From 2b94bbb140645a5ec3a11f9cd9749e2fa68cdbd4 Mon Sep 17 00:00:00 2001 From: Craig Topper Date: Tue, 8 Nov 2016 06:58:53 +0000 Subject: [PATCH] [AVX-512] Add an avx512f without avx512vl command line to vec_fp_to_int.ll and regenerate. This will make a change in a future patch easier to see. NFC git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@286216 91177308-0d34-0410-b5e6-96231b3b80d8 --- test/CodeGen/X86/vec_fp_to_int.ll | 583 +++++++++++++++++++++++------- 1 file changed, 454 insertions(+), 129 deletions(-) diff --git a/test/CodeGen/X86/vec_fp_to_int.ll b/test/CodeGen/X86/vec_fp_to_int.ll index f6b8965622eb..b79403f1a6cc 100644 --- a/test/CodeGen/X86/vec_fp_to_int.ll +++ b/test/CodeGen/X86/vec_fp_to_int.ll @@ -1,8 +1,9 @@ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2 -; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512F -; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512DQ +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512F +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512VL +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512VLDQ ; ; 32-bit tests to make sure we're not doing anything stupid. ; RUN: llc < %s -mtriple=i686-unknown-unknown @@ -45,10 +46,20 @@ define <2 x i64> @fptosi_2f64_to_2i64(<2 x double> %a) { ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptosi_2f64_to_2i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttpd2qq %xmm0, %xmm0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptosi_2f64_to_2i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvttsd2si %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0] +; AVX512VL-NEXT: vcvttsd2si %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_2f64_to_2i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttpd2qq %xmm0, %xmm0 +; AVX512VLDQ-NEXT: retq %cvt = fptosi <2 x double> %a to <2 x i64> ret <2 x i64> %cvt } @@ -110,11 +121,23 @@ define <4 x i32> @fptosi_4f64_to_2i32(<2 x double> %a) { ; AVX-NEXT: vzeroupper ; AVX-NEXT: retq ; -; AVX512-LABEL: fptosi_4f64_to_2i32: -; AVX512: # BB#0: -; AVX512-NEXT: # kill: %XMM0 %XMM0 %YMM0 -; AVX512-NEXT: vcvttpd2dq %ymm0, %xmm0 -; AVX512-NEXT: retq +; AVX512F-LABEL: fptosi_4f64_to_2i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512F-NEXT: vcvttpd2dqy %ymm0, %xmm0 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: fptosi_4f64_to_2i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512VL-NEXT: vcvttpd2dq %ymm0, %xmm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_4f64_to_2i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512VLDQ-NEXT: vcvttpd2dq %ymm0, %xmm0 +; AVX512VLDQ-NEXT: retq %ext = shufflevector <2 x double> %a, <2 x double> undef, <4 x i32> %cvt = fptosi <4 x double> %ext to <4 x i32> ret <4 x i32> %cvt @@ -177,7 +200,7 @@ define <4 x i64> @fptosi_4f64_to_4i64(<4 x double> %a) { ; ; AVX512F-LABEL: fptosi_4f64_to_4i64: ; AVX512F: # BB#0: -; AVX512F-NEXT: vextractf32x4 $1, %ymm0, %xmm1 +; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm1 ; AVX512F-NEXT: vcvttsd2si %xmm1, %rax ; AVX512F-NEXT: vmovq %rax, %xmm2 ; AVX512F-NEXT: vpermilpd {{.*#+}} xmm1 = xmm1[1,0] @@ -190,13 +213,31 @@ define <4 x i64> @fptosi_4f64_to_4i64(<4 x double> %a) { ; AVX512F-NEXT: vcvttsd2si %xmm0, %rax ; AVX512F-NEXT: vmovq %rax, %xmm0 ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] -; AVX512F-NEXT: vinserti32x4 $1, %xmm1, %ymm0, %ymm0 +; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptosi_4f64_to_4i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttpd2qq %ymm0, %ymm0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptosi_4f64_to_4i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm1 +; AVX512VL-NEXT: