diff --git a/source/source_basis/module_pw/pw_transform.cpp b/source/source_basis/module_pw/pw_transform.cpp
index 220b353e9d4..06eecba1c9f 100644
--- a/source/source_basis/module_pw/pw_transform.cpp
+++ b/source/source_basis/module_pw/pw_transform.cpp
@@ -5,11 +5,22 @@
 #include "pw_basis.h"
 #include "pw_gatherscatter.h"
 
+#include <algorithm>
 #include <cassert>
 #include <complex>
 
 namespace ModulePW
 {
+namespace
+{
+constexpr int pw_transform_cache_block = 1024;
+
+inline int block_end(const int begin, const int size)
+{
+    return std::min(begin + pw_transform_cache_block, size);
+}
+} // namespace
+
 //     const base_device::DEVICE_CPU* PW_Basis::get_default_device_ctx() {
 //         static const base_device::DEVICE_CPU* default_device_cpu;
 //     return default_device_cpu;
@@ -34,28 +45,48 @@ void PW_Basis::real2recip(const std::complex<FPTYPE>* in,
     const int npw_ = this->npw;
     const int nxyz_ = this->nxyz;
     const int* ig2isz_ = this->ig2isz;
+    const std::complex<FPTYPE>* in_ = in;
+    std::complex<FPTYPE>* auxr = this->fft_bundle.get_auxr_data<FPTYPE>();
+    std::complex<FPTYPE>* auxg = this->fft_bundle.get_auxg_data<FPTYPE>();
+    ModuleBase::timer::start(this->classname, "real2recip_copy_r");
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-    for (int ir = 0; ir < nrxx_; ++ir)
+    for (int ib = 0; ib < nrxx_; ib += pw_transform_cache_block)
     {
-        this->fft_bundle.get_auxr_data<FPTYPE>()[ir] = in[ir];
+        const int iend = block_end(ib, nrxx_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+        for (int ir = ib; ir < iend; ++ir)
+        {
+            auxr[ir] = in_[ir];
+        }
     }
-    this->fft_bundle.fftxyfor(fft_bundle.get_auxr_data<FPTYPE>(), fft_bundle.get_auxr_data<FPTYPE>());
+    ModuleBase::timer::end(this->classname, "real2recip_copy_r");
+    this->fft_bundle.fftxyfor(auxr, auxr);
 
-    this->gatherp_scatters(this->fft_bundle.get_auxr_data<FPTYPE>(), this->fft_bundle.get_auxg_data<FPTYPE>());
+    this->gatherp_scatters(auxr, auxg);
 
-    this->fft_bundle.fftzfor(fft_bundle.get_auxg_data<FPTYPE>(), fft_bundle.get_auxg_data<FPTYPE>());
+    this->fft_bundle.fftzfor(auxg, auxg);
 
+    ModuleBase::timer::start(this->classname, "real2recip_copy_g");
     if (add)
     {
         FPTYPE tmpfac = factor / FPTYPE(nxyz_);
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-        for (int ig = 0; ig < npw_; ++ig)
+        for (int ib = 0; ib < npw_; ib += pw_transform_cache_block)
         {
-            out[ig] += tmpfac * this->fft_bundle.get_auxg_data<FPTYPE>()[ig2isz_[ig]];
+            const int iend = block_end(ib, npw_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+            for (int ig = ib; ig < iend; ++ig)
+            {
+                out[ig] += tmpfac * auxg[ig2isz_[ig]];
+            }
         }
     }
     else
@@ -64,11 +95,19 @@ void PW_Basis::real2recip(const std::complex<FPTYPE>* in,
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-        for (int ig = 0; ig < npw_; ++ig)
+        for (int ib = 0; ib < npw_; ib += pw_transform_cache_block)
         {
-            out[ig] = tmpfac * this->fft_bundle.get_auxg_data<FPTYPE>()[ig2isz_[ig]];
+            const int iend = block_end(ib, npw_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+            for (int ig = ib; ig < iend; ++ig)
+            {
+                out[ig] = tmpfac * auxg[ig2isz_[ig]];
+            }
         }
     }
+    ModuleBase::timer::end(this->classname, "real2recip_copy_g");
     ModuleBase::timer::end(this->classname, "real2recip");
 }
 
