NFC: Refactor library-specific mappings of scalar maths functions to their vector counterparts

[pjeeva01]

Summary

This patch factors out mappings of scalar maths functions to their vector counterparts from TargetLibraryInfo.cpp to a separate VecFuncs.def file. Such mappings are currently available for Accelerate framework, and SVML library.

This patch is a precursor to Initial support for vectorization using MASS (IBM MASS vector library).

[jsji]

llvm/include/llvm/Analysis/VecFuncs.def

14

Why we want to check TLI_DEFINE_VECDESCS to define TLI_DEFINE_VECFUNC?
Can't we just check #ifndef TLI_DEFINE_VECFUNC here?

175

This is not always defined, should we call #undef when we are sure they are defined?

[pjeeva01]

llvm/include/llvm/Analysis/VecFuncs.def

14

The plan is to define TLI_DEFINE_VECFUNC in a different way when a different identifier (eg: TLI_DEFINE_ARR) is defined.

https://reviews.llvm.org/D59881 shows the intention behind this structure.

If you prefer for me to restructure this ifdef only to suit this patch, I will go ahead and change it.

175

The #undef is ignored if the specified identifier is not currently defined as a macro name. So, it should be alright to leave the #undef as it is.

[jsji]

llvm/include/llvm/Analysis/VecFuncs.def

14

OK, I think you meant https://reviews.llvm.org/D59883, where you want to extract the function name only.

Why not check #ifdef TLI_DEFINE_VECFUNC here, then we can add something like

#ifdef TLI_DEFINE_MASSV_VECFUNCS_NAMES  
#define TLI_DEFINE_MASSV_VECFUNCS
#define TLI_DEFINE_VECFUNC(SCAL, VEC, VF) VEC,
#endif

before this to override the default,
then use #define TLI_DEFINE_MASSV_VECFUNCS_NAMES in PPCLowerMASSVEntries.cpp ?

Then we will only need to define one macro before the #include in all places.

175

OK. Some old compiler (or pre-processor) might error due to this, but consider this is now in C standard now. It is OK to leave it as it is.

[pjeeva01]

This updated patch uses a single macro to define the mapping from scalar maths functions to their vector counterparts ( as suggested in the earlier review of the original patch).

[jsji]

LGTM. Thanks for refactoring.

[nemanjai]

LGTM. Thanks for pulling out this change into a separate patch. I'll commit it for you on Monday. This way others have an extra opportunity to add any comments or concerns.

