1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-- -mattr=avx | FileCheck %s --check-prefixes=AVX,AVX1
3 ; RUN: llc < %s -mtriple=x86_64-- -mattr=avx512f,avx512vl | FileCheck %s --check-prefixes=AVX,AVX512
5 ; PR37751 - https://bugs.llvm.org/show_bug.cgi?id=37751
6 ; We can't combine into 'round' instructions because the behavior is different for out-of-range values.
8 declare <8 x i32> @llvm.x86.avx.cvtt.ps2dq.256(<8 x float>)
9 declare <4 x i32> @llvm.x86.avx.cvtt.pd2dq.256(<4 x double>)
11 define <8 x float> @float_to_int_to_float_mem_v8f32(<8 x float>* %p) #0 {
12 ; AVX-LABEL: float_to_int_to_float_mem_v8f32:
14 ; AVX-NEXT: vcvttps2dq (%rdi), %ymm0
15 ; AVX-NEXT: vcvtdq2ps %ymm0, %ymm0
17 %x = load <8 x float>, <8 x float>* %p, align 16
18 %fptosi = tail call <8 x i32> @llvm.x86.avx.cvtt.ps2dq.256(<8 x float> %x)
19 %sitofp = sitofp <8 x i32> %fptosi to <8 x float>
20 ret <8 x float> %sitofp
23 define <8 x float> @float_to_int_to_float_reg_v8f32(<8 x float> %x) #0 {
24 ; AVX-LABEL: float_to_int_to_float_reg_v8f32:
26 ; AVX-NEXT: vcvttps2dq %ymm0, %ymm0
27 ; AVX-NEXT: vcvtdq2ps %ymm0, %ymm0
29 %fptosi = tail call <8 x i32> @llvm.x86.avx.cvtt.ps2dq.256(<8 x float> %x)
30 %sitofp = sitofp <8 x i32> %fptosi to <8 x float>
31 ret <8 x float> %sitofp
34 define <4 x double> @float_to_int_to_float_mem_v4f64(<4 x double>* %p) #0 {
35 ; AVX-LABEL: float_to_int_to_float_mem_v4f64:
37 ; AVX-NEXT: vcvttpd2dqy (%rdi), %xmm0
38 ; AVX-NEXT: vcvtdq2pd %xmm0, %ymm0
40 %x = load <4 x double>, <4 x double>* %p, align 16
41 %fptosi = tail call <4 x i32> @llvm.x86.avx.cvtt.pd2dq.256(<4 x double> %x)
42 %sitofp = sitofp <4 x i32> %fptosi to <4 x double>
43 ret <4 x double> %sitofp
46 define <4 x double> @float_to_int_to_float_reg_v4f64(<4 x double> %x) #0 {
47 ; AVX-LABEL: float_to_int_to_float_reg_v4f64:
49 ; AVX-NEXT: vcvttpd2dq %ymm0, %xmm0
50 ; AVX-NEXT: vcvtdq2pd %xmm0, %ymm0
52 %fptosi = tail call <4 x i32> @llvm.x86.avx.cvtt.pd2dq.256(<4 x double> %x)
53 %sitofp = sitofp <4 x i32> %fptosi to <4 x double>
54 ret <4 x double> %sitofp
57 attributes #0 = { "no-signed-zeros-fp-math"="true" }