vcvttsd2si %xmm1, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm1 = xmm1[1,0] +; AVX512VL-NEXT: vcvttsd2si %xmm1, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] +; AVX512VL-NEXT: vcvttsd2si %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0] +; AVX512VL-NEXT: vcvttsd2si %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] +; AVX512VL-NEXT: vinserti32x4 $1, %xmm1, %ymm0, %ymm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_4f64_to_4i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttpd2qq %ymm0, %ymm0 +; AVX512VLDQ-NEXT: retq %cvt = fptosi <4 x double> %a to <4 x i64> ret <4 x i64> %cvt } @@ -215,10 +256,20 @@ define <4 x i32> @fptosi_4f64_to_4i32(<4 x double> %a) { ; AVX-NEXT: vzeroupper ; AVX-NEXT: retq ; -; AVX512-LABEL: fptosi_4f64_to_4i32: -; AVX512: # BB#0: -; AVX512-NEXT: vcvttpd2dq %ymm0, %xmm0 -; AVX512-NEXT: retq +; AVX512F-LABEL: fptosi_4f64_to_4i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: vcvttpd2dqy %ymm0, %xmm0 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: fptosi_4f64_to_4i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvttpd2dq %ymm0, %xmm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_4f64_to_4i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttpd2dq %ymm0, %xmm0 +; AVX512VLDQ-NEXT: retq %cvt = fptosi <4 x double> %a to <4 x i32> ret <4 x i32> %cvt } @@ -285,10 +336,20 @@ define <2 x i64> @fptoui_2f64_to_2i64(<2 x double> %a) { ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptoui_2f64_to_2i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttpd2uqq %xmm0, %xmm0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptoui_2f64_to_2i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0] +; AVX512VL-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_2f64_to_2i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttpd2uqq %xmm0, %xmm0 +; AVX512VLDQ-NEXT: retq %cvt = fptoui <2 x double> %a to <2 x i64> ret <2 x i64> %cvt } @@ -356,12 +417,24 @@ define <4 x i32> @fptoui_2f64_to_4i32(<2 x double> %a) { ; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptoui_2f64_to_4i32: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttpd2uqq %xmm0, %xmm0 -; AVX512DQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] -; AVX512DQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptoui_2f64_to_4i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0] +; AVX512VL-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] +; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_2f64_to_4i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttpd2uqq %xmm0, %xmm0 +; AVX512VLDQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero +; AVX512VLDQ-NEXT: retq %cvt = fptoui <2 x double> %a to <2 x i32> %ext = shufflevector <2 x i32> %cvt, <2 x i32> zeroinitializer, <4 x i32> ret <4 x i32> %ext @@ -427,11 +500,22 @@ define <4 x i32> @fptoui_2f64_to_2i32(<2 x double> %a) { ; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptoui_2f64_to_2i32: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttpd2uqq %xmm0, %xmm0 -; AVX512DQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptoui_2f64_to_2i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0] +; AVX512VL-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] +; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_2f64_to_2i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttpd2uqq %xmm0, %xmm0 +; AVX512VLDQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512VLDQ-NEXT: retq %cvt = fptoui <2 x double> %a to <2 x i32> %ext = shufflevector <2 x i32> %cvt, <2 x i32> undef, <4 x i32> ret <4 x i32> %ext @@ -483,11 +567,24 @@ define <4 x i32> @fptoui_4f64_to_2i32(<2 x