r186490 - Propagate alignment for _Complex
JF Bastien
jfb at google.com
Wed Jul 17 17:45:44 PDT 2013
The issue seems to be that i386 sets double alignment at 32-bits, and
that's used in ASTContext.cpp (case Type::Complex) to determine a _Complex
double's alignment. I'm not sure if that's the right thing to do: the C
standard says "alignment requirements as an array type containing exactly
two elements of the corresponding real type", and I can't figure out if
arrays have extra alignment requirements compared to their base type.
A simple workaround is to run the test with a specific triple, say
x86-64-unknown-unknown. Should I commit such a change, until the above can
be resolved?
On Wed, Jul 17, 2013 at 5:05 PM, Quentin Colombet <qcolombet at apple.com>wrote:
> Hi jfb,
>
> This breaks a buildbot:
> http://lab.llvm.org:8013/builders/clang-i386-darwin11-RA/builds/1255
>
> ******************** TEST 'Clang :: CodeGen/volatile-complex.c' FAILED ********************
> Script:
> --
> /Users/buildslave/zorg/buildbot/smooshlab/slave-0.8/build.clang-i386-darwin11-RA/clang-build/Release+Asserts/bin/clang -cc1 -internal-isystem /Users/buildslave/zorg/buildbot/smooshlab/slave-0.8/build.clang-i386-darwin11-RA/clang-build/Release+Asserts/bin/../lib/clang/3.4/include -emit-llvm /Users/buildslave/zorg/buildbot/smooshlab/slave-0.8/build.clang-i386-darwin11-RA/clang.src/test/CodeGen/volatile-complex.c -o - | FileCheck /Users/buildslave/zorg/buildbot/smooshlab/slave-0.8/build.clang-i386-darwin11-RA/clang.src/test/CodeGen/volatile-complex.c
> --
> Exit Code: 1
> Command Output (stderr):
> --
> /Users/buildslave/zorg/buildbot/smooshlab/slave-0.8/build.clang-i386-darwin11-RA/clang.src/test/CodeGen/volatile-complex.c:25:17: error: expected string not found in input
> // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 1), align 8
> ^
> <stdin>:26:2: note: scanning from here
> %cd.imag = load volatile double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 1), align 4
> ^
> <stdin>:26:13: note: possible intended match here
> %cd.imag = load volatile double* getelementptr inbounds ({ double, double }* @cd, i32 0, i32 1), align 4
> ^
> --
>
> ********************
>
> Could you have a look please?
>
> Thanks,
>
> -Quentin
>
> On Jul 16, 2013, at 10:57 PM, JF Bastien <jfb at google.com> wrote:
>
> Author: jfb
> Date: Wed Jul 17 00:57:42 2013
> New Revision: 186490
>
> URL: http://llvm.org/viewvc/llvm-project?rev=186490&view=rev
> Log:
> Propagate alignment for _Complex
>
> _Complex load/store didn't have their alignment set properly, which was
> visible when GCC's torture tests use volatile _Complex.
>
> Update some existing tests to check for alignment, and add a new test
> which also has over-aligned volatile _Complex (since the imaginary part
> shouldn't be overaligned, only the real part).
>
> Added:
> cfe/trunk/test/CodeGen/volatile-complex.c (with props)
> Modified:
> cfe/trunk/lib/CodeGen/CGExprComplex.cpp
> cfe/trunk/test/CodeGen/volatile-1.c
> cfe/trunk/test/CodeGen/volatile-2.c
>
> Modified: cfe/trunk/lib/CodeGen/CGExprComplex.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGExprComplex.cpp?rev=186490&r1=186489&r2=186490&view=diff
>
> ==============================================================================
> --- cfe/trunk/lib/CodeGen/CGExprComplex.cpp (original)
> +++ cfe/trunk/lib/CodeGen/CGExprComplex.cpp Wed Jul 17 00:57:42 2013
> @@ -18,6 +18,7 @@
> #include "llvm/ADT/SmallString.h"
> #include "llvm/IR/Constants.h"
> #include "llvm/IR/Function.h"
> +#include <algorithm>
> using namespace clang;
> using namespace CodeGen;
>
> @@ -297,19 +298,26 @@ ComplexPairTy ComplexExprEmitter::EmitLo
>
> llvm::Value *SrcPtr = lvalue.getAddress();
> bool isVolatile = lvalue.isVolatileQualified();
> + unsigned AlignR = lvalue.getAlignment().getQuantity();
> + ASTContext &C = CGF.getContext();
> + QualType ComplexTy = lvalue.getType();
> + unsigned ComplexAlign = C.getTypeAlignInChars(ComplexTy).getQuantity();