llvm/include/llvm/Analysis/VecFuncs.def

+//===-- VecFuncs.def - Library information -------------*- C++ -*-----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+// This .def file will create mappings from scalar math functions to vector
+// functions along with their vectorization factor. The current support includes
+// such mappings for Accelerate framework and SVML library. 
+
+#if !(defined(TLI_DEFINE_VECFUNC))
+#define TLI_DEFINE_VECFUNC(SCAL, VEC, VF) {SCAL, VEC, VF},
+#endif 
+
+#if defined(TLI_DEFINE_ACCELERATE_VECFUNCS)
+// Accelerate framework's Vector Functions
+
+// Floating-Point Arithmetic and Auxiliary Functions
+TLI_DEFINE_VECFUNC("ceilf", "vceilf", 4)
+TLI_DEFINE_VECFUNC("fabsf", "vfabsf", 4)
+TLI_DEFINE_VECFUNC("llvm.fabs.f32", "vfabsf", 4)
+TLI_DEFINE_VECFUNC("floorf", "vfloorf", 4)
+TLI_DEFINE_VECFUNC("sqrtf", "vsqrtf", 4)
+TLI_DEFINE_VECFUNC("llvm.sqrt.f32", "vsqrtf", 4)
+
+// Exponential and Logarithmic Functions
+TLI_DEFINE_VECFUNC("expf", "vexpf", 4)
+TLI_DEFINE_VECFUNC("llvm.exp.f32", "vexpf", 4)
+TLI_DEFINE_VECFUNC("expm1f", "vexpm1f", 4)
+TLI_DEFINE_VECFUNC("logf", "vlogf", 4)
+TLI_DEFINE_VECFUNC("llvm.log.f32", "vlogf", 4)
+TLI_DEFINE_VECFUNC("log1pf", "vlog1pf", 4)
+TLI_DEFINE_VECFUNC("log10f", "vlog10f", 4)
+TLI_DEFINE_VECFUNC("llvm.log10.f32", "vlog10f", 4)
+TLI_DEFINE_VECFUNC("logbf", "vlogbf", 4)
+
+// Trigonometric Functions
+TLI_DEFINE_VECFUNC("sinf", "vsinf", 4)
+TLI_DEFINE_VECFUNC("llvm.sin.f32", "vsinf", 4)
+TLI_DEFINE_VECFUNC("cosf", "vcosf", 4)
+TLI_DEFINE_VECFUNC("llvm.cos.f32", "vcosf", 4)
+TLI_DEFINE_VECFUNC("tanf", "vtanf", 4)
+TLI_DEFINE_VECFUNC("asinf", "vasinf", 4)
+TLI_DEFINE_VECFUNC("acosf", "vacosf", 4)
+TLI_DEFINE_VECFUNC("atanf", "vatanf", 4)
+
+// Hyperbolic Functions
+TLI_DEFINE_VECFUNC("sinhf", "vsinhf", 4)
+TLI_DEFINE_VECFUNC("coshf", "vcoshf", 4)
+TLI_DEFINE_VECFUNC("tanhf", "vtanhf", 4)
+TLI_DEFINE_VECFUNC("asinhf", "vasinhf", 4)
+TLI_DEFINE_VECFUNC("acoshf", "vacoshf", 4)
+TLI_DEFINE_VECFUNC("atanhf", "vatanhf", 4)
+
+
+#elif defined(TLI_DEFINE_SVML_VECFUNCS)
+// Intel SVM library's Vector Functions
+
+TLI_DEFINE_VECFUNC("sin", "__svml_sin2", 2)
+TLI_DEFINE_VECFUNC("sin", "__svml_sin4", 4)
+TLI_DEFINE_VECFUNC("sin", "__svml_sin8", 8)
+
+TLI_DEFINE_VECFUNC("sinf", "__svml_sinf4", 4)
+TLI_DEFINE_VECFUNC("sinf", "__svml_sinf8", 8)
+TLI_DEFINE_VECFUNC("sinf", "__svml_sinf16", 16)
+
+TLI_DEFINE_VECFUNC("llvm.sin.f64", "__svml_sin2", 2)
+TLI_DEFINE_VECFUNC("llvm.sin.f64", "__svml_sin4", 4)
+TLI_DEFINE_VECFUNC("llvm.sin.f64", "__svml_sin8", 8)
+
+TLI_DEFINE_VECFUNC("llvm.sin.f32", "__svml_sinf4", 4)
+TLI_DEFINE_VECFUNC("llvm.sin.f32", "__svml_sinf8", 8)
+TLI_DEFINE_VECFUNC("llvm.sin.f32", "__svml_sinf16", 16)
+
+TLI_DEFINE_VECFUNC("cos", "__svml_cos2", 2)
+TLI_DEFINE_VECFUNC("cos", "__svml_cos4", 4)
+TLI_DEFINE_VECFUNC("cos", "__svml_cos8", 8)
+
+TLI_DEFINE_VECFUNC("cosf", "__svml_cosf4", 4)
+TLI_DEFINE_VECFUNC("cosf", "__svml_cosf8", 8)
+TLI_DEFINE_VECFUNC("cosf", "__svml_cosf16", 16)
+
+TLI_DEFINE_VECFUNC("llvm.cos.f64", "__svml_cos2", 2)
+TLI_DEFINE_VECFUNC("llvm.cos.f64", "__svml_cos4", 4)
+TLI_DEFINE_VECFUNC("llvm.cos.f64", "__svml_cos8", 8)
+
