[SYCL] [NATIVECPU] Implement missing math builtins for scalar data types

libclc/x86_64-unknown-linux/libspirv/SOURCES

@@ -16,3 +16,4 @@ math/native_sqrt.cl
 math/rint.cl
 math/round.cl
 math/trunc.cl
+shared/helpers.ll

libclc/x86_64-unknown-linux/libspirv/integer/helpers.h

@@ -0,0 +1,9 @@
+#include "func.h"
+
+#define GEN_UNARY_BUILTIN_T(NAME, TYPE)                                        \
+  _CLC_OVERLOAD TYPE __##NAME##_helper(TYPE);                                  \
+  _CLC_OVERLOAD TYPE __spirv_ocl_##NAME(TYPE n) { return __##NAME##_helper(n); }
+
+#define GEN_UNARY_BUILTIN(NAME)                                                \
+  GEN_UNARY_BUILTIN_T(NAME, int)                                               \
+  GEN_UNARY_BUILTIN_T(NAME, signed char)

libclc/x86_64-unknown-linux/libspirv/integer/popcount.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(popcount)

libclc/x86_64-unknown-linux/libspirv/math/ceil.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(ceil)

libclc/x86_64-unknown-linux/libspirv/math/fabs.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(fabs)

libclc/x86_64-unknown-linux/libspirv/math/floor.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(floor)

libclc/x86_64-unknown-linux/libspirv/math/fma.cl

@@ -0,0 +1,4 @@
+#include "helpers.h"
+
+GEN_TERNARY_BUILTIN(fma);
+

libclc/x86_64-unknown-linux/libspirv/math/helpers.h

@@ -0,0 +1,19 @@
+#include "func.h"
+
+#define GEN_UNARY_BUILTIN_T(NAME, TYPE)                                        \
+  _CLC_OVERLOAD TYPE __##NAME##_helper(TYPE);                                  \
+  _CLC_OVERLOAD TYPE __spirv_ocl_##NAME(TYPE n) { return __##NAME##_helper(n); }
+
+#define GEN_TERNARY_BUILTIN_T(NAME, TYPE)                                      \
+  _CLC_OVERLOAD TYPE __##NAME##_helper(TYPE, TYPE, TYPE);                      \
+  _CLC_OVERLOAD TYPE __spirv_ocl_##NAME(TYPE a, TYPE b, TYPE c) {              \
+    return __##NAME##_helper(a, b, c);                                         \
+  }
+
+#define GEN_UNARY_BUILTIN(NAME)                                                \
+  GEN_UNARY_BUILTIN_T(NAME, float)                                             \
+  GEN_UNARY_BUILTIN_T(NAME, double)
+
+#define GEN_TERNARY_BUILTIN(NAME)                                              \
+  GEN_TERNARY_BUILTIN_T(NAME, float)                                           \
+  GEN_TERNARY_BUILTIN_T(NAME, double)\

libclc/x86_64-unknown-linux/libspirv/math/native_cos.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(native_cos)

libclc/x86_64-unknown-linux/libspirv/math/native_exp.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(native_exp)

libclc/x86_64-unknown-linux/libspirv/math/native_exp2.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(native_exp2)

libclc/x86_64-unknown-linux/libspirv/math/native_log.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(native_log)

libclc/x86_64-unknown-linux/libspirv/math/native_log10.cl

@@ -0,0 +1,4 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(native_log10)
+

libclc/x86_64-unknown-linux/libspirv/math/native_log2.cl

@@ -0,0 +1,4 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(native_log2)
+

libclc/x86_64-unknown-linux/libspirv/math/native_sin.cl

@@ -0,0 +1,4 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(native_sin)
+

libclc/x86_64-unknown-linux/libspirv/math/native_sqrt.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(native_sqrt)

libclc/x86_64-unknown-linux/libspirv/math/rint.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(rint)

libclc/x86_64-unknown-linux/libspirv/math/round.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(round)

libclc/x86_64-unknown-linux/libspirv/math/sqrt.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(sqrt)

libclc/x86_64-unknown-linux/libspirv/math/trunc.cl

@@ -0,0 +1,3 @@
+#include "helpers.h"
+
+GEN_UNARY_BUILTIN(trunc)

