going on big buffer, need three diff task scheduler

StgSharp.Native/sn-src/k_bufop.c

@@ -0,0 +1,133 @@
+#include "sn_intrinsic.h"
+
+DECLARE_BUF_PROC_ANS_SCALAR(float, sse, fill)
+{
+        register __m128 vec = _mm_set_ps1(((float *)(scalar->data))[0]);
+        for (size_t i = 0; i < count; i++) {
+                ans[i].xmm[0] = vec;
+                ans[i].xmm[1] = vec;
+                ans[i].xmm[2] = vec;
+                ans[i].xmm[3] = vec;
+        }
+}
+
+DECLARE_BUF_PROC_ANS_SCALAR(float, avx, fill)
+{
+        register __m256 vec = _mm256_broadcast_ss(((float *)(scalar->data)));
+        for (size_t i = 0; i < count; i++) {
+                ans[i].ymm[0] = vec;
+                ans[i].ymm[1] = vec;
+        }
+}
+
+DECLARE_BUF_PROC_ANS_SCALAR(float, 512, fill)
+{
+        register __m512 vec = _mm512_broadcastss_ps(((__m128 *)(scalar->data))[0]);
+        for (size_t i = 0; i < count; i++) {
+                ans[i].zmm[0] = vec;
+        }
+}
+
+DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, add)
+{
+        for (size_t i = 0; i < count; i++) {
+                register __m128 c0 = _mm_add_ps(left[i].xmm[0], right[i].xmm[0]);
+                register __m128 c1 = _mm_add_ps(left[i].xmm[1], right[i].xmm[1]);
+                register __m128 c2 = _mm_add_ps(left[i].xmm[2], right[i].xmm[2]);
+                register __m128 c3 = _mm_add_ps(left[i].xmm[3], right[i].xmm[3]);
+
+                ans[i].xmm[0] = c0;
+                ans[i].xmm[1] = c1;
+                ans[i].xmm[2] = c2;
+                ans[i].xmm[3] = c3;
+        }
+}
+
+DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, add)
+{
+        for (size_t i = 0; i < count; i++) {
+                register __m256 c0 = _mm256_add_ps(left[i].ymm[0], right[i].ymm[0]);
+                register __m256 c1 = _mm256_add_ps(left[i].ymm[1], right[i].ymm[1]);
+                ans[i].ymm[0] = c0;
+                ans[i].ymm[1] = c1;
+        }
+}
+
+DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, add)
+{
+        for (size_t i = 0; i < count; i++) {
+                register __m512 c0 = _mm512_add_ps(left[i].zmm[0], right[i].zmm[0]);
+                ans[i].zmm[0] = c0;
+        }
+}
+
+DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, sub)
+{
+        for (size_t i = 0; i < count; i++) {
+                register __m128 c0 = _mm_sub_ps(left[i].xmm[0], right[i].xmm[0]);
+                register __m128 c1 = _mm_sub_ps(left[i].xmm[1], right[i].xmm[1]);
+                register __m128 c2 = _mm_sub_ps(left[i].xmm[2], right[i].xmm[2]);
+                register __m128 c3 = _mm_sub_ps(left[i].xmm[3], right[i].xmm[3]);
+
+                ans[i].xmm[0] = c0;
+                ans[i].xmm[1] = c1;
+                ans[i].xmm[2] = c2;
+                ans[i].xmm[3] = c3;
+        }
+}
+
+DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, sub)
+{
+        for (size_t i = 0; i < count; i++) {
+                register __m256 c0 = _mm256_sub_ps(left[i].ymm[0], right[i].ymm[0]);
+                register __m256 c1 = _mm256_sub_ps(left[i].ymm[1], right[i].ymm[1]);
+                ans[i].ymm[0] = c0;
+                ans[i].ymm[1] = c1;
+        }
+}
+
+DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, sub)
+{
+        for (size_t i = 0; i < count; i++) {
+                register __m512 c0 = _mm512_sub_ps(left[i].zmm[0], right[i].zmm[0]);
+                ans[i].zmm[0] = c0;
+        }
+}
+
+DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, sse, scalar_mul)
+{
+        register __m128 scalar_vec = ((__m128 *)(scalar->data))[0];
+        for (size_t i = 0; i < count; i++) {
+                register __m128 c0 = _mm_mul_ps(right[i].xmm[0], scalar_vec);
+                register __m128 c1 = _mm_mul_ps(right[i].xmm[1], scalar_vec);
+                register __m128 c2 = _mm_mul_ps(right[i].xmm[2], scalar_vec);
+                register __m128 c3 = _mm_mul_ps(right[i].xmm[3], scalar_vec);
+
+                ans[i].xmm[0] = c0;
+                ans[i].xmm[1] = c1;
+                ans[i].xmm[2] = c2;
+                ans[i].xmm[3] = c3;
+        }
+}
+
+DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, avx, scalar_mul)
+{
+        register __m256 scalar_vec = ((__m256 *)(scalar->data))[0];
+        for (size_t i = 0; i < count; i++) {
+                register __m256 c0 = _mm256_mul_ps(right[i].ymm[0], scalar_vec);
+                register __m256 c1 = _mm256_mul_ps(right[i].ymm[1], scalar_vec);
+
+                ans[i].ymm[0] = c0;
+                ans[i].ymm[1] = c1;
+        }
+}
+
+DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, 512, scalar_mul)
+{
+        register __m512 scalar_vec = ((__m512 *)(scalar->data))[0];
+        for (size_t i = 0; i < count; i++) {
+                register __m512 c0 = _mm512_mul_ps(right[i].zmm[0], scalar_vec);
+
+                ans[i].zmm[0] = c0;
+        }
+}

StgSharp.Native/sn-src/sn_native.c

@@ -1,6 +1,9 @@
-#include "StgSharpNative.h"
+#ifndef STB_IMAGE_IMPLEMENTATION
+#define STB_IMAGE_IMPLEMENTATION
+#endif
+#include "StgSharpNative.h"
 #include "gl.h"
-#include "../glfw/glfw3.h"
+#include "../lib/glfw/glfw3.h"
 #include <stdio.h>
 #include <string.h>
 
@@ -32,7 +35,7 @@ SN_API GLFWglproc SN_DECL loadGlfuncDefault(char const *procName)
         void *ret = glfwGetProcAddress(procName);
         // printf("%llu\n", (uint64_t)ret);
         if (ret == NULL) {
-                strcpy_s(infolog,30, "Failed to load OpenGL api:");
+                strcpy_s(infolog, 30, "Failed to load OpenGL api:");
                 for (size_t i = 0; i < sizeof(procName); i++) {
                         infolog[30 + i] = procName[i];
                 }
@@ -62,8 +65,7 @@ SN_API void SN_DECL loadImageData(char const *location, Image *out, imageLoader
         }
         out->pixelPtr = loader(location, &(out->width), &(out->height), &(out->channel), 0);
 }
-
-SN_API void SN_DECL unloadImageData(Image *restrict out)
+SN_API void SN_DECL unloadImageData(Image *out)
 {
         free(out->pixelPtr);
 }