+TLI_DEFINE_VECFUNC("llvm.cos.f32", "__svml_cosf4", 4)
+TLI_DEFINE_VECFUNC("llvm.cos.f32", "__svml_cosf8", 8)
+TLI_DEFINE_VECFUNC("llvm.cos.f32", "__svml_cosf16", 16)
+
+TLI_DEFINE_VECFUNC("pow", "__svml_pow2", 2)
+TLI_DEFINE_VECFUNC("pow", "__svml_pow4", 4)
+TLI_DEFINE_VECFUNC("pow", "__svml_pow8", 8)
+
+TLI_DEFINE_VECFUNC("powf", "__svml_powf4", 4)
+TLI_DEFINE_VECFUNC("powf", "__svml_powf8", 8)
+TLI_DEFINE_VECFUNC("powf", "__svml_powf16", 16)
+
+TLI_DEFINE_VECFUNC("__pow_finite", "__svml_pow2", 2)
+TLI_DEFINE_VECFUNC("__pow_finite", "__svml_pow4", 4)
+TLI_DEFINE_VECFUNC("__pow_finite", "__svml_pow8", 8)
+
+TLI_DEFINE_VECFUNC("__powf_finite", "__svml_powf4", 4)
+TLI_DEFINE_VECFUNC("__powf_finite", "__svml_powf8", 8)
+TLI_DEFINE_VECFUNC("__powf_finite", "__svml_powf16", 16)
+
+TLI_DEFINE_VECFUNC("llvm.pow.f64", "__svml_pow2", 2)
+TLI_DEFINE_VECFUNC("llvm.pow.f64", "__svml_pow4", 4)
+TLI_DEFINE_VECFUNC("llvm.pow.f64", "__svml_pow8", 8)
+
+TLI_DEFINE_VECFUNC("llvm.pow.f32", "__svml_powf4", 4)
+TLI_DEFINE_VECFUNC("llvm.pow.f32", "__svml_powf8", 8)
+TLI_DEFINE_VECFUNC("llvm.pow.f32", "__svml_powf16", 16)
+
+TLI_DEFINE_VECFUNC("exp", "__svml_exp2", 2)
+TLI_DEFINE_VECFUNC("exp", "__svml_exp4", 4)
+TLI_DEFINE_VECFUNC("exp", "__svml_exp8", 8)
+
+TLI_DEFINE_VECFUNC("expf", "__svml_expf4", 4)
+TLI_DEFINE_VECFUNC("expf", "__svml_expf8", 8)
+TLI_DEFINE_VECFUNC("expf", "__svml_expf16", 16)
+
+TLI_DEFINE_VECFUNC("__exp_finite", "__svml_exp2", 2)
+TLI_DEFINE_VECFUNC("__exp_finite", "__svml_exp4", 4)
+TLI_DEFINE_VECFUNC("__exp_finite", "__svml_exp8", 8)
+
+TLI_DEFINE_VECFUNC("__expf_finite", "__svml_expf4", 4)
+TLI_DEFINE_VECFUNC("__expf_finite", "__svml_expf8", 8)
+TLI_DEFINE_VECFUNC("__expf_finite", "__svml_expf16", 16)
+
+TLI_DEFINE_VECFUNC("llvm.exp.f64", "__svml_exp2", 2)
+TLI_DEFINE_VECFUNC("llvm.exp.f64", "__svml_exp4", 4)
+TLI_DEFINE_VECFUNC("llvm.exp.f64", "__svml_exp8", 8)
+
+TLI_DEFINE_VECFUNC("llvm.exp.f32", "__svml_expf4", 4)
+TLI_DEFINE_VECFUNC("llvm.exp.f32", "__svml_expf8", 8)
+TLI_DEFINE_VECFUNC("llvm.exp.f32", "__svml_expf16", 16)
+
+TLI_DEFINE_VECFUNC("log", "__svml_log2", 2)
+TLI_DEFINE_VECFUNC("log", "__svml_log4", 4)
+TLI_DEFINE_VECFUNC("log", "__svml_log8", 8)
+
+TLI_DEFINE_VECFUNC("logf", "__svml_logf4", 4)
+TLI_DEFINE_VECFUNC("logf", "__svml_logf8", 8)
+TLI_DEFINE_VECFUNC("logf", "__svml_logf16", 16)
+
+TLI_DEFINE_VECFUNC("__log_finite", "__svml_log2", 2)
+TLI_DEFINE_VECFUNC("__log_finite", "__svml_log4", 4)
+TLI_DEFINE_VECFUNC("__log_finite", "__svml_log8", 8)
+
+TLI_DEFINE_VECFUNC("__logf_finite", "__svml_logf4", 4)
+TLI_DEFINE_VECFUNC("__logf_finite", "__svml_logf8", 8)
+TLI_DEFINE_VECFUNC("__logf_finite", "__svml_logf16", 16)
+
+TLI_DEFINE_VECFUNC("llvm.log.f64", "__svml_log2", 2)
+TLI_DEFINE_VECFUNC("llvm.log.f64", "__svml_log4", 4)
+TLI_DEFINE_VECFUNC("llvm.log.f64", "__svml_log8", 8)
+
+TLI_DEFINE_VECFUNC("llvm.log.f32", "__svml_logf4", 4)
+TLI_DEFINE_VECFUNC("llvm.log.f32", "__svml_logf8", 8)
+TLI_DEFINE_VECFUNC("llvm.log.f32", "__svml_logf16", 16)
+
+
+#else
+#error "Must choose which vector library functions are to be defined."
+#endif
+
+#undef TLI_DEFINE_VECFUNC
+#undef TLI_DEFINE_ACCELERATE_VECFUNCS
+#undef TLI_DEFINE_SVML_VECFUNCS
+

llvm/lib/Analysis/TargetLibraryInfo.cpp

void TargetLibraryInfoImpl::addVectorizableFunctions(ArrayRef<VecDesc> Fns) {
   llvm::sort(ScalarDescs, compareByVectorFnName);
 }
 
 void TargetLibraryInfoImpl::addVectorizableFunctionsFromVecLib(
     enum VectorLibrary VecLib) {
   switch (VecLib) {
   case Accelerate: {
     const VecDesc VecFuncs[] = {
-        // Floating-Point Arithmetic and Auxiliary Functions
-        {"ceilf", "vceilf", 4},
-        {"fabsf", "vfabsf", 4},
-        {"llvm.fabs.f32", "vfabsf", 4},
-        {"floorf", "vfloorf", 4},
-        {"sqrtf", "vsqrtf", 4},
-        {"llvm.sqrt.f32", "vsqrtf", 4},
-
-        // Exponential and Logarithmic Functions
-        {"expf", "vexpf", 4},
-        {"llvm.exp.f32", "vexpf", 4},
-        {"expm1f", "vexpm1f", 4},
-        {"logf", "vlogf", 4},
-        {"llvm.log.f32", "vlogf", 4},
-        {"log1pf", "vlog1pf", 4},
-        {"log10f", "vlog10f", 4},
-        {"llvm.log10.f32", "vlog10f", 4},
-        {"logbf", "vlogbf", 4},
-
-        // Trigonometric Functions
-        {"sinf", "vsinf", 4},
-        {"llvm.sin.f32", "vsinf", 4},
-        {"cosf", "vcosf", 4},
-        {"llvm.cos.f32", "vcosf", 4},
-        {"tanf", "vtanf", 4},
-        {"asinf", "vasinf", 4},
-        {"acosf", "vacosf", 4},
-        {"atanf", "vatanf", 4},
-
-        // Hyperbolic Functions
-        {"sinhf", "vsinhf", 4},
-        {"coshf", "vcoshf", 4},
-        {"tanhf", "vtanhf", 4},
-        {"asinhf", "vasinhf", 4},
-        {"acoshf", "vacoshf", 4},
-        {"atanhf", "vatanhf", 4},
+    #define TLI_DEFINE_ACCELERATE_VECFUNCS
+    #include "llvm/Analysis/VecFuncs.def"
     };
     addVectorizableFunctions(VecFuncs);
     break;
   }
   case SVML: {
     const VecDesc VecFuncs[] = {
-        {"sin", "__svml_sin2", 2},
-        {"sin", "__svml_sin4", 4},
-        {"sin", "__svml_sin8", 8},
-
-        {"sinf", "__svml_sinf4", 4},
-        {"sinf", "__svml_sinf8", 8},
-        {"sinf", "__svml_sinf16", 16},
-
-        {"llvm.sin.f64", "__svml_sin2", 2},
-        {"llvm.sin.f64", "__svml_sin4", 4},
-        {"llvm.sin.f64", "__svml_sin8", 8},
-
-        {"llvm.sin.f32", "__svml_sinf4", 4},
-        {"llvm.sin.f32", "__svml_sinf8", 8},
-        {"llvm.sin.f32", "__svml_sinf16", 16},
-
-        {"cos", "__svml_cos2", 2},
-        {"cos", "__svml_cos4", 4},
-        {"cos", "__svml_cos8", 8},
-
-        {"cosf", "__svml_cosf4", 4},
-        {"cosf", "__svml_cosf8", 8},
-        {"cosf", "__svml_cosf16", 16},
-
-        {"llvm.cos.f64", "__svml_cos2", 2},
-        {"llvm.cos.f64", "__svml_cos4", 4},
-        {"llvm.cos.f64", "__svml_cos8", 8},
-
-        {"llvm.cos.f32", "__svml_cosf4", 4},