@@ -90,46 +129,73 @@ void PW_Basis::real2recip(const FPTYPE* in, std::complex<FPTYPE>* out, const boo
     const int nx_ = this->nx;
     const int ny_ = this->ny;
     const int nplane_ = this->nplane;
+    const FPTYPE* in_ = in;
+    std::complex<FPTYPE>* auxr = this->fft_bundle.get_auxr_data<FPTYPE>();
+    std::complex<FPTYPE>* auxg = this->fft_bundle.get_auxg_data<FPTYPE>();
+    FPTYPE* rspace = this->fft_bundle.get_rspace_data<FPTYPE>();
+    ModuleBase::timer::start(this->classname, "real2recip_copy_r");
     if (this->gamma_only)
     {
         const int npy = ny_ * nplane_;
+        const int nreal = nx_ * npy;
 #ifdef _OPENMP
-#pragma omp parallel for collapse(2) schedule(static)
+#pragma omp parallel for schedule(static)
 #endif
-        for (int ix = 0; ix < nx_; ++ix)
+        for (int ib = 0; ib < nreal; ib += pw_transform_cache_block)
         {
-            for (int ipy = 0; ipy < npy; ++ipy)
+            const int iend = block_end(ib, nreal);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+            for (int ir = ib; ir < iend; ++ir)
             {
-                this->fft_bundle.get_rspace_data<FPTYPE>()[ix * npy + ipy] = in[ix * npy + ipy];
+                rspace[ir] = in_[ir];
             }
         }
 
-        this->fft_bundle.fftxyr2c(fft_bundle.get_rspace_data<FPTYPE>(), fft_bundle.get_auxr_data<FPTYPE>());
+        ModuleBase::timer::end(this->classname, "real2recip_copy_r");
+        this->fft_bundle.fftxyr2c(rspace, auxr);
     }
     else
     {
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-        for (int ir = 0; ir < nrxx_; ++ir)
+        for (int ib = 0; ib < nrxx_; ib += pw_transform_cache_block)
         {
-            this->fft_bundle.get_auxr_data<FPTYPE>()[ir] = std::complex<FPTYPE>(in[ir], 0);
+            const int iend = block_end(ib, nrxx_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+            for (int ir = ib; ir < iend; ++ir)
+            {
+                auxr[ir] = std::complex<FPTYPE>(in_[ir], 0);
+            }
         }
-        this->fft_bundle.fftxyfor(fft_bundle.get_auxr_data<FPTYPE>(), fft_bundle.get_auxr_data<FPTYPE>());
+        ModuleBase::timer::end(this->classname, "real2recip_copy_r");
+        this->fft_bundle.fftxyfor(auxr, auxr);
     }
-    this->gatherp_scatters(this->fft_bundle.get_auxr_data<FPTYPE>(), this->fft_bundle.get_auxg_data<FPTYPE>());
+    this->gatherp_scatters(auxr, auxg);
 
-    this->fft_bundle.fftzfor(fft_bundle.get_auxg_data<FPTYPE>(), fft_bundle.get_auxg_data<FPTYPE>());
+    this->fft_bundle.fftzfor(auxg, auxg);
 
+    ModuleBase::timer::start(this->classname, "real2recip_copy_g");
     if (add)
     {
         FPTYPE tmpfac = factor / FPTYPE(nxyz_);
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-        for (int ig = 0; ig < npw_; ++ig)
+        for (int ib = 0; ib < npw_; ib += pw_transform_cache_block)
         {
-            out[ig] += tmpfac * this->fft_bundle.get_auxg_data<FPTYPE>()[ig2isz_[ig]];
+            const int iend = block_end(ib, npw_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+            for (int ig = ib; ig < iend; ++ig)
+            {
+                out[ig] += tmpfac * auxg[ig2isz_[ig]];
+            }
         }
     }
     else
@@ -138,11 +204,19 @@ void PW_Basis::real2recip(const FPTYPE* in, std::complex<FPTYPE>* out, const boo
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-        for (int ig = 0; ig < npw_; ++ig)
+        for (int ib = 0; ib < npw_; ib += pw_transform_cache_block)
         {
-            out[ig] = tmpfac * this->fft_bundle.get_auxg_data<FPTYPE>()[ig2isz_[ig]];
+            const int iend = block_end(ib, npw_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+            for (int ig = ib; ig < iend; ++ig)
+            {
+                out[ig] = tmpfac * auxg[ig2isz_[ig]];
+            }
         }
     }
+    ModuleBase::timer::end(this->classname, "real2recip_copy_g");
     ModuleBase::timer::end(this->classname, "real2recip");
 }
 
