Draft: Porting from C to Cpp using Multidimensional Subscript Operator

In ifos3d a 3D field e.g. vx is represented by the type float ***.

• vx is an array of size nrow with elements of type float **
• vx[.] is mapped to an array of size nrow * ncol with elements of float *
• vx[.][.] is mapped to an array of nrow * ncol * ndep with elements of float
• As the index of the fields may not start at zero, this data structure requires to substract offsets from the pointers, which may lead to pointer underflows.

With C++ multidimensional subscript operators:

• The pointer underflow problem can be avoided
• The memory for the pointer structure can be saved
• Alignment information can be added to the data array to improve performance
• Assertions can be used to check that all tensor accesses are valid.

The simplest form of this subscript operator would look like:

constexpr float &float3DTensorT::operator[](const int &i, const int &j, const int &k) noexcept
{
    assert(_lb_i <= i and i <= _ub_i);
    assert(_lb_j <= j and j <= _ub_j);
    assert(_lb_k <= k and k <= _ub_k);
    const size_t index = i * _stride_i + j * _stride_j + k - offset;
    assert(index < _size);
    return data[index];
}

Tensor element access syntax would change from vel->vx[j][i][k] to vel.vx[j, i, k]

Performance Comparison

Hardware	Compiler	Runtime C-Code	Runtime C++-Code	Performance
AMD Zen4	GCC 13	248.45s	237.04s	5%
AMD Zen4	GCC 13 (fast-math)	242.12s	235.10s	3%
AMD Zen4	AOCC 5.0	223.31s	219.49s	2%
Intel Xeon	GCC 13	198.61s	196.36s	1%
Intel Raptor Lake	icx 2025.1	306.77s	293.90s	4%

build/Makefile

@@ -214,6 +214,7 @@ IFOS_FILES := absorb.cpp \
               exchange_s_acoustic.cpp \
               exchange_v.cpp \
               filt_seis.cpp \
+	      float3DTensorT.cpp \
               freemem.cpp \
               freemem_seis.cpp \
               fsource.cpp \
@@ -345,6 +346,7 @@ help:
 	@echo "   CXX: $(CXX)"
 	@echo "   CPPFLAGS: $(CPPFLAGS) [$(EXTRA_CPPFLAGS)]"
 	@echo "   CFLAGS: $(CFLAGS)"
+	@echo "   CXXFLAGS: $(CXXFLAGS)"
 	@echo "   LD: $(LD)"
 	@echo "   LDFLAGS: $(LDFLAGS) [$(EXTRA_LDFLAGS)]"
 	@echo 
@@ -444,7 +446,7 @@ quiet_cmd_compile_cxx = \e[32m[CC]\e[0m    $@
       cmd_compile_cxx = $(CXX) -o$@ $(CDEPEND) $(CXXFLAGS.$@) $(CPPFLAGS) -I$(SRCTREE) $(CXXFLAGS) -c $<
 
 quiet_cmd_link = \e[94m[LD]\e[0m    $@
-      cmd_link = $(LD) $(CFLAGS) -o$@ $^ $(LDFLAGS.$@) $(LDFLAGS)
+      cmd_link = $(LD) $(CXXFLAGS) -o$@ $^ $(LDFLAGS.$@) $(LDFLAGS)
 
 quiet_cmd_ar = \e[32m[AR]\e[0m    $@
       cmd_ar = $(AR) $(ARFLAGS) $@ $^ && ranlib $@