double> %a) { ; AVX-NEXT: vpinsrd $3, %eax, %xmm0, %xmm0 ; AVX-NEXT: retq ; -; AVX512-LABEL: fptoui_4f64_to_2i32: -; AVX512: # BB#0: -; AVX512-NEXT: # kill: %XMM0 %XMM0 %YMM0 -; AVX512-NEXT: vcvttpd2udq %ymm0, %xmm0 -; AVX512-NEXT: retq +; AVX512F-LABEL: fptoui_4f64_to_2i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: # kill: %XMM0 %XMM0 %ZMM0 +; AVX512F-NEXT: vcvttpd2udq %zmm0, %ymm0 +; AVX512F-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: fptoui_4f64_to_2i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512VL-NEXT: vcvttpd2udq %ymm0, %xmm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_4f64_to_2i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512VLDQ-NEXT: vcvttpd2udq %ymm0, %xmm0 +; AVX512VLDQ-NEXT: retq %ext = shufflevector <2 x double> %a, <2 x double> undef, <4 x i32> %cvt = fptoui <4 x double> %ext to <4 x i32> ret <4 x i32> %cvt @@ -619,7 +716,7 @@ define <4 x i64> @fptoui_4f64_to_4i64(<4 x double> %a) { ; ; AVX512F-LABEL: fptoui_4f64_to_4i64: ; AVX512F: # BB#0: -; AVX512F-NEXT: vextractf32x4 $1, %ymm0, %xmm1 +; AVX512F-NEXT: vextractf128 $1, %ymm0, %xmm1 ; AVX512F-NEXT: vcvttsd2usi %xmm1, %rax ; AVX512F-NEXT: vmovq %rax, %xmm2 ; AVX512F-NEXT: vpermilpd {{.*#+}} xmm1 = xmm1[1,0] @@ -632,13 +729,31 @@ define <4 x i64> @fptoui_4f64_to_4i64(<4 x double> %a) { ; AVX512F-NEXT: vcvttsd2usi %xmm0, %rax ; AVX512F-NEXT: vmovq %rax, %xmm0 ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] -; AVX512F-NEXT: vinserti32x4 $1, %xmm1, %ymm0, %ymm0 +; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptoui_4f64_to_4i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttpd2uqq %ymm0, %ymm0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptoui_4f64_to_4i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vextractf32x4 $1, %ymm0, %xmm1 +; AVX512VL-NEXT: vcvttsd2usi %xmm1, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm1 = xmm1[1,0] +; AVX512VL-NEXT: vcvttsd2usi %xmm1, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] +; AVX512VL-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm0 = xmm0[1,0] +; AVX512VL-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] +; AVX512VL-NEXT: vinserti32x4 $1, %xmm1, %ymm0, %ymm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_4f64_to_4i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttpd2uqq %ymm0, %ymm0 +; AVX512VLDQ-NEXT: retq %cvt = fptoui <4 x double> %a to <4 x i64> ret <4 x i64> %cvt } @@ -705,10 +820,22 @@ define <4 x i32> @fptoui_4f64_to_4i32(<4 x double> %a) { ; AVX-NEXT: vzeroupper ; AVX-NEXT: retq ; -; AVX512-LABEL: fptoui_4f64_to_4i32: -; AVX512: # BB#0: -; AVX512-NEXT: vcvttpd2udq %ymm0, %xmm0 -; AVX512-NEXT: retq +; AVX512F-LABEL: fptoui_4f64_to_4i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: # kill: %YMM0 %YMM0 %ZMM0 +; AVX512F-NEXT: vcvttpd2udq %zmm0, %ymm0 +; AVX512F-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: fptoui_4f64_to_4i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvttpd2udq %ymm0, %xmm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_4f64_to_4i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttpd2udq %ymm0, %xmm0 +; AVX512VLDQ-NEXT: retq %cvt = fptoui <4 x double> %a to <4 x i32> ret <4 x i32> %cvt } @@ -826,11 +953,21 @@ define <2 x i64> @fptosi_4f32_to_2i64(<4 x float> %a) { ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptosi_4f32_to_2i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttps2qq %xmm0, %ymm0 -; AVX512DQ-NEXT: # kill: %XMM0 %XMM0 %YMM0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptosi_4f32_to_2i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] +; AVX512VL-NEXT: vcvttss2si %xmm1, %rax +; AVX512VL-NEXT: vcvttss2si %xmm0, %rcx +; AVX512VL-NEXT: vmovq %rcx, %xmm0 +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_4f32_to_2i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttps2qq %xmm0, %ymm0 +; AVX512VLDQ-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512VLDQ-NEXT: retq %cvt = fptosi <4 x float> %a to <4 x i64> %shuf = shufflevector <4 x i64> %cvt, <4 x i64> undef, <2 x i32> ret <2 x i64> %shuf @@ -928,13 +1065,31 @@ define <4 x i64> @fptosi_4f32_to_4i64(<8 x float> %a) { ; AVX512F-NEXT: vcvttss2si %xmm0, %rax ; AVX512F-NEXT: vmovq %rax, %xmm0 ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] -; AVX512F-NEXT: vinserti32x4 $1, %xmm1, %ymm0, %ymm0 +; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptosi_4f32_to_4i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttps2qq %xmm0, %ymm0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptosi_4f32_to_4i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3] +; AVX512VL-NEXT: vcvttss2si %xmm1, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm2 = xmm0[1,0] +; AVX512VL-NEXT: vcvttss2si %xmm2, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] +; AVX512VL-NEXT: vcvttss2si %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] +; AVX512VL-NEXT: vcvttss2si %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] +; AVX512VL-NEXT: vinserti32x4 $1, %xmm1, %ymm0, %ymm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_4f32_to_4i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttps2qq %xmm0, %ymm0 +; AVX512VLDQ-NEXT: retq %shuf = shufflevector <8 x float> %a, <8 x float> undef, <4 x i32> %cvt = fptosi <4 x float> %shuf to <4 x i64> ret <4 x i64> %cvt @@ -1012,14 +1167,32 @@ define <4 x i64> @fptosi_8f32_to_4i64(<8 x float> %a) { ; AVX512F-NEXT: vmovq %rcx, %xmm1 ; AVX512F-NEXT: vmovq %rax, %xmm2 ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] -; AVX512F-NEXT: vinserti32x4 $1, %xmm0, %ymm1, %ymm0 +; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptosi_8f32_to_4i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttps2qq %ymm0, %zmm0 -; AVX512DQ-NEXT: # kill: %YMM0 %YMM0 %ZMM0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptosi_8f32_to_4i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] +; AVX512VL-NEXT: vcvttss2si %xmm1, %rax +; AVX512VL-NEXT: vcvttss2si %xmm0, %rcx +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0] +; AVX512VL-NEXT: vcvttss2si %xmm1, %rdx +; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3] +; AVX512VL-NEXT: vcvttss2si %xmm0, %rsi +; AVX512VL-NEXT: vmovq %rsi, %xmm0 +; AVX512VL-NEXT: vmovq %rdx, %xmm1 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] +; AVX512VL-NEXT: vmovq %rcx, %xmm1 +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] +; AVX512VL-NEXT: vinserti32x4 $1, %xmm0, %ymm1, %ymm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_8f32_to_4i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttps2qq %ymm0, %zmm0 +; AVX512VLDQ-NEXT: # kill: %YMM0 %YMM0 %ZMM0 +; AVX512VLDQ-NEXT: retq %cvt = fptosi <8 x float> %a to <8 x i64> %shuf = shufflevector <8 x i64> %cvt, <8 x i64> undef, <4 x i32> ret <4 x i64> %shuf @@ -1127,10 +1300,22 @@ define <4 x i32> @fptoui_4f32_to_4i32(<4 x float> %a) { ; AVX-NEXT: vpinsrd $3, %eax, %xmm1, %xmm0 ; AVX-NEXT: retq ; -; AVX512-LABEL: fptoui_4f32_to_4i32: -; AVX512: # BB#0: -; AVX512-NEXT: vcvttps2udq %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX512F-LABEL: fptoui_4f32_to_4i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: # kill: %XMM0 %XMM0 %ZMM0 +; AVX512F-NEXT: vcvttps2udq %zmm0, %zmm0 +; AVX512F-NEXT: # kill: %XMM0 %XMM0 %ZMM0 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: fptoui_4f32_to_4i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvttps2udq %xmm0, %xmm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_4f32_to_4i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttps2udq %xmm0, %xmm0 +; AVX512VLDQ-NEXT: retq %cvt = fptoui <4 x float> %a to <4 x i32> ret <4 x i32> %cvt } @@ -1255,11 +1440,21 @@ define <2 x i64> @fptoui_4f32_to_2i64(<4 x float> %a) { ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptoui_4f32_to_2i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttps2uqq %xmm0, %ymm0 -; AVX512DQ-NEXT: # kill: %XMM0 %XMM0 %YMM0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptoui_4f32_to_2i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] +; AVX512VL-NEXT: vcvttss2usi %xmm1, %rax +; AVX512VL-NEXT: vcvttss2usi %xmm0, %rcx +; AVX512VL-NEXT: vmovq %rcx, %xmm0 +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_4f32_to_2i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttps2uqq %xmm0, %ymm0 +; AVX512VLDQ-NEXT: # kill: %XMM0 %XMM0 %YMM0 +; AVX512VLDQ-NEXT: retq %cvt = fptoui <4 x float> %a to <4 x i64> %shuf = shufflevector <4 x i64> %cvt, <4 x i64> undef, <2 x i32> ret <2 x i64> %shuf @@ -1359,10 +1554,22 @@ define <8 x i32> @fptoui_8f32_to_8i32(<8 x float> %a) { ; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 ; AVX2-NEXT: retq ; -; AVX512-LABEL: fptoui_8f32_to_8i32: -; AVX512: # BB#0: -; AVX512-NEXT: vcvttps2udq %ymm0, %ymm0 -; AVX512-NEXT: retq +; AVX512F-LABEL: fptoui_8f32_to_8i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: # kill: %YMM0 %YMM0 %ZMM0 +; AVX512F-NEXT: vcvttps2udq %zmm0, %zmm0 +; AVX512F-NEXT: # kill: %YMM0 %YMM0 %ZMM0 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: fptoui_8f32_to_8i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvttps2udq %ymm0, %ymm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_8f32_to_8i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttps2udq %ymm0, %ymm0 +; AVX512VLDQ-NEXT: retq %cvt = fptoui <8 x float> %a to <8 x i32> ret <8 x i32> %cvt } @@ -1509,13 +1716,31 @@ define <4 x i64> @fptoui_4f32_to_4i64(<8 x float> %a) { ; AVX512F-NEXT: vcvttss2usi %xmm0, %rax ; AVX512F-NEXT: vmovq %rax, %xmm0 ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] -; AVX512F-NEXT: vinserti32x4 $1, %xmm1, %ymm0, %ymm0 +; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0 ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptoui_4f32_to_4i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttps2uqq %xmm0, %ymm0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptoui_4f32_to_4i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3] +; AVX512VL-NEXT: vcvttss2usi %xmm1, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm2 = xmm0[1,0] +; AVX512VL-NEXT: vcvttss2usi %xmm2, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0] +; AVX512VL-NEXT: vcvttss2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] +; AVX512VL-NEXT: vcvttss2usi %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0] +; AVX512VL-NEXT: vinserti32x4 $1, %xmm1, %ymm0, %ymm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_4f32_to_4i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttps2uqq %xmm0, %ymm0 +; AVX512VLDQ-NEXT: retq %shuf = shufflevector <8 x float> %a, <8 x float> undef, <4 x i32> %cvt = fptoui <4 x float> %shuf to <4 x i64> ret <4 x i64> %cvt @@ -1663,14 +1888,32 @@ define <4 x i64> @fptoui_8f32_to_4i64(<8 x float> %a) { ; AVX512F-NEXT: vmovq %rcx, %xmm1 ; AVX512F-NEXT: vmovq %rax, %xmm2 ; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] -; AVX512F-NEXT: vinserti32x4 $1, %xmm0, %ymm1, %ymm0 +; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 ; AVX512F-NEXT: retq ; -; AVX512DQ-LABEL: fptoui_8f32_to_4i64: -; AVX512DQ: # BB#0: -; AVX512DQ-NEXT: vcvttps2uqq %ymm0, %zmm0 -; AVX512DQ-NEXT: # kill: %YMM0 %YMM0 %ZMM0 -; AVX512DQ-NEXT: retq +; AVX512VL-LABEL: fptoui_8f32_to_4i64: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vmovshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] +; AVX512VL-NEXT: vcvttss2usi %xmm1, %rax +; AVX512VL-NEXT: vcvttss2usi %xmm0, %rcx +; AVX512VL-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0] +; AVX512VL-NEXT: vcvttss2usi %xmm1, %rdx +; AVX512VL-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3] +; AVX512VL-NEXT: vcvttss2usi %xmm0, %rsi +; AVX512VL-NEXT: vmovq %rsi, %xmm0 +; AVX512VL-NEXT: vmovq %rdx, %xmm1 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] +; AVX512VL-NEXT: vmovq %rcx, %xmm1 +; AVX512VL-NEXT: vmovq %rax, %xmm2 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] +; AVX512VL-NEXT: vinserti32x4 $1, %xmm0, %ymm1, %ymm0 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptoui_8f32_to_4i64: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvttps2uqq %ymm0, %zmm0 +; AVX512VLDQ-NEXT: # kill: %YMM0 %YMM0 %ZMM0 +; AVX512VLDQ-NEXT: retq %cvt = fptoui <8 x float> %a to <8 x i64> %shuf = shufflevector <8 x i64> %cvt, <8 x i64> undef, <4 x i32> ret <4 x i64> %shuf @@ -2005,20 +2248,52 @@ define <4 x i32> @fptosi_2f16_to_4i32(<2 x half> %a) nounwind { ; AVX-NEXT: popq %rax ; AVX-NEXT: retq ; -; AVX512-LABEL: fptosi_2f16_to_4i32: -; AVX512: # BB#0: -; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0 -; AVX512-NEXT: vcvtph2ps %xmm0, %xmm0 -; AVX512-NEXT: vcvtps2ph $4, %xmm1, %xmm1 -; AVX512-NEXT: vcvtph2ps %xmm1, %xmm1 -; AVX512-NEXT: vcvttss2si %xmm1, %rax -; AVX512-NEXT: vmovq %rax, %xmm1 -; AVX512-NEXT: vcvttss2si %xmm0, %rax -; AVX512-NEXT: vmovq %rax, %xmm0 -; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] -; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] -; AVX512-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero -; AVX512-NEXT: retq +; AVX512F-LABEL: fptosi_2f16_to_4i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: # kill: %XMM1 %XMM1 %ZMM1 +; AVX512F-NEXT: # kill: %XMM0 %XMM0 %ZMM0 +; AVX512F-NEXT: vcvtps2ph $4, %zmm0, %ymm0 +; AVX512F-NEXT: vcvtph2ps %ymm0, %zmm0 +; AVX512F-NEXT: vcvtps2ph $4, %zmm1, %ymm1 +; AVX512F-NEXT: vcvtph2ps %ymm1, %zmm1 +; AVX512F-NEXT: vcvttss2si %xmm1, %rax +; AVX512F-NEXT: vmovq %rax, %xmm1 +; AVX512F-NEXT: vcvttss2si %xmm0, %rax +; AVX512F-NEXT: vmovq %rax, %xmm0 +; AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: fptosi_2f16_to_4i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: vcvtps2ph $4, %xmm0, %xmm0 +; AVX512VL-NEXT: vcvtph2ps %xmm0, %xmm0 +; AVX512VL-NEXT: vcvtps2ph $4, %xmm1, %xmm1 +; AVX512VL-NEXT: vcvtph2ps %xmm1, %xmm1 +; AVX512VL-NEXT: vcvttss2si %xmm1, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm1 +; AVX512VL-NEXT: vcvttss2si %xmm0, %rax +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_2f16_to_4i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: vcvtps2ph $4, %xmm0, %xmm0 +; AVX512VLDQ-NEXT: vcvtph2ps %xmm0, %xmm0 +; AVX512VLDQ-NEXT: vcvtps2ph $4, %xmm1, %xmm1 +; AVX512VLDQ-NEXT: vcvtph2ps %xmm1, %xmm1 +; AVX512VLDQ-NEXT: vcvttss2si %xmm1, %rax +; AVX512VLDQ-NEXT: vmovq %rax, %xmm1 +; AVX512VLDQ-NEXT: vcvttss2si %xmm0, %rax +; AVX512VLDQ-NEXT: vmovq %rax, %xmm0 +; AVX512VLDQ-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; AVX512VLDQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero +; AVX512VLDQ-NEXT: retq %cvt = fptosi <2 x half> %a to <2 x i32> %ext = shufflevector <2 x i32> %cvt, <2 x i32> zeroinitializer, <4 x i32> ret <4 x i32> %ext @@ -2131,30 +2406,80 @@ define <4 x i32> @fptosi_2f128_to_4i32(<2 x fp128> %a) nounwind { ; AVX-NEXT: popq %r14 ; AVX-NEXT: retq ; -; AVX512-LABEL: fptosi_2f128_to_4i32: -; AVX512: # BB#0: -; AVX512-NEXT: pushq %r14 -; AVX512-NEXT: pushq %rbx -; AVX512-NEXT: subq $24, %rsp -; AVX512-NEXT: movq %rsi, %r14 -; AVX512-NEXT: movq %rdi, %rbx -; AVX512-NEXT: movq %rdx, %rdi -; AVX512-NEXT: movq %rcx, %rsi -; AVX512-NEXT: callq __fixtfdi -; AVX512-NEXT: vmovq %rax, %xmm0 -; AVX512-NEXT: vmovdqa64 %xmm0, (%rsp) # 16-byte Spill -; AVX512-NEXT: movq %rbx, %rdi -; AVX512-NEXT: movq %r14, %rsi -; AVX512-NEXT: callq __fixtfdi -; AVX512-NEXT: vmovq %rax, %xmm0 -; AVX512-NEXT: vmovdqa64 (%rsp), %xmm1 # 16-byte Reload -; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] -; AVX512-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] -; AVX512-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero -; AVX512-NEXT: addq $24, %rsp -; AVX512-NEXT: popq %rbx -; AVX512-NEXT: popq %r14 -; AVX512-NEXT: retq +; AVX512F-LABEL: fptosi_2f128_to_4i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: pushq %r14 +; AVX512F-NEXT: pushq %rbx +; AVX512F-NEXT: subq $24, %rsp +; AVX512F-NEXT: movq %rsi, %r14 +; AVX512F-NEXT: movq %rdi, %rbx +; AVX512F-NEXT: movq %rdx, %rdi +; AVX512F-NEXT: movq %rcx, %rsi +; AVX512F-NEXT: callq __fixtfdi +; AVX512F-NEXT: vmovq %rax, %xmm0 +; AVX512F-NEXT: vmovdqa %xmm0, (%rsp) # 16-byte Spill +; AVX512F-NEXT: movq %rbx, %rdi +; AVX512F-NEXT: movq %r14, %rsi +; AVX512F-NEXT: callq __fixtfdi +; AVX512F-NEXT: vmovq %rax, %xmm0 +; AVX512F-NEXT: vpunpcklqdq (%rsp), %xmm0, %xmm0 # 16-byte Folded Reload +; AVX512F-NEXT: # xmm0 = xmm0[0],mem[0] +; AVX512F-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero +; AVX512F-NEXT: addq $24, %rsp +; AVX512F-NEXT: popq %rbx +; AVX512F-NEXT: popq %r14 +; AVX512F-NEXT: retq +; +; AVX512VL-LABEL: fptosi_2f128_to_4i32: +; AVX512VL: # BB#0: +; AVX512VL-NEXT: pushq %r14 +; AVX512VL-NEXT: pushq %rbx +; AVX512VL-NEXT: subq $24, %rsp +; AVX512VL-NEXT: movq %rsi, %r14 +; AVX512VL-NEXT: movq %rdi, %rbx +; AVX512VL-NEXT: movq %rdx, %rdi +; AVX512VL-NEXT: movq %rcx, %rsi +; AVX512VL-NEXT: callq __fixtfdi +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vmovdqa64 %xmm0, (%rsp) # 16-byte Spill +; AVX512VL-NEXT: movq %rbx, %rdi +; AVX512VL-NEXT: movq %r14, %rsi +; AVX512VL-NEXT: callq __fixtfdi +; AVX512VL-NEXT: vmovq %rax, %xmm0 +; AVX512VL-NEXT: vmovdqa64 (%rsp), %xmm1 # 16-byte Reload +; AVX512VL-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512VL-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero +; AVX512VL-NEXT: addq $24, %rsp +; AVX512VL-NEXT: popq %rbx +; AVX512VL-NEXT: popq %r14 +; AVX512VL-NEXT: retq +; +; AVX512VLDQ-LABEL: fptosi_2f128_to_4i32: +; AVX512VLDQ: # BB#0: +; AVX512VLDQ-NEXT: pushq %r14 +; AVX512VLDQ-NEXT: pushq %rbx +; AVX512VLDQ-NEXT: subq $24, %rsp +; AVX512VLDQ-NEXT: movq %rsi, %r14 +; AVX512VLDQ-NEXT: movq %rdi, %rbx +; AVX512VLDQ-NEXT: movq %rdx, %rdi +; AVX512VLDQ-NEXT: movq %rcx, %rsi +; AVX512VLDQ-NEXT: callq __fixtfdi +; AVX512VLDQ-NEXT: vmovq %rax, %xmm0 +; AVX512VLDQ-NEXT: vmovdqa64 %xmm0, (%rsp) # 16-byte Spill +; AVX512VLDQ-NEXT: movq %rbx, %rdi +; AVX512VLDQ-NEXT: movq %r14, %rsi +; AVX512VLDQ-NEXT: callq __fixtfdi +; AVX512VLDQ-NEXT: vmovq %rax, %xmm0 +; AVX512VLDQ-NEXT: vmovdqa64 (%rsp), %xmm1 # 16-byte Reload +; AVX512VLDQ-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] +; AVX512VLDQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero +; AVX512VLDQ-NEXT: addq $24, %rsp +; AVX512VLDQ-NEXT: popq %rbx +; AVX512VLDQ-NEXT: popq %r14 +; AVX512VLDQ-NEXT: retq %cvt = fptosi <2 x fp128> %a to <2 x i32> %ext = shufflevector <2 x i32> %cvt, <2 x i32> zeroinitializer, <4 x i32> ret <4 x i32> %ext