> + unsigned AlignI = std::min(AlignR, ComplexAlign);
>
> llvm::Value *Real=0, *Imag=0;
>
> if (!IgnoreReal || isVolatile) {
> llvm::Value *RealP = Builder.CreateStructGEP(SrcPtr, 0,
> SrcPtr->getName() +
> ".realp");
> - Real = Builder.CreateLoad(RealP, isVolatile, SrcPtr->getName() +
> ".real");
> + Real = Builder.CreateAlignedLoad(RealP, AlignR, isVolatile,
> + SrcPtr->getName() + ".real");
> }
>
> if (!IgnoreImag || isVolatile) {
> llvm::Value *ImagP = Builder.CreateStructGEP(SrcPtr, 1,
> SrcPtr->getName() +
> ".imagp");
> - Imag = Builder.CreateLoad(ImagP, isVolatile, SrcPtr->getName() +
> ".imag");
> + Imag = Builder.CreateAlignedLoad(ImagP, AlignI, isVolatile,
> + SrcPtr->getName() + ".imag");
> }
> return ComplexPairTy(Real, Imag);
> }
> @@ -325,10 +333,16 @@ void ComplexExprEmitter::EmitStoreOfComp
> llvm::Value *Ptr = lvalue.getAddress();
> llvm::Value *RealPtr = Builder.CreateStructGEP(Ptr, 0, "real");
> llvm::Value *ImagPtr = Builder.CreateStructGEP(Ptr, 1, "imag");
> -
> - // TODO: alignment
> - Builder.CreateStore(Val.first, RealPtr, lvalue.isVolatileQualified());
> - Builder.CreateStore(Val.second, ImagPtr, lvalue.isVolatileQualified());
> + unsigned AlignR = lvalue.getAlignment().getQuantity();
> + ASTContext &C = CGF.getContext();
> + QualType ComplexTy = lvalue.getType();
> + unsigned ComplexAlign = C.getTypeAlignInChars(ComplexTy).getQuantity();
> + unsigned AlignI = std::min(AlignR, ComplexAlign);
> +
> + Builder.CreateAlignedStore(Val.first, RealPtr, AlignR,
> + lvalue.isVolatileQualified());
> + Builder.CreateAlignedStore(Val.second, ImagPtr, AlignI,
> + lvalue.isVolatileQualified());
> }
>
>
>
> Modified: cfe/trunk/test/CodeGen/volatile-1.c
> URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/volatile-1.c?rev=186490&r1=186489&r2=186490&view=diff
>
> ==============================================================================
> --- cfe/trunk/test/CodeGen/volatile-1.c (original)
> +++ cfe/trunk/test/CodeGen/volatile-1.c Wed Jul 17 00:57:42 2013
> @@ -26,45 +26,45 @@ int printf(const char *, ...);
> void test() {
> // CHECK: load volatile [[INT]]* @i
> i;
> - // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]*
> @ci, i32 0, i32 0)
> - // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]*
> @ci, i32 0, i32 1)
> + // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]*
> @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]*
> @ci, i32 0, i32 1), align 4
> // CHECK-NEXT: sitofp [[INT]]
> (float)(ci);
> - // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]*
> @ci, i32 0, i32 0)
> - // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]*
> @ci, i32 0, i32 1)
> + // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]*
> @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: load volatile [[INT]]* getelementptr inbounds ([[CINT]]*
> @ci, i32 0, i32 1), align 4
> (void)ci;
> // CHECK-NEXT: bitcast
> // CHECK-NEXT: memcpy
> (void)a;
> - // CHECK-NEXT: [[R:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: [[I:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> - // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> + // CHECK-NEXT: [[R:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: [[I:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> + // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> (void)(ci=ci);
> // CHECK-NEXT: [[T:%.*]] = load volatile [[INT]]* @j
> // CHECK-NEXT: store volatile [[INT]] [[T]], [[INT]]* @i
> (void)(i=j);
> - // CHECK-NEXT: [[R1:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: [[I1:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> - // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> + // CHECK-NEXT: [[R1:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: [[I1:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> + // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> // Not sure why they're ordered this way.
