How oneAPI vectorize the given code?

Roin — Fri, 04 Mar 2022 22:01:19 GMT

Hi,

I am working on DPC++ on CPU backend.

I have the following function:

void Evaluate_Kernel(Creature *creatures, double *fitness_function, uint32_t count, uint32_t num_vars, sycl::nd_item<3> item_ct1) { double fitness = 0; for (int j = 0; j < num_vars; j++) { double pow = 127; for (int k = 0; k < j + 1; k++) { pow *= 7.3; printf("add\n"); } fitness += pow; } printf("fitness: %f\n", fitness); }

And when I see the dumped code, it has:

0000000000407a60 <_Z15Evaluate_KernelP8CreaturePdjjN2cl4sycl7nd_itemILi3EEE>: 407a60: 55 push %rbp 407a61: 48 89 e5 mov %rsp,%rbp 407a64: 48 83 ec 40 sub $0x40,%rsp 407a68: 48 89 7d f8 mov %rdi,-0x8(%rbp) 407a6c: 48 89 75 f0 mov %rsi,-0x10(%rbp) 407a70: 89 55 ec mov %edx,-0x14(%rbp) 407a73: 89 4d e8 mov %ecx,-0x18(%rbp) 407a76: 0f 57 c0 xorps %xmm0,%xmm0 407a79: f2 0f 11 45 e0 movsd %xmm0,-0x20(%rbp) 407a7e: c7 45 dc 00 00 00 00 movl $0x0,-0x24(%rbp) 407a85: 8b 45 dc mov -0x24(%rbp),%eax 407a88: 3b 45 e8 cmp -0x18(%rbp),%eax 407a8b: 0f 83 73 00 00 00 jae 407b04 <_Z15Evaluate_KernelP8CreaturePdjjN2cl4sycl7nd_itemILi3EEE+0xa4> 407a91: f2 0f 10 05 87 16 02 movsd 0x21687(%rip),%xmm0 # 429120 <_ZTI5Timer+0x10> 407a98: 00 407a99: f2 0f 11 45 d0 movsd %xmm0,-0x30(%rbp) 407a9e: c7 45 cc 00 00 00 00 movl $0x0,-0x34(%rbp) 407aa5: 8b 45 cc mov -0x34(%rbp),%eax 407aa8: 8b 4d dc mov -0x24(%rbp),%ecx 407aab: 83 c1 01 add $0x1,%ecx 407aae: 39 c8 cmp %ecx,%eax 407ab0: 0f 8d 31 00 00 00 jge 407ae7 <_Z15Evaluate_KernelP8CreaturePdjjN2cl4sycl7nd_itemILi3EEE+0x87> 407ab6: f2 0f 10 05 6a 16 02 movsd 0x2166a(%rip),%xmm0 # 429128 <_ZTI5Timer+0x18> 407abd: 00 407abe: f2 0f 59 45 d0 mulsd -0x30(%rbp),%xmm0 407ac3: f2 0f 11 45 d0 movsd %xmm0,-0x30(%rbp) 407ac8: 48 bf 70 95 42 00 00 movabs $0x429570,%rdi 407acf: 00 00 00 407ad2: b0 00 mov $0x0,%al 407ad4: e8 97 c5 ff ff callq 404070 <printf@plt> 407ad9: 8b 45 cc mov -0x34(%rbp),%eax 407adc: 83 c0 01 add $0x1,%eax 407adf: 89 45 cc mov %eax,-0x34(%rbp) 407ae2: e9 be ff ff ff jmpq 407aa5 <_Z15Evaluate_KernelP8CreaturePdjjN2cl4sycl7nd_itemILi3EEE+0x45> 407ae7: f2 0f 10 45 d0 movsd -0x30(%rbp),%xmm0 407aec: f2 0f 58 45 e0 addsd -0x20(%rbp),%xmm0 407af1: f2 0f 11 45 e0 movsd %xmm0,-0x20(%rbp) 407af6: 8b 45 dc mov -0x24(%rbp),%eax 407af9: 83 c0 01 add $0x1,%eax 407afc: 89 45 dc mov %eax,-0x24(%rbp) 407aff: e9 81 ff ff ff jmpq 407a85 <_Z15Evaluate_KernelP8CreaturePdjjN2cl4sycl7nd_itemILi3EEE+0x25> 407b04: f2 0f 10 45 e0 movsd -0x20(%rbp),%xmm0 407b09: 48 bf 75 95 42 00 00 movabs $0x429575,%rdi 407b10: 00 00 00 407b13: b0 01 mov $0x1,%al 407b15: e8 56 c5 ff ff callq 404070 <printf@plt> 407b1a: 48 83 c4 40 add $0x40,%rsp 407b1e: 5d pop %rbp 407b1f: c3 retq

I am quite confused as I cannot understand what the optimized instructions do. Would anyone please give me some hints of how to understand the optimized code?

Thanks

topic How oneAPI vectorize the given code? in Intel® oneAPI DPC++/C++ Compiler

How oneAPI vectorize the given code?