1 ; RUN: llc -mtriple=x86_64-linux -mcpu=nehalem < %s | FileCheck %s 2 ; RUN: llc -mtriple=x86_64-win32 -mcpu=nehalem < %s | FileCheck %s 3 ; rdar://7398554 4 5 ; When doing vector gather-scatter index calculation with 32-bit indices, 6 ; bounce the vector off of cache rather than shuffling each individual 7 ; element out of the index vector. 8 9 ; CHECK: andps ([[H:%rdx|%r8]]), %xmm0 10 ; CHECK: movaps %xmm0, {{(-24)?}}(%rsp) 11 ; CHECK: movslq {{(-24)?}}(%rsp), %rax 12 ; CHECK: movsd ([[P:%rdi|%rcx]],%rax,8), %xmm0 13 ; CHECK: movslq {{-20|4}}(%rsp), %rax 14 ; CHECK: movhpd ([[P]],%rax,8), %xmm0 15 ; CHECK: movslq {{-16|8}}(%rsp), %rax 16 ; CHECK: movsd ([[P]],%rax,8), %xmm1 17 ; CHECK: movslq {{-12|12}}(%rsp), %rax 18 ; CHECK: movhpd ([[P]],%rax,8), %xmm1 19 20 define <4 x double> @foo(double* %p, <4 x i32>* %i, <4 x i32>* %h) nounwind { 21 %a = load <4 x i32>* %i 22 %b = load <4 x i32>* %h 23 %j = and <4 x i32> %a, %b 24 %d0 = extractelement <4 x i32> %j, i32 0 25 %d1 = extractelement <4 x i32> %j, i32 1 26 %d2 = extractelement <4 x i32> %j, i32 2 27 %d3 = extractelement <4 x i32> %j, i32 3 28 %q0 = getelementptr double* %p, i32 %d0 29 %q1 = getelementptr double* %p, i32 %d1 30 %q2 = getelementptr double* %p, i32 %d2 31 %q3 = getelementptr double* %p, i32 %d3 32 %r0 = load double* %q0 33 %r1 = load double* %q1 34 %r2 = load double* %q2 35 %r3 = load double* %q3 36 %v0 = insertelement <4 x double> undef, double %r0, i32 0 37 %v1 = insertelement <4 x double> %v0, double %r1, i32 1 38 %v2 = insertelement <4 x double> %v1, double %r2, i32 2 39 %v3 = insertelement <4 x double> %v2, double %r3, i32 3 40 ret <4 x double> %v3 41 } 42