> // CHECK-NEXT: [[R:%.*]] = add [[INT]] [[R2]], [[R1]]
> // CHECK-NEXT: [[I:%.*]] = add [[INT]] [[I2]], [[I1]]
> - // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> + // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> ci+=ci;
>
> - // CHECK-NEXT: [[R1:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: [[I1:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> - // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> + // CHECK-NEXT: [[R1:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: [[I1:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> + // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> // CHECK-NEXT: [[R:%.*]] = add [[INT]] [[R2]], [[R1]]
> // CHECK-NEXT: [[I:%.*]] = add [[INT]] [[I2]], [[I1]]
> - // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> - // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0)
> - // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1)
> + // CHECK-NEXT: store volatile [[INT]] [[R]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: store volatile [[INT]] [[I]], [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> + // CHECK-NEXT: [[R2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 0), align 4
> + // CHECK-NEXT: [[I2:%.*]] = load volatile [[INT]]* getelementptr
> inbounds ([[CINT]]* @ci, i32 0, i32 1), align 4
> // These additions can be elided
> // CHECK-NEXT: add [[INT]] [[R]], [[R2]]
> // CHECK-NEXT: add [[INT]] [[I]], [[I2]]
>
> Modified: cfe/trunk/test/CodeGen/volatile-2.c
> URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/volatile-2.c?rev=186490&r1=186489&r2=186490&view=diff
>
> ==============================================================================
> --- cfe/trunk/test/CodeGen/volatile-2.c (original)
> +++ cfe/trunk/test/CodeGen/volatile-2.c Wed Jul 17 00:57:42 2013
> @@ -3,8 +3,8 @@
> void test0() {
> // CHECK: define void @test0()
> // CHECK: [[F:%.*]] = alloca float
> - // CHECK-NEXT: [[REAL:%.*]] = load volatile float* getelementptr
> inbounds ({ float, float }* @test0_v, i32 0, i32 0)
> - // CHECK-NEXT: load volatile float* getelementptr inbounds ({{.*}}
> @test0_v, i32 0, i32 1)
> + // CHECK-NEXT: [[REAL:%.*]] = load volatile float* getelementptr
> inbounds ({ float, float }* @test0_v, i32 0, i32 0), align 4
> + // CHECK-NEXT: load volatile float* getelementptr inbounds ({{.*}}
> @test0_v, i32 0, i32 1), align 4
> // CHECK-NEXT: store float [[REAL]], float* [[F]], align 4
> // CHECK-NEXT: ret void
> extern volatile _Complex float test0_v;
> @@ -13,10 +13,10 @@ void test0() {
>
> void test1() {
> // CHECK: define void @test1()
> - // CHECK: [[REAL:%.*]] = load volatile float* getelementptr
> inbounds ({{.*}} @test1_v, i32 0, i32 0)
> - // CHECK-NEXT: [[IMAG:%.*]] = load volatile float* getelementptr
> inbounds ({{.*}} @test1_v, i32 0, i32 1)
> - // CHECK-NEXT: store volatile float [[REAL]], float* getelementptr
> inbounds ({{.*}} @test1_v, i32 0, i32 0)
> - // CHECK-NEXT: store volatile float [[IMAG]], float* getelementptr
> inbounds ({{.*}} @test1_v, i32 0, i32 1)
> + // CHECK: [[REAL:%.*]] = load volatile float* getelementptr
> inbounds ({{.*}} @test1_v, i32 0, i32 0), align 4
> + // CHECK-NEXT: [[IMAG:%.*]] = load volatile float* getelementptr
> inbounds ({{.*}} @test1_v, i32 0, i32 1), align 4
> + // CHECK-NEXT: store volatile float [[REAL]], float* getelementptr
> inbounds ({{.*}} @test1_v, i32 0, i32 0), align 4
> + // CHECK-NEXT: store volatile float [[IMAG]], float* getelementptr
> inbounds ({{.*}} @test1_v, i32 0, i32 1), align 4
> // CHECK-NEXT: ret void
> extern volatile _Complex float test1_v;
> test1_v = test1_v;
>
> Added: cfe/trunk/test/CodeGen/volatile-complex.c
> URL:
> http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/volatile-complex.c?rev=186490&view=auto
>
> ==============================================================================
> --- cfe/trunk/test/CodeGen/volatile-complex.c (added)
> +++ cfe/trunk/test/CodeGen/volatile-complex.c Wed Jul 17 00:57:42 2013
> @@ -0,0 +1,63 @@
> +// RUN: %clang_cc1 -emit-llvm %s -o - | FileCheck %s
> +