libclc/x86_64-unknown-linux/libspirv/shared/helpers.ll

@@ -0,0 +1,270 @@
+declare float @llvm.sqrt.f32(float %n)
+declare double @llvm.sqrt.f64(double %n)
+declare float @llvm.fabs.f32(float %n)
+declare double @llvm.fabs.f64(double %n)
+declare float @llvm.trunc.f32(float %n)
+declare double @llvm.trunc.f64(double %n)
+declare float @llvm.ceil.f32(float %n)
+declare double @llvm.ceil.f64(double %n)
+declare float @llvm.floor.f32(float %n)
+declare double @llvm.floor.f64(double %n)
+declare float @llvm.round.f32(float %n)
+declare double @llvm.round.f64(double %n)
+declare float @llvm.rint.f32(float %n)
+declare double @llvm.rint.f64(double %n)
+declare float @llvm.cos.f32(float %n)
+declare double @llvm.cos.f64(double %n)
+declare float @llvm.sin.f32(float %n)
+declare double @llvm.sin.f64(double %n)
+declare float @llvm.exp2.f32(float %n)
+declare double @llvm.exp2.f64(double %n)
+declare float @llvm.exp.f32(float %n)
+declare double @llvm.exp.f64(double %n)
+declare float @llvm.log10.f32(float %n)
+declare double @llvm.log10.f64(double %n)
+declare float @llvm.log.f32(float %n)
+declare double @llvm.log.f64(double %n)
+declare float @llvm.log2.f32(float %n)
+declare double @llvm.log2.f64(double %n)
+declare float @llvm.fma.f32(float %n1, float %n2, float %n3)
+declare double @llvm.fma.f64(double %n1, double %n2, double %n3)
+declare i32 @llvm.ctpop.i32(i32 %n)
+declare i8 @llvm.ctpop.i8(i8 %n)
+
+define dso_local float @_Z13__sqrt_helperf(float %x) {
+entry:
+  %call = call float @llvm.sqrt.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z13__sqrt_helperd(double %x) {
+entry:
+  %call = call double @llvm.sqrt.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z13__fabs_helperf(float %x) {
+entry:
+  %call = call float @llvm.fabs.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z13__fabs_helperd(double %x) {
+entry:
+  %call = call double @llvm.fabs.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z14__trunc_helperf(float %x) {
+entry:
+  %call = call float @llvm.trunc.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z14__trunc_helperd(double %x) {
+entry:
+  %call = call double @llvm.trunc.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z13__ceil_helperf(float %x) {
+entry:
+  %call = call float @llvm.ceil.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z13__ceil_helperd(double %x) {
+entry:
+  %call = call double @llvm.ceil.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z14__floor_helperf(float %x) {
+entry:
+  %call = call float @llvm.floor.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z14__floor_helperd(double %x) {
+entry:
+  %call = call double @llvm.floor.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z14__round_helperf(float %x) {
+entry:
+  %call = call float @llvm.round.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z14__round_helperd(double %x) {
+entry:
+  %call = call double @llvm.round.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z13__rint_helperf(float %x) {
+entry:
+  %call = call float @llvm.rint.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z13__rint_helperd(double %x) {
+entry:
+  %call = call double @llvm.rint.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z20__native_sqrt_helperf(float %x) {
+entry:
+  %call = call float @llvm.sqrt.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z20__native_sqrt_helperd(double %x) {
+entry:
+  %call = call double @llvm.sqrt.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z19__native_cos_helperf(float %x) {
+entry:
+  %call = call float @llvm.cos.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z19__native_cos_helperd(double %x) {
+entry:
+  %call = call double @llvm.cos.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z19__native_sin_helperf(float %x) {
+entry:
+  %call = call float @llvm.sin.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z19__native_sin_helperd(double %x) {
+entry:
+  %call = call double @llvm.sin.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z20__native_exp2_helperf(float %x) {
+entry:
+  %call = call float @llvm.exp2.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z20__native_exp2_helperd(double %x) {
+entry:
+  %call = call double @llvm.exp2.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z19__native_exp_helperf(float %x) {
+entry:
+  %call = call float @llvm.exp.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z19__native_exp_helperd(double %x) {
+entry:
+  %call = call double @llvm.exp.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z21__native_log10_helperf(float %x) {
+entry:
+  %call = call float @llvm.log10.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z21__native_log10_helperd(double %x) {
+entry:
+  %call = call double @llvm.log10.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z19__native_log_helperf(float %x) {
+entry:
+  %call = call float @llvm.log.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z19__native_log_helperd(double %x) {
+entry:
+  %call = call double @llvm.log.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z20__native_log2_helperf(float %x) {
+entry:
+  %call = call float @llvm.log2.f32(float %x) 
+  ret float %call
+}
+
+
+define dso_local double @_Z20__native_log2_helperd(double %x) {
+entry:
+  %call = call double @llvm.log2.f64(double %x) 
+  ret double %call
+}
+
+
+define dso_local float @_Z12__fma_helperfff(float %a, float %b, float %c) {
+entry:
+  %call = call float @llvm.fma.f32(float %a, float %b, float %c) 
+  ret float %call
+}
+
+
+define dso_local double @_Z12__fma_helperddd(double %a, double %b, double %c) {
+entry:
+  %call = call double @llvm.fma.f64(double %a, double %b, double %c) 
+  ret double %call
+}
+
+
+define dso_local i32 @_Z17__popcount_helperi(i32 %x) {
+entry:
+  %call = call i32 @llvm.ctpop.i32(i32 %x) 
+  ret i32 %call
+}
+
+
+define dso_local i8 @_Z17__popcount_helpera(i8 %x) {
+entry:
+  %call = call i8 @llvm.ctpop.i8(i8 %x) 
+  ret i8 %call
+}
+

llvm/lib/SYCLLowerIR/PrepareSYCLNativeCPU.cpp

@@ -245,8 +245,6 @@ Value *getStateArg(const Function *F) {
   return F->getArg(FT->getNumParams() - 1);
 }
 
-static constexpr unsigned int NativeCPUGlobalAS = 1;
-
 } // namespace
 
 PreservedAnalyses PrepareSYCLNativeCPUPass::run(Module &M,