[llvm-branch-commits] [llvm-branch] r286248 - [3.9.1] Merging r282613 - [X86][AVX] Add test showing that VBROADCAST loads don't correctly respect dependencies

[Simon Pilgrim via llvm-branch-commits]

Author: rksimon
Date: Tue Nov  8 10:45:26 2016
New Revision: 286248

URL: http://llvm.org/viewvc/llvm-project?rev=286248&view=rev
Log:
[3.9.1] Merging r282613 - [X86][AVX] Add test showing that VBROADCAST loads don't correctly respect dependencies

As discussed in PR30596, this is a preliminary test update before we can merge r283070

Note: This required the test to be regenerated after the merge as 3.9.1 doesn't have trunk's latest lea -> mov simplifications

Modified:
    llvm/branches/release_39/   (props changed)
    llvm/branches/release_39/test/CodeGen/X86/avx-vbroadcast.ll

Propchange: llvm/branches/release_39/
------------------------------------------------------------------------------
--- svn:mergeinfo (original)
+++ svn:mergeinfo Tue Nov  8 10:45:26 2016
@@ -1,3 +1,3 @@
 /llvm/branches/Apple/Pertwee:110850,110961
 /llvm/branches/type-system-rewrite:133420-134817
-/llvm/trunk:155241,275868-275870,275879,275898,275928,275935,275946,275978,275981,276015,276051,276077,276109,276119,276181,276209,276236-276237,276358,276364,276368,276389,276435,276438,276479,276510,276648,276676,276712,276740,276823,276956,276980,277093,277114,277135,277371,277399,277500,277504,277625,277691,277693,277773,278002,278086,278133,278157,278343,278370,278413,278558-278559,278562,278569,278571,278573,278575,278584,278841,278900,278938,278999,279125,279268,279369,279647,280837,281957,283129
+/llvm/trunk:155241,275868-275870,275879,275898,275928,275935,275946,275978,275981,276015,276051,276077,276109,276119,276181,276209,276236-276237,276358,276364,276368,276389,276435,276438,276479,276510,276648,276676,276712,276740,276823,276956,276980,277093,277114,277135,277371,277399,277500,277504,277625,277691,277693,277773,278002,278086,278133,278157,278343,278370,278413,278558-278559,278562,278569,278571,278573,278575,278584,278841,278900,278938,278999,279125,279268,279369,279647,280837,281957,282613,283129

Modified: llvm/branches/release_39/test/CodeGen/X86/avx-vbroadcast.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/release_39/test/CodeGen/X86/avx-vbroadcast.ll?rev=286248&r1=286247&r2=286248&view=diff
==============================================================================
--- llvm/branches/release_39/test/CodeGen/X86/avx-vbroadcast.ll (original)
+++ llvm/branches/release_39/test/CodeGen/X86/avx-vbroadcast.ll Tue Nov  8 10:45:26 2016
@@ -546,3 +546,62 @@ define <4 x double> @splat_concat4(doubl
   %6 = shufflevector <2 x double> %3, <2 x double> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   ret <4 x double> %6
 }
+
+;
+; FIXME: When VBROADCAST replaces an existing load, ensure it still respects lifetime dependencies.
+;
+define float @broadcast_lifetime() nounwind {
+; X32-LABEL: broadcast_lifetime:
+; X32:       ## BB#0:
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    subl $40, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, (%esp)
+; X32-NEXT:    calll _gfunc
+; X32-NEXT:    movl %esi, (%esp)
+; X32-NEXT:    calll _gfunc
+; X32-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %xmm0
+; X32-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %xmm1
+; X32-NEXT:    vsubss %xmm0, %xmm1, %xmm0
+; X32-NEXT:    vmovss %xmm0, {{[0-9]+}}(%esp)
+; X32-NEXT:    flds {{[0-9]+}}(%esp)
+; X32-NEXT:    addl $40, %esp
+; X32-NEXT:    popl %esi
+; X32-NEXT:    retl
+;
+; X64-LABEL: broadcast_lifetime:
+; X64:       ## BB#0:
+; X64-NEXT:    subq $24, %rsp
+; X64-NEXT:    leaq (%rsp), %rdi
+; X64-NEXT:    callq _gfunc
+; X64-NEXT:    leaq (%rsp), %rdi
+; X64-NEXT:    callq _gfunc
+; X64-NEXT:    vbroadcastss {{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    vbroadcastss {{[0-9]+}}(%rsp), %xmm1
+; X64-NEXT:    vsubss %xmm0, %xmm1, %xmm0
+; X64-NEXT:    addq $24, %rsp
+; X64-NEXT:    retq
+  %1 = alloca <4 x float>, align 16
+  %2 = alloca <4 x float>, align 16
+  %3 = bitcast <4 x float>* %1 to i8*
+  %4 = bitcast <4 x float>* %2 to i8*
+
+  call void @llvm.lifetime.start(i64 16, i8* %3)
+  call void @gfunc(<4 x float>* %1)
+  %5 = load <4 x float>, <4 x float>* %1, align 16
+  call void @llvm.lifetime.end(i64 16, i8* %3)
+
+  call void @llvm.lifetime.start(i64 16, i8* %4)
+  call void @gfunc(<4 x float>* %2)
+  %6 = load <4 x float>, <4 x float>* %2, align 16
+  call void @llvm.lifetime.end(i64 16, i8* %4)
+
+  %7 = extractelement <4 x float> %5, i32 1
+  %8 = extractelement <4 x float> %6, i32 1
+  %9 = fsub float %8, %7
+  ret float %9
+}
+
+declare void @gfunc(<4 x float>*)
+declare void @llvm.lifetime.start(i64, i8*)
+declare void @llvm.lifetime.end(i64, i8*)