@@ -166,35 +240,62 @@ void PW_Basis::recip2real(const std::complex<FPTYPE>* in,
     const int npw_ = this->npw;
     const int nrxx_ = this->nrxx;
     const int* ig2isz_ = this->ig2isz;
+    const int nstnz_ = nst_ * nz_;
+    std::complex<FPTYPE>* auxg = this->fft_bundle.get_auxg_data<FPTYPE>();
+    std::complex<FPTYPE>* auxr = this->fft_bundle.get_auxr_data<FPTYPE>();
+    ModuleBase::timer::start(this->classname, "recip2real_copy_g");
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-    for (int i = 0; i < nst_ * nz_; ++i)
+    for (int ib = 0; ib < nstnz_; ib += pw_transform_cache_block)
     {
-        fft_bundle.get_auxg_data<FPTYPE>()[i] = std::complex<FPTYPE>(0, 0);
+        const int iend = block_end(ib, nstnz_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+        for (int i = ib; i < iend; ++i)
+        {
+            auxg[i] = std::complex<FPTYPE>(0, 0);
+        }
     }
 
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-    for (int ig = 0; ig < npw_; ++ig)
+    for (int ib = 0; ib < npw_; ib += pw_transform_cache_block)
     {
-        this->fft_bundle.get_auxg_data<FPTYPE>()[ig2isz_[ig]] = in[ig];
+        const int iend = block_end(ib, npw_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+        for (int ig = ib; ig < iend; ++ig)
+        {
+            auxg[ig2isz_[ig]] = in[ig];
+        }
     }
-    this->fft_bundle.fftzbac(fft_bundle.get_auxg_data<FPTYPE>(), fft_bundle.get_auxg_data<FPTYPE>());
+    ModuleBase::timer::end(this->classname, "recip2real_copy_g");
+    this->fft_bundle.fftzbac(auxg, auxg);
 
-    this->gathers_scatterp(this->fft_bundle.get_auxg_data<FPTYPE>(), this->fft_bundle.get_auxr_data<FPTYPE>());
+    this->gathers_scatterp(auxg, auxr);
 
-    this->fft_bundle.fftxybac(fft_bundle.get_auxr_data<FPTYPE>(), this->fft_bundle.get_auxr_data<FPTYPE>());
+    this->fft_bundle.fftxybac(auxr, auxr);
 
+    ModuleBase::timer::start(this->classname, "recip2real_copy_r");
     if (add)
     {
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-        for (int ir = 0; ir < nrxx_; ++ir)
+        for (int ib = 0; ib < nrxx_; ib += pw_transform_cache_block)
         {
-            out[ir] += factor * this->fft_bundle.get_auxr_data<FPTYPE>()[ir];
+            const int iend = block_end(ib, nrxx_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+            for (int ir = ib; ir < iend; ++ir)
+            {
+                out[ir] += factor * auxr[ir];
+            }
         }
     }
     else
@@ -202,11 +303,19 @@ void PW_Basis::recip2real(const std::complex<FPTYPE>* in,
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-        for (int ir = 0; ir < nrxx_; ++ir)
+        for (int ib = 0; ib < nrxx_; ib += pw_transform_cache_block)
         {
-            out[ir] = this->fft_bundle.get_auxr_data<FPTYPE>()[ir];
+            const int iend = block_end(ib, nrxx_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+            for (int ir = ib; ir < iend; ++ir)
+            {
+                out[ir] = auxr[ir];
+            }
         }
     }
+    ModuleBase::timer::end(this->classname, "recip2real_copy_r");
     ModuleBase::timer::end(this->classname, "recip2real");
 }
 
@@ -229,69 +338,108 @@ void PW_Basis::recip2real(const std::complex<FPTYPE>* in, FPTYPE* out, const boo
     const int ny_ = this->ny;
     const int nplane_ = this->nplane;
     const int* ig2isz_ = this->ig2isz;
+    const int nstnz_ = nst_ * nz_;
+    std::complex<FPTYPE>* auxg = this->fft_bundle.get_auxg_data<FPTYPE>();
+    std::complex<FPTYPE>* auxr = this->fft_bundle.get_auxr_data<FPTYPE>();
+    FPTYPE* rspace = this->fft_bundle.get_rspace_data<FPTYPE>();
+    ModuleBase::timer::start(this->classname, "recip2real_copy_g");
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-    for (int i = 0; i < nst_ * nz_; ++i)
+    for (int ib = 0; ib < nstnz_; ib += pw_transform_cache_block)
     {
-        fft_bundle.get_auxg_data<FPTYPE>()[i] = std::complex<FPTYPE>(0, 0);
+        const int iend = block_end(ib, nstnz_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+        for (int i = ib; i < iend; ++i)
+        {
+            auxg[i] = std::complex<FPTYPE>(0, 0);
+        }
     }
 
