1; RUN: llc < %s -mtriple=i686-apple-darwin9 -mattr=sse4.1 -mcpu=penryn | FileCheck %s -check-prefix=X32 2; RUN: llc < %s -mtriple=x86_64-apple-darwin9 -mattr=sse4.1 -mcpu=penryn | FileCheck %s -check-prefix=X64 3 4@g16 = external global i16 5 6define <4 x i32> @pinsrd_1(i32 %s, <4 x i32> %tmp) nounwind { 7 %tmp1 = insertelement <4 x i32> %tmp, i32 %s, i32 1 8 ret <4 x i32> %tmp1 9; X32-LABEL: pinsrd_1: 10; X32: pinsrd $1, 4(%esp), %xmm0 11 12; X64-LABEL: pinsrd_1: 13; X64: pinsrd $1, %edi, %xmm0 14} 15 16define <16 x i8> @pinsrb_1(i8 %s, <16 x i8> %tmp) nounwind { 17 %tmp1 = insertelement <16 x i8> %tmp, i8 %s, i32 1 18 ret <16 x i8> %tmp1 19; X32-LABEL: pinsrb_1: 20; X32: pinsrb $1, 4(%esp), %xmm0 21 22; X64-LABEL: pinsrb_1: 23; X64: pinsrb $1, %edi, %xmm0 24} 25 26 27define <2 x i64> @pmovsxbd_1(i32* %p) nounwind { 28entry: 29 %0 = load i32* %p, align 4 30 %1 = insertelement <4 x i32> undef, i32 %0, i32 0 31 %2 = insertelement <4 x i32> %1, i32 0, i32 1 32 %3 = insertelement <4 x i32> %2, i32 0, i32 2 33 %4 = insertelement <4 x i32> %3, i32 0, i32 3 34 %5 = bitcast <4 x i32> %4 to <16 x i8> 35 %6 = tail call <4 x i32> @llvm.x86.sse41.pmovsxbd(<16 x i8> %5) nounwind readnone 36 %7 = bitcast <4 x i32> %6 to <2 x i64> 37 ret <2 x i64> %7 38 39; X32: _pmovsxbd_1: 40; X32: movl 4(%esp), %eax 41; X32: pmovsxbd (%eax), %xmm0 42 43; X64: _pmovsxbd_1: 44; X64: pmovsxbd (%rdi), %xmm0 45} 46 47define <2 x i64> @pmovsxwd_1(i64* %p) nounwind readonly { 48entry: 49 %0 = load i64* %p ; <i64> [#uses=1] 50 %tmp2 = insertelement <2 x i64> zeroinitializer, i64 %0, i32 0 ; <<2 x i64>> [#uses=1] 51 %1 = bitcast <2 x i64> %tmp2 to <8 x i16> ; <<8 x i16>> [#uses=1] 52 %2 = tail call <4 x i32> @llvm.x86.sse41.pmovsxwd(<8 x i16> %1) nounwind readnone ; <<4 x i32>> [#uses=1] 53 %3 = bitcast <4 x i32> %2 to <2 x i64> ; <<2 x i64>> [#uses=1] 54 ret <2 x i64> %3 55 56; X32: _pmovsxwd_1: 57; X32: movl 4(%esp), %eax 58; X32: pmovsxwd (%eax), %xmm0 59 60; X64: _pmovsxwd_1: 61; X64: pmovsxwd (%rdi), %xmm0 62} 63 64 65 66 67define <2 x i64> @pmovzxbq_1() nounwind { 68entry: 69 %0 = load i16* @g16, align 2 ; <i16> [#uses=1] 70 %1 = insertelement <8 x i16> undef, i16 %0, i32 0 ; <<8 x i16>> [#uses=1] 71 %2 = bitcast <8 x i16> %1 to <16 x i8> ; <<16 x i8>> [#uses=1] 72 %3 = tail call <2 x i64> @llvm.x86.sse41.pmovzxbq(<16 x i8> %2) nounwind readnone ; <<2 x i64>> [#uses=1] 73 ret <2 x i64> %3 74 75; X32: _pmovzxbq_1: 76; X32: movl L_g16$non_lazy_ptr, %eax 77; X32: pmovzxbq (%eax), %xmm0 78 79; X64: _pmovzxbq_1: 80; X64: movq _g16@GOTPCREL(%rip), %rax 81; X64: pmovzxbq (%rax), %xmm0 82} 83 84declare <4 x i32> @llvm.x86.sse41.pmovsxbd(<16 x i8>) nounwind readnone 85declare <4 x i32> @llvm.x86.sse41.pmovsxwd(<8 x i16>) nounwind readnone 86declare <2 x i64> @llvm.x86.sse41.pmovzxbq(<16 x i8>) nounwind readnone 87 88 89 90 91define i32 @extractps_1(<4 x float> %v) nounwind { 92 %s = extractelement <4 x float> %v, i32 3 93 %i = bitcast float %s to i32 94 ret i32 %i 95 96; X32: _extractps_1: 97; X32: extractps $3, %xmm0, %eax 98 99; X64: _extractps_1: 100; X64: extractps $3, %xmm0, %eax 101} 102define i32 @extractps_2(<4 x float> %v) nounwind { 103 %t = bitcast <4 x float> %v to <4 x i32> 104 %s = extractelement <4 x i32> %t, i32 3 105 ret i32 %s 106 107; X32: _extractps_2: 108; X32: extractps $3, %xmm0, %eax 109 110; X64: _extractps_2: 111; X64: extractps $3, %xmm0, %eax 112} 113 114 115; The non-store form of extractps puts its result into a GPR. 116; This makes it suitable for an extract from a <4 x float> that 117; is bitcasted to i32, but unsuitable for much of anything else. 118 119define float @ext_1(<4 x float> %v) nounwind { 120 %s = extractelement <4 x float> %v, i32 3 121 %t = fadd float %s, 1.0 122 ret float %t 123 124; X32: _ext_1: 125; X32: pshufd $3, %xmm0, %xmm0 126; X32: addss LCPI7_0, %xmm0 127 128; X64: _ext_1: 129; X64: pshufd $3, %xmm0, %xmm0 130; X64: addss LCPI7_0(%rip), %xmm0 131} 132define float @ext_2(<4 x float> %v) nounwind { 133 %s = extractelement <4 x float> %v, i32 3 134 ret float %s 135 136; X32: _ext_2: 137; X32: pshufd $3, %xmm0, %xmm0 138 139; X64: _ext_2: 140; X64: pshufd $3, %xmm0, %xmm0 141} 142define i32 @ext_3(<4 x i32> %v) nounwind { 143 %i = extractelement <4 x i32> %v, i32 3 144 ret i32 %i 145 146; X32: _ext_3: 147; X32: pextrd $3, %xmm0, %eax 148 149; X64: _ext_3: 150; X64: pextrd $3, %xmm0, %eax 151} 152 153define <4 x float> @insertps_1(<4 x float> %t1, <4 x float> %t2) nounwind { 154 %tmp1 = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %t1, <4 x float> %t2, i32 1) nounwind readnone 155 ret <4 x float> %tmp1 156; X32: _insertps_1: 157; X32: insertps $1, %xmm1, %xmm0 158 159; X64: _insertps_1: 160; X64: insertps $1, %xmm1, %xmm0 161} 162 163declare <4 x float> @llvm.x86.sse41.insertps(<4 x float>, <4 x float>, i32) nounwind readnone 164 165define <4 x float> @insertps_2(<4 x float> %t1, float %t2) nounwind { 166 %tmp1 = insertelement <4 x float> %t1, float %t2, i32 0 167 ret <4 x float> %tmp1 168; X32: _insertps_2: 169; X32: insertps $0, 4(%esp), %xmm0 170 171; X64: _insertps_2: 172; X64: insertps $0, %xmm1, %xmm0 173} 174 175define <4 x float> @insertps_3(<4 x float> %t1, <4 x float> %t2) nounwind { 176 %tmp2 = extractelement <4 x float> %t2, i32 0 177 %tmp1 = insertelement <4 x float> %t1, float %tmp2, i32 0 178 ret <4 x float> %tmp1 179; X32: _insertps_3: 180; X32: insertps $0, %xmm1, %xmm0 181 182; X64: _insertps_3: 183; X64: insertps $0, %xmm1, %xmm0 184} 185 186define i32 @ptestz_1(<2 x i64> %t1, <2 x i64> %t2) nounwind { 187 %tmp1 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> %t1, <2 x i64> %t2) nounwind readnone 188 ret i32 %tmp1 189; X32: _ptestz_1: 190; X32: ptest %xmm1, %xmm0 191; X32: sete %al 192 193; X64: _ptestz_1: 194; X64: ptest %xmm1, %xmm0 195; X64: sete %al 196} 197 198define i32 @ptestz_2(<2 x i64> %t1, <2 x i64> %t2) nounwind { 199 %tmp1 = call i32 @llvm.x86.sse41.ptestc(<2 x i64> %t1, <2 x i64> %t2) nounwind readnone 200 ret i32 %tmp1 201; X32: _ptestz_2: 202; X32: ptest %xmm1, %xmm0 203; X32: sbbl %eax 204 205; X64: _ptestz_2: 206; X64: ptest %xmm1, %xmm0 207; X64: sbbl %eax 208} 209 210define i32 @ptestz_3(<2 x i64> %t1, <2 x i64> %t2) nounwind { 211 %tmp1 = call i32 @llvm.x86.sse41.ptestnzc(<2 x i64> %t1, <2 x i64> %t2) nounwind readnone 212 ret i32 %tmp1 213; X32: _ptestz_3: 214; X32: ptest %xmm1, %xmm0 215; X32: seta %al 216 217; X64: _ptestz_3: 218; X64: ptest %xmm1, %xmm0 219; X64: seta %al 220} 221 222 223declare i32 @llvm.x86.sse41.ptestz(<2 x i64>, <2 x i64>) nounwind readnone 224declare i32 @llvm.x86.sse41.ptestc(<2 x i64>, <2 x i64>) nounwind readnone 225declare i32 @llvm.x86.sse41.ptestnzc(<2 x i64>, <2 x i64>) nounwind readnone 226 227; This used to compile to insertps $0 + insertps $16. insertps $0 is always 228; pointless. 229define <2 x float> @buildvector(<2 x float> %A, <2 x float> %B) nounwind { 230entry: 231 %tmp7 = extractelement <2 x float> %A, i32 0 232 %tmp5 = extractelement <2 x float> %A, i32 1 233 %tmp3 = extractelement <2 x float> %B, i32 0 234 %tmp1 = extractelement <2 x float> %B, i32 1 235 %add.r = fadd float %tmp7, %tmp3 236 %add.i = fadd float %tmp5, %tmp1 237 %tmp11 = insertelement <2 x float> undef, float %add.r, i32 0 238 %tmp9 = insertelement <2 x float> %tmp11, float %add.i, i32 1 239 ret <2 x float> %tmp9 240; X32-LABEL: buildvector: 241; X32-NOT: insertps $0 242; X32: insertps $16 243; X32-NOT: insertps $0 244; X32: ret 245; X64-LABEL: buildvector: 246; X64-NOT: insertps $0 247; X64: insertps $16 248; X64-NOT: insertps $0 249; X64: ret 250} 251 252