Skip to content

First draft: [Clang] Add constant evaluation support for x86 psadbw/p… #169253

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
+74 −0
Draft

First draft: [Clang] Add constant evaluation support for x86 psadbw/p… #169253

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
53 changes: 53 additions & 0 deletions clang/lib/AST/ExprConstant.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1901,6 +1901,8 @@ static bool EvaluateFixedPointOrInteger(const Expr *E, APFixedPoint &Result,
static bool EvaluateFixedPoint(const Expr *E, APFixedPoint &Result,
EvalInfo &Info);

static bool EvaluatePSADBW128(const CallExpr *E, EvalInfo &Info, APValue &Result);

//===----------------------------------------------------------------------===//
// Misc utilities
//===----------------------------------------------------------------------===//
Expand Down Expand Up @@ -12078,6 +12080,54 @@ static bool evalPackBuiltin(const CallExpr *E, EvalInfo &Info, APValue &Result,
return true;
}

static bool EvaluatePSADBW128(const CallExpr *E, EvalInfo &Info, APValue &Result) {
// 1) Evaluate the arguments into APValues
APValue A, B;
if (!Evaluate(A, Info, E->getArg(0)) ||
!Evaluate(B, Info, E->getArg(1)))
return false;

if (!A.isVector() || !B.isVector())
Copy link
Collaborator

@RKSimon RKSimon Nov 25, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The ia32_psadbw builtins have fixed types - don't bother with so much isVector/getVectorLength checks - assert((getVectorLength() % 16) == 0) should be enough.

return false;

unsigned Len = A.getVectorLength();
if (Len != 16) // psadbw128 uses 16 bytes (2 × 8)
return false;

// 2) Compute SAD over two 8-byte blocks
uint64_t Sum0 = 0;
uint64_t Sum1 = 0;

// bytes 0..7
for (unsigned i = 0; i < 8; ++i) {
uint64_t a = A.getVectorElt(i).getInt().getZExtValue();
uint64_t b = B.getVectorElt(i).getInt().getZExtValue();
Sum0 += (a > b ? a - b : b - a);
}
Copy link
Collaborator

@RKSimon RKSimon Nov 25, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Use APIntOps::abdu ?


// bytes 8..15
for (unsigned i = 8; i < 16; ++i) {
uint64_t a = A.getVectorElt(i).getInt().getZExtValue();
uint64_t b = B.getVectorElt(i).getInt().getZExtValue();
Sum1 += (a > b ? a - b : b - a);
}
Copy link
Collaborator

@RKSimon RKSimon Nov 25, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Convert to a format that will handle 256/512 variants now - otherwise you'll just end up rewriting all of this later on.

SmallVector<APValue, 8> Elts;
for (unsigned Lane = 0; Lane != Len; Lane += 8) {
  for (unsigend I = 0; I != 8; ++I) {
    APInt A = A.getVectorElt(Lane + I).getInt();
    APInt B = B.getVectorElt(Lane + I).getInt();
  }
etc.
  Elts.emplace_back(APValue(APSInt(APInt(64, Sum), Unsigned)));
}


// 3) Build result vector of two 64-bit elements
SmallVector<APValue, 2> Elts;
QualType ElemTy = E->getType()->castAs<VectorType>()->getElementType();
bool Unsigned = ElemTy->isUnsignedIntegerType();
unsigned BW = Info.Ctx.getIntWidth(ElemTy); // usually 64

Elts.emplace_back(APValue(APSInt(APInt(BW, Sum0), Unsigned)));
Elts.emplace_back(APValue(APSInt(APInt(BW, Sum1), Unsigned)));

// APValue(const APValue *E, unsigned N) – copies the elements
Result = APValue(Elts.data(), Elts.size());

return true;
}


static bool evalShuffleGeneric(
EvalInfo &Info, const CallExpr *Call, APValue &Out,
llvm::function_ref<std::pair<unsigned, int>(unsigned, unsigned)>
Expand Down Expand Up @@ -12342,6 +12392,9 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
case clang::X86::BI__builtin_ia32_pavgw512:
return EvaluateBinOpExpr(llvm::APIntOps::avgCeilU);

case X86::BI__builtin_ia32_psadbw128:
return EvaluatePSADBW128(E, Info, Result);

case clang::X86::BI__builtin_ia32_pmulhrsw128:
case clang::X86::BI__builtin_ia32_pmulhrsw256:
case clang::X86::BI__builtin_ia32_pmulhrsw512:
Expand Down
21 changes: 21 additions & 0 deletions clang/test/AST/ByteCode/x86-psadbw-psadbw128.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,21 @@
// RUN: %clang_cc1 -std=c++20 -triple x86_64-unknown-linux-gnu -O0 \
// RUN: -fexperimental-new-constant-interpreter -verify %s

// No headers allowed. Use Clang's vector types directly.

typedef char v16qi __attribute__((vector_size(16)));
typedef long long v2di __attribute__((vector_size(16)));

constexpr v2di test_psadbw128() {
v16qi a = {10,20,30,40,50,60,70,80,
1,2,3,4,5,6,7,8};

v16qi b = {5,15,25,45,55,55,75,85,
10,0,3,9,1,10,2,8};

// Call the builtin directly
return __builtin_ia32_psadbw128(a, b);
}

static_assert(test_psadbw128()[0] == 40, "block0 mismatch");
static_assert(test_psadbw128()[1] == 29, "block1 mismatch");
Loading