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-    for (int ig = 0; ig < npw_; ++ig)
+    for (int ib = 0; ib < npw_; ib += pw_transform_cache_block)
     {
-        this->fft_bundle.get_auxg_data<FPTYPE>()[ig2isz_[ig]] = in[ig];
+        const int iend = block_end(ib, npw_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+        for (int ig = ib; ig < iend; ++ig)
+        {
+            auxg[ig2isz_[ig]] = in[ig];
+        }
     }
-    this->fft_bundle.fftzbac(fft_bundle.get_auxg_data<FPTYPE>(), fft_bundle.get_auxg_data<FPTYPE>());
+    ModuleBase::timer::end(this->classname, "recip2real_copy_g");
+    this->fft_bundle.fftzbac(auxg, auxg);
 
-    this->gathers_scatterp(this->fft_bundle.get_auxg_data<FPTYPE>(), this->fft_bundle.get_auxr_data<FPTYPE>());
+    this->gathers_scatterp(auxg, auxr);
 
     if (this->gamma_only)
     {
-        this->fft_bundle.fftxyc2r(fft_bundle.get_auxr_data<FPTYPE>(), fft_bundle.get_rspace_data<FPTYPE>());
+        this->fft_bundle.fftxyc2r(auxr, rspace);
 
         const int npy = ny_ * nplane_;
+        const int nreal = nx_ * npy;
 
+        ModuleBase::timer::start(this->classname, "recip2real_copy_r");
         if (add)
         {
 #ifdef _OPENMP
-#pragma omp parallel for collapse(2) schedule(static)
+#pragma omp parallel for schedule(static)
 #endif
-            for (int ix = 0; ix < nx_; ++ix)
+            for (int ib = 0; ib < nreal; ib += pw_transform_cache_block)
             {
-                for (int ipy = 0; ipy < npy; ++ipy)
+                const int iend = block_end(ib, nreal);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+                for (int ir = ib; ir < iend; ++ir)
                 {
-                    out[ix * npy + ipy] += factor * this->fft_bundle.get_rspace_data<FPTYPE>()[ix * npy + ipy];
+                    out[ir] += factor * rspace[ir];
                 }
             }
         }
         else
         {
 #ifdef _OPENMP
-#pragma omp parallel for collapse(2) schedule(static)
+#pragma omp parallel for schedule(static)
 #endif
-            for (int ix = 0; ix < nx_; ++ix)
+            for (int ib = 0; ib < nreal; ib += pw_transform_cache_block)
             {
-                for (int ipy = 0; ipy < npy; ++ipy)
+                const int iend = block_end(ib, nreal);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+                for (int ir = ib; ir < iend; ++ir)
                 {
-                    out[ix * npy + ipy] = this->fft_bundle.get_rspace_data<FPTYPE>()[ix * npy + ipy];
+                    out[ir] = rspace[ir];
                 }
             }
         }
+        ModuleBase::timer::end(this->classname, "recip2real_copy_r");
     }
     else
     {
-        this->fft_bundle.fftxybac(fft_bundle.get_auxr_data<FPTYPE>(), fft_bundle.get_auxr_data<FPTYPE>());
+        this->fft_bundle.fftxybac(auxr, auxr);
+        ModuleBase::timer::start(this->classname, "recip2real_copy_r");
         if (add)
         {
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-            for (int ir = 0; ir < nrxx_; ++ir)
+            for (int ib = 0; ib < nrxx_; ib += pw_transform_cache_block)
             {
-                out[ir] += factor * this->fft_bundle.get_auxr_data<FPTYPE>()[ir].real();
+                const int iend = block_end(ib, nrxx_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+                for (int ir = ib; ir < iend; ++ir)
+                {
+                    out[ir] += factor * auxr[ir].real();
+                }
             }
         }
         else
@@ -299,11 +447,19 @@ void PW_Basis::recip2real(const std::complex<FPTYPE>* in, FPTYPE* out, const boo
 #ifdef _OPENMP
 #pragma omp parallel for schedule(static)
 #endif
-            for (int ir = 0; ir < nrxx_; ++ir)
+            for (int ib = 0; ib < nrxx_; ib += pw_transform_cache_block)
             {
-                out[ir] = this->fft_bundle.get_auxr_data<FPTYPE>()[ir].real();
+                const int iend = block_end(ib, nrxx_);
+#ifdef _OPENMP
+#pragma omp simd
+#endif
+                for (int ir = ib; ir < iend; ++ir)
+                {
+                    out[ir] = auxr[ir].real();
+                }
             }
         }
+        ModuleBase::timer::end(this->classname, "recip2real_copy_r");
     }
     ModuleBase::timer::end(this->classname, "recip2real");
 }
@@ -340,4 +496,4 @@ template void PW_Basis::recip2real<double>(const std::complex<double>* in,
                                            std::complex<double>* out,
                                            const bool add,
                                            const double factor) const;
-} // namespace ModulePW
\ No newline at end of file
+} // namespace ModulePW