From: Simon Pilgrim Date: Thu, 24 Nov 2016 14:02:30 +0000 (+0000) Subject: [X86][AVX512DQVL] Add awareness of vcvtqq2ps and vcvtuqq2ps implicit zeroing of upper... X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=1a88fd21d555faec17c3e438ca58377aa515eda5;p=llvm [X86][AVX512DQVL] Add awareness of vcvtqq2ps and vcvtuqq2ps implicit zeroing of upper 64-bits of xmm result git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287878 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td index 07e58386c2d..2793df3e584 100644 --- a/lib/Target/X86/X86InstrAVX512.td +++ b/lib/Target/X86/X86InstrAVX512.td @@ -6569,6 +6569,17 @@ let Predicates = [HasAVX512] in { (VCVTPS2PDZrm addr:$src)>; } +let Predicates = [HasDQI, HasVLX] in { + let AddedComplexity = 15 in { + def : Pat<(X86vzmovl (v2f64 (bitconvert + (v4f32 (X86VSintToFP (v2i64 VR128X:$src)))))), + (VCVTQQ2PSZ128rr VR128:$src)>; + def : Pat<(X86vzmovl (v2f64 (bitconvert + (v4f32 (X86VUintToFP (v2i64 VR128X:$src)))))), + (VCVTUQQ2PSZ128rr VR128:$src)>; + } +} + let Predicates = [HasDQI, NoVLX] in { def : Pat<(v4i64 (fp_to_sint (v4f32 VR128X:$src1))), (EXTRACT_SUBREG (v8i64 (VCVTTPS2QQZrr diff --git a/test/CodeGen/X86/avx512dqvl-intrinsics.ll b/test/CodeGen/X86/avx512dqvl-intrinsics.ll index 6de1db0fbde..c6ab1c36770 100644 --- a/test/CodeGen/X86/avx512dqvl-intrinsics.ll +++ b/test/CodeGen/X86/avx512dqvl-intrinsics.ll @@ -183,8 +183,6 @@ define <4 x float>@test_int_x86_avx512_mask_cvt_qq2ps_128_zext(<2 x i64> %x0, <4 ; CHECK-NEXT: kmovb %edi, %k1 ## encoding: [0xc5,0xf9,0x92,0xcf] ; CHECK-NEXT: vcvtqq2ps %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0xfc,0x09,0x5b,0xc8] ; CHECK-NEXT: vcvtqq2ps %xmm0, %xmm0 ## encoding: [0x62,0xf1,0xfc,0x08,0x5b,0xc0] -; CHECK-NEXT: vmovq %xmm0, %xmm0 ## encoding: [0x62,0xf1,0xfe,0x08,0x7e,0xc0] -; CHECK-NEXT: ## xmm0 = xmm0[0],zero ; CHECK-NEXT: vaddps %xmm0, %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x74,0x08,0x58,0xc0] ; CHECK-NEXT: retq ## encoding: [0xc3] %res = call <4 x float> @llvm.x86.avx512.mask.cvtqq2ps.128(<2 x i64> %x0, <4 x float> %x1, i8 %x2) @@ -360,8 +358,6 @@ define <4 x float>@test_int_x86_avx512_mask_cvt_uqq2ps_128_zext(<2 x i64> %x0, < ; CHECK-NEXT: kmovb %edi, %k1 ## encoding: [0xc5,0xf9,0x92,0xcf] ; CHECK-NEXT: vcvtuqq2ps %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0xff,0x09,0x7a,0xc8] ; CHECK-NEXT: vcvtuqq2ps %xmm0, %xmm0 ## encoding: [0x62,0xf1,0xff,0x08,0x7a,0xc0] -; CHECK-NEXT: vmovq %xmm0, %xmm0 ## encoding: [0x62,0xf1,0xfe,0x08,0x7e,0xc0] -; CHECK-NEXT: ## xmm0 = xmm0[0],zero ; CHECK-NEXT: vaddps %xmm0, %xmm1, %xmm0 ## encoding: [0x62,0xf1,0x74,0x08,0x58,0xc0] ; CHECK-NEXT: retq ## encoding: [0xc3] %res = call <4 x float> @llvm.x86.avx512.mask.cvtuqq2ps.128(<2 x i64> %x0, <4 x float> %x1, i8 %x2) diff --git a/test/CodeGen/X86/vec_int_to_fp.ll b/test/CodeGen/X86/vec_int_to_fp.ll index 8fbc142fa3d..13cdc48d5d9 100644 --- a/test/CodeGen/X86/vec_int_to_fp.ll +++ b/test/CodeGen/X86/vec_int_to_fp.ll @@ -1182,7 +1182,6 @@ define <4 x float> @sitofp_2i64_to_4f32_zero(<2 x i64> %a) { ; AVX512VLDQ-LABEL: sitofp_2i64_to_4f32_zero: ; AVX512VLDQ: # BB#0: ; AVX512VLDQ-NEXT: vcvtqq2ps %xmm0, %xmm0 -; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero ; AVX512VLDQ-NEXT: retq %cvt = sitofp <2 x i64> %a to <2 x float> %ext = shufflevector <2 x float> %cvt, <2 x float> zeroinitializer, <4 x i32> @@ -1847,7 +1846,6 @@ define <4 x float> @uitofp_2i64_to_2f32(<2 x i64> %a) { ; AVX512VLDQ-LABEL: uitofp_2i64_to_2f32: ; AVX512VLDQ: # BB#0: ; AVX512VLDQ-NEXT: vcvtuqq2ps %xmm0, %xmm0 -; AVX512VLDQ-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero ; AVX512VLDQ-NEXT: retq %cvt = uitofp <2 x i64> %a to <2 x float> %ext = shufflevector <2 x float> %cvt, <2 x float> zeroinitializer, <4 x i32>