Removed unnecessary min and max calls for rdsum avx512.

PiperOrigin-RevId: 660299297
google · Aug 7, 2024 · 1ecee10 · 1ecee10
1 parent 1cda6cd
commit 1ecee10
Show file tree

Hide file tree

Showing 5 changed files with 0 additions and 20 deletions.
diff --git a/src/f32-rdsum/avx512.c.in b/src/f32-rdsum/avx512.c.in
@@ -126,17 +126,13 @@ void xnn_f32_rdsum_ukernel_${ACCUMULATORS}p${ACCUMULATORS}x__avx512f_c${CHANNELS
     }
     for (int i = 0; i < channels >> 4; ++i) {
       vacc[i] = _mm512_add_ps(vo[i], vacc[i]);
-      vacc[i] = _mm512_max_ps(vacc[i], vmin);
-      vacc[i] = _mm512_min_ps(vacc[i], vmax);
     }
     for (int i = 0; i < channels >> 4; ++i) {
       _mm512_storeu_ps(output, vacc[i]); output += 16;
     }
     if (remainder) {
       const size_t pos = num_full_chunks;
       __m512 vout = vacc[pos];
-      vout = _mm512_max_ps(vout, vmin);
-      vout = _mm512_min_ps(vout, vmax);
       vout = _mm512_maskz_add_ps(vmask, vout,  _mm512_maskz_loadu_ps(vmask, output));
       _mm512_mask_storeu_ps(output, vmask, vout);
     }

diff --git a/src/f32-rdsum/gen/f32-rdsum-7p7x-minmax-avx512f-c128.c b/src/f32-rdsum/gen/f32-rdsum-7p7x-minmax-avx512f-c128.c
@@ -341,17 +341,13 @@ void xnn_f32_rdsum_ukernel_7p7x__avx512f_c128(
     }
     for (int i = 0; i < channels >> 4; ++i) {
       vacc[i] = _mm512_add_ps(vo[i], vacc[i]);
-      vacc[i] = _mm512_max_ps(vacc[i], vmin);
-      vacc[i] = _mm512_min_ps(vacc[i], vmax);
     }
     for (int i = 0; i < channels >> 4; ++i) {
       _mm512_storeu_ps(output, vacc[i]); output += 16;
     }
     if (remainder) {
       const size_t pos = num_full_chunks;
       __m512 vout = vacc[pos];
-      vout = _mm512_max_ps(vout, vmin);
-      vout = _mm512_min_ps(vout, vmax);
       vout = _mm512_maskz_add_ps(vmask, vout,  _mm512_maskz_loadu_ps(vmask, output));
       _mm512_mask_storeu_ps(output, vmask, vout);
     }

diff --git a/src/f32-rdsum/gen/f32-rdsum-7p7x-minmax-avx512f-c16.c b/src/f32-rdsum/gen/f32-rdsum-7p7x-minmax-avx512f-c16.c
@@ -180,17 +180,13 @@ void xnn_f32_rdsum_ukernel_7p7x__avx512f_c16(
     }
     for (int i = 0; i < channels >> 4; ++i) {
       vacc[i] = _mm512_add_ps(vo[i], vacc[i]);
-      vacc[i] = _mm512_max_ps(vacc[i], vmin);
-      vacc[i] = _mm512_min_ps(vacc[i], vmax);
     }
     for (int i = 0; i < channels >> 4; ++i) {
       _mm512_storeu_ps(output, vacc[i]); output += 16;
     }
     if (remainder) {
       const size_t pos = num_full_chunks;
       __m512 vout = vacc[pos];
-      vout = _mm512_max_ps(vout, vmin);
-      vout = _mm512_min_ps(vout, vmax);
       vout = _mm512_maskz_add_ps(vmask, vout,  _mm512_maskz_loadu_ps(vmask, output));
       _mm512_mask_storeu_ps(output, vmask, vout);
     }

diff --git a/src/f32-rdsum/gen/f32-rdsum-7p7x-minmax-avx512f-c32.c b/src/f32-rdsum/gen/f32-rdsum-7p7x-minmax-avx512f-c32.c
@@ -203,17 +203,13 @@ void xnn_f32_rdsum_ukernel_7p7x__avx512f_c32(
     }
     for (int i = 0; i < channels >> 4; ++i) {
       vacc[i] = _mm512_add_ps(vo[i], vacc[i]);
-      vacc[i] = _mm512_max_ps(vacc[i], vmin);
-      vacc[i] = _mm512_min_ps(vacc[i], vmax);
     }
     for (int i = 0; i < channels >> 4; ++i) {
       _mm512_storeu_ps(output, vacc[i]); output += 16;
     }
     if (remainder) {
       const size_t pos = num_full_chunks;
       __m512 vout = vacc[pos];
-      vout = _mm512_max_ps(vout, vmin);
-      vout = _mm512_min_ps(vout, vmax);
       vout = _mm512_maskz_add_ps(vmask, vout,  _mm512_maskz_loadu_ps(vmask, output));
       _mm512_mask_storeu_ps(output, vmask, vout);
     }

diff --git a/src/f32-rdsum/gen/f32-rdsum-7p7x-minmax-avx512f-c64.c b/src/f32-rdsum/gen/f32-rdsum-7p7x-minmax-avx512f-c64.c
@@ -249,17 +249,13 @@ void xnn_f32_rdsum_ukernel_7p7x__avx512f_c64(
     }
     for (int i = 0; i < channels >> 4; ++i) {
       vacc[i] = _mm512_add_ps(vo[i], vacc[i]);
-      vacc[i] = _mm512_max_ps(vacc[i], vmin);
-      vacc[i] = _mm512_min_ps(vacc[i], vmax);
     }
     for (int i = 0; i < channels >> 4; ++i) {
       _mm512_storeu_ps(output, vacc[i]); output += 16;
     }
     if (remainder) {
       const size_t pos = num_full_chunks;
       __m512 vout = vacc[pos];
-      vout = _mm512_max_ps(vout, vmin);
-      vout = _mm512_min_ps(vout, vmax);
       vout = _mm512_maskz_add_ps(vmask, vout,  _mm512_maskz_loadu_ps(vmask, output));
       _mm512_mask_storeu_ps(output, vmask, vout);
     }