Skip to content

Commit

Permalink
[SingleSource/Vectorizer] Add unit tests for FindLastIV pattern. (#193)
Browse files Browse the repository at this point in the history 
This patch adds runtime test case for vectorization of FindLastIV reduction idiom:

  int32_t Rdx = -1;
  for (int32_t I = 0; I < TC; I++) {
    Rdx = A[I] > B[I] ? I : Rdx;
  }
  return Rdx;

Improving test coverage for llvm/llvm-project#67812 and llvm/llvm-project#120395.
  • Loading branch information
@Mel-Chen
Mel-Chen authored Jan 23, 2025
1 parent 72993e7 commit d694427
Show file tree
Hide file tree
Showing 3 changed files with 416 additions and 0 deletions.
8 changes: 8 additions & 0 deletions SingleSource/UnitTests/Vectorizer/common.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,14 @@
Init _Pragma("clang loop vectorize(enable)") Loop \
};

#define DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(Init, Loop, Type) \
auto ScalarFn = [](auto *A, auto *B, Type TC) -> Type { \
Init _Pragma("clang loop vectorize(disable) interleave_count(1)") Loop \
}; \
auto VectorFn = [](auto *A, auto *B, Type TC) -> Type { \
Init _Pragma("clang loop vectorize(enable)") Loop \
};

#define DEFINE_SCALAR_AND_VECTOR_FN3(Loop) \
auto ScalarFn = [](auto *A, auto *B, auto *C, unsigned TC) { \
_Pragma("clang loop vectorize(disable) interleave_count(1)") Loop \
Expand Down
380 changes: 380 additions & 0 deletions SingleSource/UnitTests/Vectorizer/find-last.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,380 @@
#include <algorithm>
#include <functional>
#include <iostream>
#include <limits>
#include <memory>
#include <stdint.h>

#include "common.h"

template <typename RetTy, typename Ty>
using Fn2Ty = std::function<RetTy(Ty *, Ty *, RetTy)>;
template <typename RetTy, typename Ty>
static void checkVectorFunction(Fn2Ty<RetTy, Ty> ScalarFn,
Fn2Ty<RetTy, Ty> VectorFn, const char *Name) {
std::cout << "Checking " << Name << "\n";

unsigned N = 1000;
std::unique_ptr<Ty[]> Src1(new Ty[N]);
std::unique_ptr<Ty[]> Src2(new Ty[N]);
init_data(Src1, N);
init_data(Src2, N);

// Test VectorFn with different input data.
{
// Check with random inputs.
auto Reference = ScalarFn(&Src1[0], &Src2[0], N);
auto ToCheck = VectorFn(&Src1[0], &Src2[0], N);
if (Reference != ToCheck) {
std::cerr << "Miscompare\n";
exit(1);
}
}

{
// Check with Src1 > Src2 for all elements.
for (unsigned I = 0; I != N; ++I) {
Src1[I] = std::numeric_limits<Ty>::max();
Src2[I] = std::numeric_limits<Ty>::min();
}
auto Reference = ScalarFn(&Src1[0], &Src2[0], N);
auto ToCheck = VectorFn(&Src1[0], &Src2[0], N);
if (Reference != ToCheck) {
std::cerr << "Miscompare\n";
exit(1);
}
}

{
// Check with Src1 < Src2 for all elements.
for (unsigned I = 0; I != N; ++I) {
Src1[I] = std::numeric_limits<Ty>::min();
Src2[I] = std::numeric_limits<Ty>::max();
}
auto Reference = ScalarFn(&Src1[0], &Src2[0], N);
auto ToCheck = VectorFn(&Src1[0], &Src2[0], N);
if (Reference != ToCheck) {
std::cerr << "Miscompare\n";
exit(1);
}
}

{
// Check with only Src1[998] > Src2[998].
for (unsigned I = 0; I != N; ++I)
Src1[I] = Src2[I] = std::numeric_limits<Ty>::min();
Src1[998] = std::numeric_limits<Ty>::max();
auto Reference = ScalarFn(&Src1[0], &Src2[0], N);
auto ToCheck = VectorFn(&Src1[0], &Src2[0], N);
if (Reference != ToCheck) {
std::cerr << "Miscompare\n";
exit(1);
}
}

{
// Check with only Src1[0] > Src2[0].
for (unsigned I = 0; I != N; ++I)
Src1[I] = Src2[I] = std::numeric_limits<Ty>::min();
Src1[0] = std::numeric_limits<Ty>::max();
auto Reference = ScalarFn(&Src1[0], &Src2[0], N);
auto ToCheck = VectorFn(&Src1[0], &Src2[0], N);
if (Reference != ToCheck) {
std::cerr << "Miscompare\n";
exit(1);
}
}

{
// Check with only Src1[N - 1] > Src2[N - 1].
for (unsigned I = 0; I != N; ++I)
Src1[I] = Src2[I] = std::numeric_limits<Ty>::min();
Src1[N - 1] = std::numeric_limits<Ty>::max();
auto Reference = ScalarFn(&Src1[0], &Src2[0], N);
auto ToCheck = VectorFn(&Src1[0], &Src2[0], N);
if (Reference != ToCheck) {
std::cerr << "Miscompare\n";
exit(1);
}
}

{
// Check with only Src1[0] > Src2[0] and Src1[N - 1] > Src2[N - 1].
for (unsigned I = 0; I != N; ++I)
Src1[I] = Src2[I] = std::numeric_limits<Ty>::min();
Src1[0] = Src1[N - 1] = std::numeric_limits<Ty>::max();
auto Reference = ScalarFn(&Src1[0], &Src2[0], N);
auto ToCheck = VectorFn(&Src1[0], &Src2[0], N);
if (Reference != ToCheck) {
std::cerr << "Miscompare\n";
exit(1);
}
}
}

int main(void) {
rng = std::mt19937(15);

#define INC_COND(Start, Step, RetTy) for (RetTy I = Start; I < TC; I += Step)
#define DEC_COND(End, Step, RetTy) for (RetTy I = TC; I > End; I -= Step)

#define DEFINE_FINDLAST_LOOP_BODY(TrueVal, FalseVal, ForCond) \
ForCond { Rdx = A[I] > B[I] ? TrueVal : FalseVal; } \
return Rdx;

{
// Find the last index where A[I] > B[I] and update 32-bits Rdx when the
// condition is true.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 0, /* Step= */ 1, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(ScalarFn, VectorFn,
"findlast_icmp_s32_true_update");
checkVectorFunction<int32_t, float>(ScalarFn, VectorFn,
"findlast_fcmp_s32_true_update");
}

{
// Find the last index where A[I] > B[I] and update 16-bits Rdx when the
// condition is true.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 0, /* Step= */ 1, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(ScalarFn, VectorFn,
"findlast_icmp_s16_true_update");
}

{
// Update 32-bits Rdx when the condition A[I] > B[I] is false.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ Rdx, /* FalseVal= */ I,
/* ForCond= */
INC_COND(/* Start= */ 0, /* Step= */ 1, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(ScalarFn, VectorFn,
"findlast_icmp_s32_false_update");
checkVectorFunction<int32_t, float>(ScalarFn, VectorFn,
"findlast_fcmp_s32_false_update");
}

{
// Update 16-bits Rdx when the condition A[I] > B[I] is false.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ Rdx, /* FalseVal= */ I,
/* ForCond= */
INC_COND(/* Start= */ 0, /* Step= */ 1, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(ScalarFn, VectorFn,
"findlast_icmp_s16_false_update");
}

{
// Find the last 32-bits index with the start value TC.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = TC;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 0, /* Step= */ 1, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(ScalarFn, VectorFn,
"findlast_icmp_s32_start_TC");
checkVectorFunction<int32_t, float>(ScalarFn, VectorFn,
"findlast_fcmp_s32_start_TC");
}

{
// Find the last 16-bits index with the start value TC.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = TC;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 0, /* Step= */ 1, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(ScalarFn, VectorFn,
"findlast_icmp_s16_start_TC");
}

{
// Increment the 32-bits induction variable by 2.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 0, /* Step= */ 2, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(ScalarFn, VectorFn,
"findlast_icmp_s32_inc_2");
checkVectorFunction<int32_t, float>(ScalarFn, VectorFn,
"findlast_fcmp_s32_inc_2");
}

{
// Increment the 16-bits induction variable by 2.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 0, /* Step= */ 2, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(ScalarFn, VectorFn,
"findlast_icmp_s16_inc_2");
}

{
// Check with decreasing 32-bits induction variable.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
DEC_COND(/* End= */ 0, /* Step= */ 1, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(
ScalarFn, VectorFn, "findlast_icmp_s32_start_decreasing_induction");
checkVectorFunction<int32_t, float>(
ScalarFn, VectorFn, "findlast_fcmp_s32_start_decreasing_induction");
}

{
// Check with decreasing 16-bits induction variable.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
DEC_COND(/* End= */ 0, /* Step= */ 1, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(
ScalarFn, VectorFn, "findlast_icmp_s16_start_decreasing_induction");
}

{
// Check with 32-bits the induction variable starts from 3.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 3, /* Step= */ 1, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(ScalarFn, VectorFn,
"findlast_icmp_s32_iv_start_3");
checkVectorFunction<int32_t, float>(ScalarFn, VectorFn,
"findlast_fcmp_s32_iv_start_3");
}

{
// Check with 16-bits the induction variable starts from 3.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = -1;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 3, /* Step= */ 1, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(ScalarFn, VectorFn,
"findlast_icmp_s16_iv_start_3");
}

{
// Check with start value of 3 and 32-bits induction variable starts at 3.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = 3;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 3, /* Step= */ 1, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(
ScalarFn, VectorFn, "findlast_icmp_s32_start_3_iv_start_3");
checkVectorFunction<int32_t, float>(ScalarFn, VectorFn,
"findlast_fcmp_s32_start_3_iv_start_3");
}

{
// Check with start value of 3 and 16-bits induction variable starts at 3.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = 3;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 3, /* Step= */ 1, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(
ScalarFn, VectorFn, "findlast_icmp_s16_start_3_iv_start_3");
}

{
// Check with start value of 2 and 32-bits induction variable starts at 3.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = 2;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 3, /* Step= */ 1, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(
ScalarFn, VectorFn, "findlast_icmp_s32_start_2_iv_start_3");
checkVectorFunction<int32_t, float>(ScalarFn, VectorFn,
"findlast_fcmp_s32_start_2_iv_start_3");
}

{
// Check with start value of 2 and 16-bits induction variable starts at 3.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = 2;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 3, /* Step= */ 1, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(
ScalarFn, VectorFn, "findlast_icmp_s16_start_2_iv_start_3");
}

{
// Check with start value of 4 and 32-bits induction variable starts at 3.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int32_t Rdx = 4;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 3, /* Step= */ 1, /* RetTy= */ int32_t)),
int32_t);
checkVectorFunction<int32_t, int32_t>(
ScalarFn, VectorFn, "findlast_icmp_s32_start_4_iv_start_3");
checkVectorFunction<int32_t, float>(ScalarFn, VectorFn,
"findlast_fcmp_s32_start_4_iv_start_3");
}

{
// Check with start value of 4 and 16-bits induction variable starts at 3.
DEFINE_SCALAR_AND_VECTOR_FN2_TYPE(
int16_t Rdx = 4;,
DEFINE_FINDLAST_LOOP_BODY(
/* TrueVal= */ I, /* FalseVal= */ Rdx,
/* ForCond= */
INC_COND(/* Start= */ 3, /* Step= */ 1, /* RetTy= */ int16_t)),
int16_t);
checkVectorFunction<int16_t, int16_t>(
ScalarFn, VectorFn, "findlast_icmp_s16_start_4_iv_start_3");
}

return 0;
}
Loading

0 comments on commit d694427

Please sign in to comment.