> +volatile _Complex float cf;
> +volatile _Complex double cd;
> +volatile _Complex float cf32 __attribute__((aligned(32)));
> +volatile _Complex double cd32 __attribute__((aligned(32)));
> +
> +
> +// CHECK: define void @test_cf()
> +// CHECK-NEXT: entry:
> +void test_cf() {
> + // CHECK-NEXT: load volatile float* getelementptr inbounds ({ float,
> float }* @cf, i32 0, i32 0), align 4
> + // CHECK-NEXT: load volatile float* getelementptr inbounds ({ float,
> float }* @cf, i32 0, i32 1), align 4
> + (void)(cf);
> + // CHECK-NEXT: [[R:%.*]] = load volatile float* getelementptr inbounds
> ({ float, float }* @cf, i32 0, i32 0), align 4
> + // CHECK-NEXT: [[I:%.*]] = load volatile float* getelementptr inbounds
> ({ float, float }* @cf, i32 0, i32 1), align 4
> + // CHECK-NEXT: store volatile float [[R]], float* getelementptr
> inbounds ({ float, float }* @cf, i32 0, i32 0), align 4
> + // CHECK-NEXT: store volatile float [[I]], float* getelementptr
> inbounds ({ float, float }* @cf, i32 0, i32 1), align 4
> + (void)(cf=cf);
> + // CHECK-NEXT: ret void
> +}
> +
> +// CHECK: define void @test_cd()
> +// CHECK-NEXT: entry:
> +void test_cd() {
> + // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double,
> double }* @cd, i32 0, i32 0), align 8
> + // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double,
> double }* @cd, i32 0, i32 1), align 8
> + (void)(cd);
> + // CHECK-NEXT: [[R:%.*]] = load volatile double* getelementptr inbounds
> ({ double, double }* @cd, i32 0, i32 0), align 8
> + // CHECK-NEXT: [[I:%.*]] = load volatile double* getelementptr inbounds
> ({ double, double }* @cd, i32 0, i32 1), align 8
> + // CHECK-NEXT: store volatile double [[R]], double* getelementptr
> inbounds ({ double, double }* @cd, i32 0, i32 0), align 8
> + // CHECK-NEXT: store volatile double [[I]], double* getelementptr
> inbounds ({ double, double }* @cd, i32 0, i32 1), align 8
> + (void)(cd=cd);
> + // CHECK-NEXT: ret void
> +}
> +
> +// CHECK: define void @test_cf32()
> +// CHECK-NEXT: entry:
> +void test_cf32() {
> + // CHECK-NEXT: load volatile float* getelementptr inbounds ({ float,
> float }* @cf32, i32 0, i32 0), align 32
> + // CHECK-NEXT: load volatile float* getelementptr inbounds ({ float,
> float }* @cf32, i32 0, i32 1), align 4
> + (void)(cf32);
> + // CHECK-NEXT: [[R:%.*]] = load volatile float* getelementptr inbounds
> ({ float, float }* @cf32, i32 0, i32 0), align 32
> + // CHECK-NEXT: [[I:%.*]] = load volatile float* getelementptr inbounds
> ({ float, float }* @cf32, i32 0, i32 1), align 4
> + // CHECK-NEXT: store volatile float [[R]], float* getelementptr
> inbounds ({ float, float }* @cf32, i32 0, i32 0), align 32
> + // CHECK-NEXT: store volatile float [[I]], float* getelementptr
> inbounds ({ float, float }* @cf32, i32 0, i32 1), align 4
> + (void)(cf32=cf32);
> + // CHECK-NEXT: ret void
> +}
> +
> +// CHECK: define void @test_cd32()
> +// CHECK-NEXT: entry:
> +void test_cd32() {
> + // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double,
> double }* @cd32, i32 0, i32 0), align 32
> + // CHECK-NEXT: load volatile double* getelementptr inbounds ({ double,
> double }* @cd32, i32 0, i32 1), align 8
> + (void)(cd32);
> + // CHECK-NEXT: [[R:%.*]] = load volatile double* getelementptr inbounds
> ({ double, double }* @cd32, i32 0, i32 0), align 32
> + // CHECK-NEXT: [[I:%.*]] = load volatile double* getelementptr inbounds
> ({ double, double }* @cd32, i32 0, i32 1), align 8
> + // CHECK-NEXT: store volatile double [[R]], double* getelementptr
> inbounds ({ double, double }* @cd32, i32 0, i32 0), align 32
> + // CHECK-NEXT: store volatile double [[I]], double* getelementptr
> inbounds ({ double, double }* @cd32, i32 0, i32 1), align 8
> + (void)(cd32=cd32);
> + // CHECK-NEXT: ret void
> +}
>
> Propchange: cfe/trunk/test/CodeGen/volatile-complex.c
>
> ------------------------------------------------------------------------------
> svn:eol-style = LF
>
>
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130717/4e6b2cdf/attachment.html>
More information about the cfe-commits
mailing list