diff --git a/llvm/test/CodeGen/X86/fp-stack-ret.ll b/llvm/test/CodeGen/X86/fp-stack-ret.ll index c83a0cbf69e0..1307f70ead17 100644 --- a/llvm/test/CodeGen/X86/fp-stack-ret.ll +++ b/llvm/test/CodeGen/X86/fp-stack-ret.ll @@ -1,25 +1,40 @@ -; RUN: llc < %s -mtriple=i686-apple-darwin8 -mcpu=yonah -march=x86 > %t -; RUN: grep fldl %t | count 1 -; RUN: not grep xmm %t -; RUN: grep {sub.*esp} %t | count 1 +; RUN: llc < %s -mtriple=i686-apple-darwin8 -mcpu=yonah -march=x86 | FileCheck %s ; These testcases shouldn't require loading into an XMM register then storing ; to memory, then reloading into an FPStack reg. +; CHECK: test1 +; CHECK: fldl +; CHECK-NEXT: ret define double @test1(double *%P) { %A = load double* %P ret double %A } -; fastcc should return a value +; fastcc should return a value +; CHECK: test2 +; CHECK-NOT: xmm +; CHECK: ret define fastcc double @test2(<2 x double> %A) { %B = extractelement <2 x double> %A, i32 0 ret double %B } +; CHECK: test3 +; CHECK: sub{{.*}}%esp +; CHECLK-NOT: xmm define fastcc double @test3(<4 x float> %A) { %B = bitcast <4 x float> %A to <2 x double> %C = call fastcc double @test2(<2 x double> %B) ret double %C } - + +; Clear the stack when not using a return value. +; CHECK: test4 +; CHECK: call +; CHECK: fstp +; CHECK: ret +define void @test4(double *%P) { + %A = call double @test1(double *%P) + ret void +} diff --git a/llvm/test/CodeGen/X86/pr1505b.ll b/llvm/test/CodeGen/X86/pr1505b.ll index 6a08dae51f8a..945ec4c6b621 100644 --- a/llvm/test/CodeGen/X86/pr1505b.ll +++ b/llvm/test/CodeGen/X86/pr1505b.ll @@ -1,5 +1,4 @@ -; RUN: llc < %s -mcpu=i486 | grep fstpl | count 5 -; RUN: llc < %s -mcpu=i486 | grep fstps | count 2 +; RUN: llc < %s -mcpu=i486 | FileCheck %s ; PR1505 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64" @@ -30,19 +29,41 @@ declare void @_ZNSt8ios_base4InitC1Ev(%"struct.std::ctype_base"*) declare i32 @__cxa_atexit(void (i8*)*, i8*, i8*) +; CHECK: main define i32 @main() { entry: +; CHECK: flds %tmp6 = volatile load float* @a ; [#uses=1] +; CHECK: fstps (%esp) +; CHECK: tanf %tmp9 = tail call float @tanf( float %tmp6 ) ; [#uses=1] +; Spill returned value: +; CHECK: fstp + +; CHECK: fldl %tmp12 = volatile load double* @b ; [#uses=1] +; CHECK: fstpl (%esp) +; CHECK: tan %tmp13 = tail call double @tan( double %tmp12 ) ; [#uses=1] +; Spill returned value: +; CHECK: fstp %tmp1314 = fptrunc double %tmp13 to float ; [#uses=1] %tmp16 = tail call %"struct.std::basic_ostream >"* @_ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_PKc( %"struct.std::basic_ostream >"* @_ZSt4cout, i8* getelementptr ([12 x i8]* @.str, i32 0, i32 0) ) ; <%"struct.std::basic_ostream >"*> [#uses=1] %tmp1920 = fpext float %tmp9 to double ; [#uses=1] +; reload: +; CHECK: fld +; CHECK: fstpl +; CHECK: ZNSolsEd %tmp22 = tail call %"struct.std::basic_ostream >"* @_ZNSolsEd( %"struct.std::basic_ostream >"* %tmp16, double %tmp1920 ) ; <%"struct.std::basic_ostream >"*> [#uses=1] %tmp30 = tail call %"struct.std::basic_ostream >"* @_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_( %"struct.std::basic_ostream >"* %tmp22 ) ; <%"struct.std::basic_ostream >"*> [#uses=0] +; reload: +; CHECK: fld +; CHECK: fstps +; CHECK: ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_PKc %tmp34 = tail call %"struct.std::basic_ostream >"* @_ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_PKc( %"struct.std::basic_ostream >"* @_ZSt4cout, i8* getelementptr ([13 x i8]* @.str1, i32 0, i32 0) ) ; <%"struct.std::basic_ostream >"*> [#uses=1] %tmp3940 = fpext float %tmp1314 to double ; [#uses=1] +; CHECK: fstpl +; CHECK: ZNSolsEd %tmp42 = tail call %"struct.std::basic_ostream >"* @_ZNSolsEd( %"struct.std::basic_ostream >"* %tmp34, double %tmp3940 ) ; <%"struct.std::basic_ostream >"*> [#uses=1] %tmp51 = tail call %"struct.std::basic_ostream >"* @_ZSt4endlIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_( %"struct.std::basic_ostream >"* %tmp42 ) ; <%"struct.std::basic_ostream >"*> [#uses=0] ret i32 0