %edi ; X32-NEXT: setb %cl ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload ; X32-NEXT: mull %ebx ; X32-NEXT: movl %edx, %ebx -; X32-NEXT: addl %esi, %eax +; X32-NEXT: addl %edi, %eax ; X32-NEXT: movzbl %cl, %ecx ; X32-NEXT: adcl %ecx, %ebx ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload @@ -1466,12 +1116,12 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload +; X32-NEXT: movl (%esp), %edi # 4-byte Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload ; X32-NEXT: movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill @@ -1481,45 +1131,45 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload ; X32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload ; X32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload -; X32-NEXT: movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload +; X32-NEXT: movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill -; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload +; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload ; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload ; X32-NEXT: adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, (%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 4(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 8(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 12(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 16(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 20(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 24(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 28(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 32(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 36(%ecx) -; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload -; X32-NEXT: movl %edi, 40(%ecx) -; X32-NEXT: movl %esi, 44(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, (%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 4(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 8(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 12(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 16(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 20(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 24(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 28(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 32(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 36(%ecx) +; X32-NEXT: movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload +; X32-NEXT: movl %esi, 40(%ecx) +; X32-NEXT: movl %edi, 44(%ecx) ; X32-NEXT: movl %edx, 48(%ecx) ; X32-NEXT: movl %ebp, 52(%ecx) ; X32-NEXT: movl %eax, 56(%ecx) ; X32-NEXT: movl %ebx, 60(%ecx) -; X32-NEXT: addl $244, %esp +; X32-NEXT: addl $180, %esp ; X32-NEXT: popl %esi ; X32-NEXT: popl %edi ; X32-NEXT: popl %ebx @@ -1536,292 +1186,250 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind { ; X64-NEXT: pushq %rbx ; X64-NEXT: pushq %rax ; X64-NEXT: movq %rdx, (%rsp) # 8-byte Spill -; X64-NEXT: movq 24(%rdi), %r11 -; X64-NEXT: movq 16(%rdi), %r15 +; X64-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: movq (%rdi), %r9 +; X64-NEXT: movq 8(%rdi), %r15 +; X64-NEXT: movq 24(%rdi), %r12 +; X64-NEXT: movq 16(%rdi), %rax ; X64-NEXT: movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: movq (%rsi), %rdx -; X64-NEXT: movq 8(%rsi), %rbp -; X64-NEXT: movq %r15, %rax -; X64-NEXT: movq %rdx, %rsi -; X64-NEXT: mulq %rdx -; X64-NEXT: movq %rdx, %r9 -; X64-NEXT: movq %rax, %r8 -; X64-NEXT: movq %r11, %rax -; X64-NEXT: movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: mulq %rsi -; X64-NEXT: movq %rsi, %r10 -; X64-NEXT: movq %rdx, %rbx +; X64-NEXT: movq (%rsi), %rdi +; X64-NEXT: movq 8(%rsi), %r14 ; X64-NEXT: movq %rax, %rsi -; X64-NEXT: addq %r9, %rsi -; X64-NEXT: adcq $0, %rbx -; X64-NEXT: movq %r15, %rax -; X64-NEXT: movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: mulq %rbp -; X64-NEXT: movq %rdx, %rcx -; X64-NEXT: movq %rax, %r9 -; X64-NEXT: addq %rsi, %r9 -; X64-NEXT: adcq %rbx, %rcx -; X64-NEXT: setb %al -; X64-NEXT: movzbl %al, %ebx -; X64-NEXT: movq %r11, %rax -; X64-NEXT: mulq %rbp -; X64-NEXT: movq %rbp, %r14 -; X64-NEXT: movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: movq %rdx, %rsi -; X64-NEXT: movq %rax, %rbp -; X64-NEXT: addq %rcx, %rbp -; X64-NEXT: adcq %rbx, %rsi -; X64-NEXT: xorl %ecx, %ecx -; X64-NEXT: movq %r10, %rbx -; X64-NEXT: movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: movq %r10, %rax -; X64-NEXT: mulq %rcx -; X64-NEXT: movq %rdx, %r13 -; X64-NEXT: movq %rax, %r10 -; X64-NEXT: movq %r15, %rax -; X64-NEXT: mulq %rcx -; X64-NEXT: movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: movq %rax, %r15 -; X64-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: addq %r10, %r15 -; X64-NEXT: adcq %r13, %rdx -; X64-NEXT: addq %rbp, %r15 -; X64-NEXT: adcq %rsi, %rdx -; X64-NEXT: movq %rdx, %r12 -; X64-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: movq (%rdi), %rcx -; X64-NEXT: movq %rcx, %rax -; X64-NEXT: mulq %rbx ; X64-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: movq %rdx, %r11 -; X64-NEXT: movq 8(%rdi), %rdi -; X64-NEXT: movq %rdi, %rax -; X64-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: mulq %rbx +; X64-NEXT: mulq %rdi +; X64-NEXT: movq %rdx, %rcx +; X64-NEXT: movq %rax, %r11 +; X64-NEXT: movq %r12, %rax +; X64-NEXT: movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: mulq %rdi ; X64-NEXT: movq %rdx, %rbp -; X64-NEXT: movq %rax, %rsi -; X64-NEXT: addq %r11, %rsi +; X64-NEXT: movq %rax, %rbx +; X64-NEXT: addq %rcx, %rbx ; X64-NEXT: adcq $0, %rbp -; X64-NEXT: movq %rcx, %rax +; X64-NEXT: movq %rsi, %rax ; X64-NEXT: mulq %r14 +; X64-NEXT: movq %rdx, %rcx +; X64-NEXT: movq %rax, %r10 +; X64-NEXT: addq %rbx, %r10 +; X64-NEXT: adcq %rbp, %rcx +; X64-NEXT: setb %al +; X64-NEXT: movzbl %al, %esi +; X64-NEXT: movq %r12, %rax +; X64-NEXT: mulq %r14 +; X64-NEXT: movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: movq %rdx, %r8 +; X64-NEXT: movq %rax, %r13 +; X64-NEXT: addq %rcx, %r13 +; X64-NEXT: adcq %rsi, %r8 +; X64-NEXT: movq %r9, %rax +; X64-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: mulq %rdi +; X64-NEXT: movq %rdx, %rcx +; X64-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: movq %r15, %rax +; X64-NEXT: mulq %rdi ; X64-NEXT: movq %rdx, %rbx -; X64-NEXT: addq %rsi, %rax +; X64-NEXT: movq %rax, %rbp +; X64-NEXT: addq %rcx, %rbp +; X64-NEXT: adcq $0, %rbx +; X64-NEXT: movq %r9, %rax +; X64-NEXT: movq %r9, %r12 +; X64-NEXT: movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: mulq %r14 +; X64-NEXT: movq %rdx, %rcx +; X64-NEXT: addq %rbp, %rax ; X64-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: adcq %rbp, %rbx -; X64-NEXT: setb %r11b -; X64-NEXT: movq %rdi, %rax +; X64-NEXT: adcq %rbx, %rcx +; X64-NEXT: setb %sil +; X64-NEXT: movq %r15, %rdi +; X64-NEXT: movq %r15, %rax ; X64-NEXT: mulq %r14 -; X64-NEXT: movq %rdx, %rsi -; X64-NEXT: movq %rax, %rbp -; X64-NEXT: addq %rbx, %rbp -; X64-NEXT: movzbl %r11b, %eax -; X64-NEXT: adcq %rax, %rsi -; X64-NEXT: movq %rcx, %rax -; X64-NEXT: xorl %edx, %edx -; X64-NEXT: mulq %rdx -; X64-NEXT: movq %rdx, %r14 -; X64-NEXT: movq %rax, %r11 -; X64-NEXT: addq %rax, %r10 -; X64-NEXT: adcq %rdx, %r13 -; X64-NEXT: addq %rbp, %r10 -; X64-NEXT: adcq %rsi, %r13 -; X64-NEXT: addq %r8, %r10 -; X64-NEXT: adcq %r9, %r13 -; X64-NEXT: adcq $0, %r15 -; X64-NEXT: adcq $0, %r12 -; X64-NEXT: movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: movq %rdx, %r15 +; X64-NEXT: movq %rax, %rbx +; X64-NEXT: addq %rcx, %rbx +; X64-NEXT: movzbl %sil, %eax +; X64-NEXT: adcq %rax, %r15 +; X64-NEXT: addq %r11, %rbx +; X64-NEXT: adcq %r10, %r15 +; X64-NEXT: adcq $0, %r13 +; X64-NEXT: movq %r8, %r14 +; X64-NEXT: adcq $0, %r14 ; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload ; X64-NEXT: movq 16(%rsi), %r8 -; X64-NEXT: movq %rcx, %rax -; X64-NEXT: movq %rcx, %r9 -; X64-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: movq %r9, %rax ; X64-NEXT: mulq %r8 -; X64-NEXT: movq %rdx, %rdi -; X64-NEXT: movq %rax, %r12 -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload -; X64-NEXT: movq %rcx, %rax +; X64-NEXT: movq %rdx, %r10 +; X64-NEXT: movq %rax, %r9 +; X64-NEXT: movq %rdi, %rax +; X64-NEXT: movq %rdi, %r11 +; X64-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill ; X64-NEXT: mulq %r8 -; X64-NEXT: movq %rdx, %rbp -; X64-NEXT: movq %rax, %rbx -; X64-NEXT: addq %rdi, %rbx -; X64-NEXT: adcq $0, %rbp +; X64-NEXT: movq %rdx, %rcx +; X64-NEXT: movq %rax, %rbp +; X64-NEXT: addq %r10, %rbp +; X64-NEXT: adcq $0, %rcx ; X64-NEXT: movq 24(%rsi), %rdi -; X64-NEXT: movq %r9, %rax +; X64-NEXT: movq %r12, %rax ; X64-NEXT: mulq %rdi ; X64-NEXT: movq %rdx, %rsi -; X64-NEXT: addq %rbx, %rax -; X64-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: adcq %rbp, %rsi -; X64-NEXT: setb %bpl -; X64-NEXT: movq %rcx, %rax -; X64-NEXT: mulq %rdi -; X64-NEXT: movq %rdx, %rbx -; X64-NEXT: movq %rax, %r9 -; X64-NEXT: addq %rsi, %r9 -; X64-NEXT: movzbl %bpl, %eax -; X64-NEXT: adcq %rax, %rbx -; X64-NEXT: movq %r8, %rax -; X64-NEXT: xorl %ecx, %ecx -; X64-NEXT: mulq %rcx -; X64-NEXT: movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: addq %rbp, %rax ; X64-NEXT: movq %rax, %rbp -; X64-NEXT: addq %rax, %r11 -; X64-NEXT: adcq %rdx, %r14 -; X64-NEXT: addq %r9, %r11 -; X64-NEXT: adcq %rbx, %r14 -; X64-NEXT: addq %r10, %r12 -; X64-NEXT: movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: adcq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill +; X64-NEXT: adcq %rcx, %rsi +; X64-NEXT: setb %cl +; X64-NEXT: movq %r11, %rax +; X64-NEXT: mulq %rdi +; X64-NEXT: movq %rdx, %r11 +; X64-NEXT: movq %rax, %r12 +; X64-NEXT: addq %rsi, %r12 +; X64-NEXT: movzbl %cl, %eax +; X64-NEXT: adcq %rax, %r11 +; X64-NEXT: addq %rbx, %r9 +; X64-NEXT: movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: adcq %r15, %rbp +; X64-NEXT: movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: adcq $0, %r12 ; X64-NEXT: adcq $0, %r11 -; X64-NEXT: adcq $0, %r14 -; X64-NEXT: addq %r15, %r11 -; X64-NEXT: adcq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload +; X64-NEXT: addq %r13, %r12 +; X64-NEXT: adcq %r14, %r11 ; X64-NEXT: setb %r9b -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload -; X64-NEXT: movq %rcx, %rax +; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload +; X64-NEXT: movq %rbx, %rax ; X64-NEXT: mulq %r8 -; X64-NEXT: movq %rdx, %r10 -; X64-NEXT: movq %rax, %r12 -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload -; X64-NEXT: movq %r15, %rax +; X64-NEXT: movq %rdx, %rcx +; X64-NEXT: movq %rax, %r14 +; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload +; X64-NEXT: movq %r10, %rax ; X64-NEXT: mulq %r8 ; X64-NEXT: movq %rdx, %rsi -; X64-NEXT: movq %rax, %rbx -; X64-NEXT: addq %r10, %rbx +; X64-NEXT: movq %rax, %rbp +; X64-NEXT: addq %rcx, %rbp ; X64-NEXT: adcq $0, %rsi -; X64-NEXT: movq %rcx, %rax +; X64-NEXT: movq %rbx, %rax ; X64-NEXT: mulq %rdi ; X64-NEXT: movq %rdx, %rcx -; X64-NEXT: addq %rbx, %rax -; X64-NEXT: movq %rax, %rbx +; X64-NEXT: addq %rbp, %rax +; X64-NEXT: movq %rax, %rbp ; X64-NEXT: adcq %rsi, %rcx -; X64-NEXT: setb %sil -; X64-NEXT: movq %r15, %rax +; X64-NEXT: setb %bl +; X64-NEXT: movq %r10, %rax ; X64-NEXT: mulq %rdi ; X64-NEXT: addq %rcx, %rax -; X64-NEXT: movzbl %sil, %ecx +; X64-NEXT: movzbl %bl, %ecx ; X64-NEXT: adcq %rcx, %rdx -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload -; X64-NEXT: addq %rbp, %rsi -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload -; X64-NEXT: adcq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload -; X64-NEXT: addq %rax, %rsi -; X64-NEXT: adcq %rdx, %rcx -; X64-NEXT: addq %r11, %r12 -; X64-NEXT: movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: adcq %r14, %rbx -; X64-NEXT: movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: movzbl %r9b, %eax -; X64-NEXT: adcq %rax, %rsi -; X64-NEXT: movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: adcq $0, %rcx -; X64-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: addq %r12, %r14 +; X64-NEXT: movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: adcq %r11, %rbp +; X64-NEXT: movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: movzbl %r9b, %ecx +; X64-NEXT: adcq %rcx, %rax +; X64-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: adcq $0, %rdx +; X64-NEXT: movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill ; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload -; X64-NEXT: movq 32(%rcx), %rsi -; X64-NEXT: imulq %rsi, %rdi -; X64-NEXT: movq %rsi, %rax +; X64-NEXT: movq 32(%rcx), %r10 +; X64-NEXT: imulq %r10, %rdi +; X64-NEXT: movq %r10, %rax ; X64-NEXT: mulq %r8 -; X64-NEXT: movq %rax, %r10 +; X64-NEXT: movq %rax, %r11 ; X64-NEXT: addq %rdi, %rdx ; X64-NEXT: movq 40(%rcx), %r9 ; X64-NEXT: imulq %r9, %r8 ; X64-NEXT: addq %rdx, %r8 ; X64-NEXT: movq 48(%rcx), %rax -; X64-NEXT: movq %rcx, %rbx ; X64-NEXT: movq %rax, %rdi -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload -; X64-NEXT: imulq %rcx, %rdi -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload -; X64-NEXT: mulq %rbp -; X64-NEXT: movq %rax, %r14 -; X64-NEXT: addq %rdi, %rdx -; X64-NEXT: movq 56(%rbx), %rbx -; X64-NEXT: imulq %rbp, %rbx -; X64-NEXT: addq %rdx, %rbx -; X64-NEXT: addq %r10, %r14 -; X64-NEXT: adcq %r8, %rbx -; X64-NEXT: movq %rbp, %rax -; X64-NEXT: movq %rbp, %r10 +; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload +; X64-NEXT: imulq %rbx, %rdi +; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload ; X64-NEXT: mulq %rsi +; X64-NEXT: movq %rax, %r12 +; X64-NEXT: addq %rdi, %rdx +; X64-NEXT: movq 56(%rcx), %rbp +; X64-NEXT: imulq %rsi, %rbp +; X64-NEXT: addq %rdx, %rbp +; X64-NEXT: addq %r11, %r12 +; X64-NEXT: adcq %r8, %rbp +; X64-NEXT: movq %rsi, %rax +; X64-NEXT: movq %rsi, %rcx +; X64-NEXT: mulq %r10 ; X64-NEXT: movq %rdx, %rdi ; X64-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; X64-NEXT: movq %rcx, %rax -; X64-NEXT: movq %rcx, %r8 -; X64-NEXT: mulq %rsi +; X64-NEXT: movq %rbx, %rax +; X64-NEXT: movq %rbx, %r11 +; X64-NEXT: mulq %r10 ; X64-NEXT: movq %rdx, %rsi -; X64-NEXT: movq %rax, %rbp -; X64-NEXT: addq %rdi, %rbp +; X64-NEXT: movq %rax, %rbx +; X64-NEXT: addq %rdi, %rbx ; X64-NEXT: adcq $0, %rsi -; X64-NEXT: movq %r10, %rax +; X64-NEXT: movq %rcx, %rax ; X64-NEXT: mulq %r9 ; X64-NEXT: movq %rdx, %rdi -; X64-NEXT: movq %rax, %r13 -; X64-NEXT: addq %rbp, %r13 +; X64-NEXT: movq %rax, %r10 +; X64-NEXT: addq %rbx, %r10 ; X64-NEXT: adcq %rsi, %rdi -; X64-NEXT: setb %cl -; X64-NEXT: movq %r8, %rax +; X64-NEXT: setb %bl +; X64-NEXT: movq %r11, %rax ; X64-NEXT: mulq %r9 -; X64-NEXT: movq %rdx, %r11 -; X64-NEXT: movq %rax, %r9 -; X64-NEXT: addq %rdi, %r9 -; X64-NEXT: movzbl %cl, %eax -; X64-NEXT: adcq %rax, %r11 -; X64-NEXT: addq %r14, %r9 -; X64-NEXT: adcq %rbx, %r11 -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload -; X64-NEXT: movq 56(%rbp), %rcx -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload -; X64-NEXT: imulq %r10, %rcx -; X64-NEXT: movq 48(%rbp), %rbx -; X64-NEXT: movq %r10, %rax -; X64-NEXT: mulq %rbx -; X64-NEXT: movq %rax, %rsi -; X64-NEXT: addq %rcx, %rdx -; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload -; X64-NEXT: imulq %r15, %rbx -; X64-NEXT: addq %rdx, %rbx -; X64-NEXT: movq 32(%rbp), %rdi -; X64-NEXT: movq 40(%rbp), %r8 +; X64-NEXT: movq %rdx, %r15 +; X64-NEXT: movq %rax, %r13 +; X64-NEXT: addq %rdi, %r13 +; X64-NEXT: movzbl %bl, %eax +; X64-NEXT: adcq %rax, %r15 +; X64-NEXT: addq %r12, %r13 +; X64-NEXT: adcq %rbp, %r15 +; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload +; X64-NEXT: movq 56(%rdx), %rcx ; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload -; X64-NEXT: movq %rax, %rcx -; X64-NEXT: imulq %r8, %rcx -; X64-NEXT: mulq %rdi -; X64-NEXT: movq %rax, %r14 +; X64-NEXT: imulq %rax, %rcx +; X64-NEXT: movq 48(%rdx), %rbp +; X64-NEXT: movq %rdx, %rsi +; X64-NEXT: movq %rax, %r11 +; X64-NEXT: mulq %rbp +; X64-NEXT: movq %rax, %r12 ; X64-NEXT: addq %rcx, %rdx +; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload +; X64-NEXT: imulq %r8, %rbp +; X64-NEXT: addq %rdx, %rbp +; X64-NEXT: movq 32(%rsi), %rdi +; X64-NEXT: movq 40(%rsi), %rbx ; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload -; X64-NEXT: imulq %rdi, %rax -; X64-NEXT: addq %rdx, %rax -; X64-NEXT: addq %rsi, %r14 -; X64-NEXT: adcq %rbx, %rax -; X64-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill +; X64-NEXT: movq %rax, %rsi +; X64-NEXT: imulq %rbx, %rsi +; X64-NEXT: mulq %rdi +; X64-NEXT: movq %rax, %rcx +; X64-NEXT: addq %rsi, %rdx +; X64-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload +; X64-NEXT: imulq %rdi, %r14 +; X64-NEXT: addq %rdx, %r14 +; X64-NEXT: addq %r12, %rcx +; X64-NEXT: adcq %rbp, %r14 ; X64-NEXT: movq %rdi, %rax -; X64-NEXT: mulq %r10 -; X64-NEXT: movq %rdx, %r12 +; X64-NEXT: mulq %r11 +; X64-NEXT: movq %rdx, %rbp ; X64-NEXT: movq %rax, %rsi -; X64-NEXT: movq %r8, %rax -; X64-NEXT: mulq %r10 -; X64-NEXT: movq %rdx, %rcx -; X64-NEXT: movq %rax, %rbx -; X64-NEXT: addq %r12, %rbx -; X64-NEXT: adcq $0, %rcx +; X64-NEXT: movq %rbx, %rax +; X64-NEXT: mulq %r11 +; X64-NEXT: movq %rdx, %r11 +; X64-NEXT: movq %rax, %r12 +; X64-NEXT: addq %rbp, %r12 +; X64-NEXT: adcq $0, %r11 ; X64-NEXT: movq %rdi, %rax -; X64-NEXT: mulq %r15 +; X64-NEXT: mulq %r8 ; X64-NEXT: movq %rdx, %rbp ; X64-NEXT: movq %rax, %rdi -; X64-NEXT: addq %rbx, %rdi -; X64-NEXT: adcq %rcx, %rbp -; X64-NEXT: setb %cl -; X64-NEXT: movq %r8, %rax -; X64-NEXT: mulq %r15 +; X64-NEXT: addq %r12, %rdi +; X64-NEXT: adcq %r11, %rbp +; X64-NEXT: setb %r9b +; X64-NEXT: movq %rbx, %rax +; X64-NEXT: mulq %r8 ; X64-NEXT: addq %rbp, %rax -; X64-NEXT: movzbl %cl, %ecx -; X64-NEXT: adcq %rcx, %rdx -; X64-NEXT: addq %r14, %rax -; X64-NEXT: adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload +; X64-NEXT: movzbl %r9b, %ebp +; X64-NEXT: adcq %rbp, %rdx +; X64-NEXT: addq %rcx, %rax +; X64-NEXT: adcq %r14, %rdx ; X64-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload -; X64-NEXT: adcq %r13, %rdi -; X64-NEXT: adcq %r9, %rax -; X64-NEXT: adcq %r11, %rdx +; X64-NEXT: adcq %r10, %rdi +; X64-NEXT: adcq %r13, %rax +; X64-NEXT: adcq %r15, %rdx ; X64-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload ; X64-NEXT: adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Folded Reload ; X64-NEXT: adcq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload |
