234 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s --check-prefix=X86-X873; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X86-SSE4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X645 6; Ideally this would compile to 5 multiplies.7 8define double @pow_wrapper(double %a) nounwind readonly ssp noredzone {9; X86-X87-LABEL: pow_wrapper:10; X86-X87: # %bb.0:11; X86-X87-NEXT: fldl {{[0-9]+}}(%esp)12; X86-X87-NEXT: fld %st(0)13; X86-X87-NEXT: fmul %st(1), %st14; X86-X87-NEXT: fmul %st, %st(1)15; X86-X87-NEXT: fmul %st, %st(0)16; X86-X87-NEXT: fmul %st, %st(1)17; X86-X87-NEXT: fmul %st, %st(0)18; X86-X87-NEXT: fmulp %st, %st(1)19; X86-X87-NEXT: retl20;21; X86-SSE-LABEL: pow_wrapper:22; X86-SSE: # %bb.0:23; X86-SSE-NEXT: pushl %ebp24; X86-SSE-NEXT: movl %esp, %ebp25; X86-SSE-NEXT: andl $-8, %esp26; X86-SSE-NEXT: subl $8, %esp27; X86-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero28; X86-SSE-NEXT: movapd %xmm0, %xmm129; X86-SSE-NEXT: mulsd %xmm0, %xmm130; X86-SSE-NEXT: mulsd %xmm1, %xmm031; X86-SSE-NEXT: mulsd %xmm1, %xmm132; X86-SSE-NEXT: mulsd %xmm1, %xmm033; X86-SSE-NEXT: mulsd %xmm1, %xmm134; X86-SSE-NEXT: mulsd %xmm0, %xmm135; X86-SSE-NEXT: movsd %xmm1, (%esp)36; X86-SSE-NEXT: fldl (%esp)37; X86-SSE-NEXT: movl %ebp, %esp38; X86-SSE-NEXT: popl %ebp39; X86-SSE-NEXT: retl40;41; X64-LABEL: pow_wrapper:42; X64: # %bb.0:43; X64-NEXT: movapd %xmm0, %xmm144; X64-NEXT: mulsd %xmm0, %xmm145; X64-NEXT: mulsd %xmm1, %xmm046; X64-NEXT: mulsd %xmm1, %xmm147; X64-NEXT: mulsd %xmm1, %xmm048; X64-NEXT: mulsd %xmm1, %xmm149; X64-NEXT: mulsd %xmm1, %xmm050; X64-NEXT: retq51 %ret = tail call double @llvm.powi.f64.i32(double %a, i32 15) nounwind ; <double> [#uses=1]52 ret double %ret53}54 55define double @pow_wrapper_optsize(double %a) nounwind optsize {56; X86-X87-LABEL: pow_wrapper_optsize:57; X86-X87: # %bb.0:58; X86-X87-NEXT: subl $12, %esp59; X86-X87-NEXT: fldl {{[0-9]+}}(%esp)60; X86-X87-NEXT: fstpl (%esp)61; X86-X87-NEXT: movl $15, {{[0-9]+}}(%esp)62; X86-X87-NEXT: calll __powidf263; X86-X87-NEXT: addl $12, %esp64; X86-X87-NEXT: retl65;66; X86-SSE-LABEL: pow_wrapper_optsize:67; X86-SSE: # %bb.0:68; X86-SSE-NEXT: subl $12, %esp69; X86-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero70; X86-SSE-NEXT: movsd %xmm0, (%esp)71; X86-SSE-NEXT: movl $15, {{[0-9]+}}(%esp)72; X86-SSE-NEXT: calll __powidf273; X86-SSE-NEXT: addl $12, %esp74; X86-SSE-NEXT: retl75;76; X64-LABEL: pow_wrapper_optsize:77; X64: # %bb.0:78; X64-NEXT: movl $15, %edi79; X64-NEXT: jmp __powidf2@PLT # TAILCALL80 %ret = tail call double @llvm.powi.f64.i32(double %a, i32 15) nounwind ; <double> [#uses=1]81 ret double %ret82}83 84define double @pow_wrapper_pgso(double %a) nounwind !prof !14 {85; X86-X87-LABEL: pow_wrapper_pgso:86; X86-X87: # %bb.0:87; X86-X87-NEXT: subl $12, %esp88; X86-X87-NEXT: fldl {{[0-9]+}}(%esp)89; X86-X87-NEXT: fstpl (%esp)90; X86-X87-NEXT: movl $15, {{[0-9]+}}(%esp)91; X86-X87-NEXT: calll __powidf292; X86-X87-NEXT: addl $12, %esp93; X86-X87-NEXT: retl94;95; X86-SSE-LABEL: pow_wrapper_pgso:96; X86-SSE: # %bb.0:97; X86-SSE-NEXT: subl $12, %esp98; X86-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero99; X86-SSE-NEXT: movsd %xmm0, (%esp)100; X86-SSE-NEXT: movl $15, {{[0-9]+}}(%esp)101; X86-SSE-NEXT: calll __powidf2102; X86-SSE-NEXT: addl $12, %esp103; X86-SSE-NEXT: retl104;105; X64-LABEL: pow_wrapper_pgso:106; X64: # %bb.0:107; X64-NEXT: movl $15, %edi108; X64-NEXT: jmp __powidf2@PLT # TAILCALL109 %ret = tail call double @llvm.powi.f64.i32(double %a, i32 15) nounwind ; <double> [#uses=1]110 ret double %ret111}112 113define double @pow_wrapper_minsize(double %a) nounwind minsize {114; X86-X87-LABEL: pow_wrapper_minsize:115; X86-X87: # %bb.0:116; X86-X87-NEXT: subl $12, %esp117; X86-X87-NEXT: fldl {{[0-9]+}}(%esp)118; X86-X87-NEXT: fstpl (%esp)119; X86-X87-NEXT: movl $15, {{[0-9]+}}(%esp)120; X86-X87-NEXT: calll __powidf2121; X86-X87-NEXT: addl $12, %esp122; X86-X87-NEXT: retl123;124; X86-SSE-LABEL: pow_wrapper_minsize:125; X86-SSE: # %bb.0:126; X86-SSE-NEXT: subl $12, %esp127; X86-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero128; X86-SSE-NEXT: movsd %xmm0, (%esp)129; X86-SSE-NEXT: movl $15, {{[0-9]+}}(%esp)130; X86-SSE-NEXT: calll __powidf2131; X86-SSE-NEXT: addl $12, %esp132; X86-SSE-NEXT: retl133;134; X64-LABEL: pow_wrapper_minsize:135; X64: # %bb.0:136; X64-NEXT: pushq $15137; X64-NEXT: popq %rdi138; X64-NEXT: jmp __powidf2@PLT # TAILCALL139 %ret = tail call double @llvm.powi.f64.i32(double %a, i32 15) nounwind ; <double> [#uses=1]140 ret double %ret141}142 143define <2 x float> @powi_v2f32(<2 x float> %a) nounwind minsize {144; X86-X87-LABEL: powi_v2f32:145; X86-X87: # %bb.0:146; X86-X87-NEXT: pushl %esi147; X86-X87-NEXT: subl $16, %esp148; X86-X87-NEXT: flds {{[0-9]+}}(%esp)149; X86-X87-NEXT: fstps {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill150; X86-X87-NEXT: flds {{[0-9]+}}(%esp)151; X86-X87-NEXT: pushl $15152; X86-X87-NEXT: popl %esi153; X86-X87-NEXT: movl %esi, {{[0-9]+}}(%esp)154; X86-X87-NEXT: fstps (%esp)155; X86-X87-NEXT: calll __powisf2156; X86-X87-NEXT: fstps {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill157; X86-X87-NEXT: movl %esi, {{[0-9]+}}(%esp)158; X86-X87-NEXT: flds {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload159; X86-X87-NEXT: fstps (%esp)160; X86-X87-NEXT: calll __powisf2161; X86-X87-NEXT: flds {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload162; X86-X87-NEXT: fxch %st(1)163; X86-X87-NEXT: addl $16, %esp164; X86-X87-NEXT: popl %esi165; X86-X87-NEXT: retl166;167; X86-SSE-LABEL: powi_v2f32:168; X86-SSE: # %bb.0:169; X86-SSE-NEXT: pushl %esi170; X86-SSE-NEXT: subl $32, %esp171; X86-SSE-NEXT: movups %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill172; X86-SSE-NEXT: pushl $15173; X86-SSE-NEXT: popl %esi174; X86-SSE-NEXT: movl %esi, {{[0-9]+}}(%esp)175; X86-SSE-NEXT: movss %xmm0, (%esp)176; X86-SSE-NEXT: calll __powisf2177; X86-SSE-NEXT: movl %esi, {{[0-9]+}}(%esp)178; X86-SSE-NEXT: movups {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload179; X86-SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]180; X86-SSE-NEXT: movss %xmm0, (%esp)181; X86-SSE-NEXT: fstps {{[0-9]+}}(%esp)182; X86-SSE-NEXT: calll __powisf2183; X86-SSE-NEXT: fstps {{[0-9]+}}(%esp)184; X86-SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero185; X86-SSE-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero186; X86-SSE-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]187; X86-SSE-NEXT: addl $32, %esp188; X86-SSE-NEXT: popl %esi189; X86-SSE-NEXT: retl190;191; X64-LABEL: powi_v2f32:192; X64: # %bb.0:193; X64-NEXT: pushq %rbx194; X64-NEXT: subq $32, %rsp195; X64-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill196; X64-NEXT: pushq $15197; X64-NEXT: popq %rbx198; X64-NEXT: movl %ebx, %edi199; X64-NEXT: callq __powisf2@PLT200; X64-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill201; X64-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload202; X64-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]203; X64-NEXT: movl %ebx, %edi204; X64-NEXT: callq __powisf2@PLT205; X64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload206; X64-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]207; X64-NEXT: movaps %xmm1, %xmm0208; X64-NEXT: addq $32, %rsp209; X64-NEXT: popq %rbx210; X64-NEXT: retq211 %ret = tail call < 2 x float> @llvm.powi.v2f32.i32(<2 x float> %a, i32 15) nounwind ;212 ret <2 x float> %ret213}214 215declare double @llvm.powi.f64.i32(double, i32) nounwind readonly216declare < 2 x float> @llvm.powi.v2f32.i32(<2 x float>, i32) nounwind readonly217 218!llvm.module.flags = !{!0}219!0 = !{i32 1, !"ProfileSummary", !1}220!1 = !{!2, !3, !4, !5, !6, !7, !8, !9}221!2 = !{!"ProfileFormat", !"InstrProf"}222!3 = !{!"TotalCount", i64 10000}223!4 = !{!"MaxCount", i64 10}224!5 = !{!"MaxInternalCount", i64 1}225!6 = !{!"MaxFunctionCount", i64 1000}226!7 = !{!"NumCounts", i64 3}227!8 = !{!"NumFunctions", i64 3}228!9 = !{!"DetailedSummary", !10}229!10 = !{!11, !12, !13}230!11 = !{i32 10000, i64 100, i32 1}231!12 = !{i32 999000, i64 100, i32 1}232!13 = !{i32 999999, i64 1, i32 2}233!14 = !{!"function_entry_count", i64 0}234