-        {"llvm.cos.f32", "__svml_cosf8", 8},
-        {"llvm.cos.f32", "__svml_cosf16", 16},
-
-        {"pow", "__svml_pow2", 2},
-        {"pow", "__svml_pow4", 4},
-        {"pow", "__svml_pow8", 8},
-
-        {"powf", "__svml_powf4", 4},
-        {"powf", "__svml_powf8", 8},
-        {"powf", "__svml_powf16", 16},
-
-        { "__pow_finite", "__svml_pow2", 2 },
-        { "__pow_finite", "__svml_pow4", 4 },
-        { "__pow_finite", "__svml_pow8", 8 },
-
-        { "__powf_finite", "__svml_powf4", 4 },
-        { "__powf_finite", "__svml_powf8", 8 },
-        { "__powf_finite", "__svml_powf16", 16 },
-
-        {"llvm.pow.f64", "__svml_pow2", 2},
-        {"llvm.pow.f64", "__svml_pow4", 4},
-        {"llvm.pow.f64", "__svml_pow8", 8},
-
-        {"llvm.pow.f32", "__svml_powf4", 4},
-        {"llvm.pow.f32", "__svml_powf8", 8},
-        {"llvm.pow.f32", "__svml_powf16", 16},
-
-        {"exp", "__svml_exp2", 2},
-        {"exp", "__svml_exp4", 4},
-        {"exp", "__svml_exp8", 8},
-
-        {"expf", "__svml_expf4", 4},
-        {"expf", "__svml_expf8", 8},
-        {"expf", "__svml_expf16", 16},
-
-        { "__exp_finite", "__svml_exp2", 2 },
-        { "__exp_finite", "__svml_exp4", 4 },
-        { "__exp_finite", "__svml_exp8", 8 },
-
-        { "__expf_finite", "__svml_expf4", 4 },
-        { "__expf_finite", "__svml_expf8", 8 },
-        { "__expf_finite", "__svml_expf16", 16 },
-
-        {"llvm.exp.f64", "__svml_exp2", 2},
-        {"llvm.exp.f64", "__svml_exp4", 4},
-        {"llvm.exp.f64", "__svml_exp8", 8},
-
-        {"llvm.exp.f32", "__svml_expf4", 4},
-        {"llvm.exp.f32", "__svml_expf8", 8},
-        {"llvm.exp.f32", "__svml_expf16", 16},
-
-        {"log", "__svml_log2", 2},
-        {"log", "__svml_log4", 4},
-        {"log", "__svml_log8", 8},
-
-        {"logf", "__svml_logf4", 4},
-        {"logf", "__svml_logf8", 8},
-        {"logf", "__svml_logf16", 16},
-
-        { "__log_finite", "__svml_log2", 2 },
-        { "__log_finite", "__svml_log4", 4 },
-        { "__log_finite", "__svml_log8", 8 },
-
-        { "__logf_finite", "__svml_logf4", 4 },
-        { "__logf_finite", "__svml_logf8", 8 },
-        { "__logf_finite", "__svml_logf16", 16 },
-
-        {"llvm.log.f64", "__svml_log2", 2},
-        {"llvm.log.f64", "__svml_log4", 4},
-        {"llvm.log.f64", "__svml_log8", 8},
-
-        {"llvm.log.f32", "__svml_logf4", 4},
-        {"llvm.log.f32", "__svml_logf8", 8},
-        {"llvm.log.f32", "__svml_logf16", 16},
+    #define TLI_DEFINE_SVML_VECFUNCS
+    #include "llvm/Analysis/VecFuncs.def"
     };
     addVectorizableFunctions(VecFuncs);
     break;
   }
   case NoLibrary:
     break;
   }
 }