Update mat.h

nihui · web-flow · commit e429fea4b8ee · 2025-01-03T10:35:31.000+08:00
diff --git a/src/mat.h b/src/mat.h
@@ -862,13 +862,13 @@ NCNN_FORCEINLINE Mat::Mat(int _w, int _h, void* _data, size_t _elemsize, Allocat
 NCNN_FORCEINLINE Mat::Mat(int _w, int _h, int _c, void* _data, size_t _elemsize, Allocator* _allocator)
     : data(_data), refcount(0), elemsize(_elemsize), elempack(1), allocator(_allocator), dims(3), w(_w), h(_h), d(1), c(_c)
 {
-    cstep = alignSize((size_t)w * h * elemsize, 16) / elemsize;
+    cstep = alignSize((size_t)w * h * elemsize, 128) / elemsize;
 }
 
 NCNN_FORCEINLINE Mat::Mat(int _w, int _h, int _d, int _c, void* _data, size_t _elemsize, Allocator* _allocator)
     : data(_data), refcount(0), elemsize(_elemsize), elempack(1), allocator(_allocator), dims(4), w(_w), h(_h), d(_d), c(_c)
 {
-    cstep = alignSize((size_t)w * h * d * elemsize, 16) / elemsize;
+    cstep = alignSize((size_t)w * h * d * elemsize, 128) / elemsize;
 }
 
 NCNN_FORCEINLINE Mat::Mat(int _w, void* _data, size_t _elemsize, int _elempack, Allocator* _allocator)
@@ -886,13 +886,13 @@ NCNN_FORCEINLINE Mat::Mat(int _w, int _h, void* _data, size_t _elemsize, int _el
 NCNN_FORCEINLINE Mat::Mat(int _w, int _h, int _c, void* _data, size_t _elemsize, int _elempack, Allocator* _allocator)
     : data(_data), refcount(0), elemsize(_elemsize), elempack(_elempack), allocator(_allocator), dims(3), w(_w), h(_h), d(1), c(_c)
 {
-    cstep = alignSize((size_t)w * h * elemsize, 16) / elemsize;
+    cstep = alignSize((size_t)w * h * elemsize, 128) / elemsize;
 }
 
 NCNN_FORCEINLINE Mat::Mat(int _w, int _h, int _d, int _c, void* _data, size_t _elemsize, int _elempack, Allocator* _allocator)
     : data(_data), refcount(0), elemsize(_elemsize), elempack(_elempack), allocator(_allocator), dims(4), w(_w), h(_h), d(_d), c(_c)
 {
-    cstep = alignSize((size_t)w * h * d * elemsize, 16) / elemsize;
+    cstep = alignSize((size_t)w * h * d * elemsize, 128) / elemsize;
 }
 
 NCNN_FORCEINLINE Mat::~Mat()
@@ -1446,13 +1446,13 @@ NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, VkBufferMemory* _data, size_t _ele
 NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, int _c, VkBufferMemory* _data, size_t _elemsize, VkAllocator* _allocator)
     : data(_data), refcount(0), elemsize(_elemsize), elempack(1), allocator(_allocator), dims(3), w(_w), h(_h), d(1), c(_c)
 {
-    cstep = alignSize(w * h * elemsize, 16) / elemsize;
+    cstep = alignSize(w * h * elemsize, 128) / elemsize;
 }
 
 NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, int _d, int _c, VkBufferMemory* _data, size_t _elemsize, VkAllocator* _allocator)
     : data(_data), refcount(0), elemsize(_elemsize), elempack(1), allocator(_allocator), dims(4), w(_w), h(_h), d(_d), c(_c)
 {
-    cstep = alignSize(w * h * d * elemsize, 16) / elemsize;
+    cstep = alignSize(w * h * d * elemsize, 128) / elemsize;
 }
 
 NCNN_FORCEINLINE VkMat::VkMat(int _w, VkBufferMemory* _data, size_t _elemsize, int _elempack, VkAllocator* _allocator)
@@ -1470,13 +1470,13 @@ NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, VkBufferMemory* _data, size_t _ele
 NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, int _c, VkBufferMemory* _data, size_t _elemsize, int _elempack, VkAllocator* _allocator)
     : data(_data), refcount(0), elemsize(_elemsize), elempack(_elempack), allocator(_allocator), dims(3), w(_w), h(_h), d(1), c(_c)
 {
-    cstep = alignSize(w * h * elemsize, 16) / elemsize;
+    cstep = alignSize(w * h * elemsize, 128) / elemsize;
 }
 
 NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, int _d, int _c, VkBufferMemory* _data, size_t _elemsize, int _elempack, VkAllocator* _allocator)
     : data(_data), refcount(0), elemsize(_elemsize), elempack(_elempack), allocator(_allocator), dims(4), w(_w), h(_h), d(_d), c(_c)
 {
-    cstep = alignSize(w * h * d * elemsize, 16) / elemsize;
+    cstep = alignSize(w * h * d * elemsize, 128) / elemsize;
 }
 
 NCNN_FORCEINLINE VkMat::~VkMat()

Original file line number	Diff line number	Diff line change
`@@ -862,13 +862,13 @@ NCNN_FORCEINLINE Mat::Mat(int _w, int _h, void* _data, size_t _elemsize, Allocat`
`862`	`862`	`NCNN_FORCEINLINE Mat::Mat(int _w, int _h, int _c, void* _data, size_t _elemsize, Allocator* _allocator)`
`863`	`863`	`: data(_data), refcount(0), elemsize(_elemsize), elempack(1), allocator(_allocator), dims(3), w(_w), h(_h), d(1), c(_c)`
`864`	`864`	`{`
`865`		`- cstep = alignSize((size_t)w * h * elemsize, 16) / elemsize;`
	`865`	`+ cstep = alignSize((size_t)w * h * elemsize, 128) / elemsize;`
`866`	`866`	`}`
`867`	`867`
`868`	`868`	`NCNN_FORCEINLINE Mat::Mat(int _w, int _h, int _d, int _c, void* _data, size_t _elemsize, Allocator* _allocator)`
`869`	`869`	`: data(_data), refcount(0), elemsize(_elemsize), elempack(1), allocator(_allocator), dims(4), w(_w), h(_h), d(_d), c(_c)`
`870`	`870`	`{`
`871`		`- cstep = alignSize((size_t)w * h * d * elemsize, 16) / elemsize;`
	`871`	`+ cstep = alignSize((size_t)w * h * d * elemsize, 128) / elemsize;`
`872`	`872`	`}`
`873`	`873`
`874`	`874`	`NCNN_FORCEINLINE Mat::Mat(int _w, void* _data, size_t _elemsize, int _elempack, Allocator* _allocator)`
`@@ -886,13 +886,13 @@ NCNN_FORCEINLINE Mat::Mat(int _w, int _h, void* _data, size_t _elemsize, int _el`
`886`	`886`	`NCNN_FORCEINLINE Mat::Mat(int _w, int _h, int _c, void* _data, size_t _elemsize, int _elempack, Allocator* _allocator)`
`887`	`887`	`: data(_data), refcount(0), elemsize(_elemsize), elempack(_elempack), allocator(_allocator), dims(3), w(_w), h(_h), d(1), c(_c)`
`888`	`888`	`{`
`889`		`- cstep = alignSize((size_t)w * h * elemsize, 16) / elemsize;`
	`889`	`+ cstep = alignSize((size_t)w * h * elemsize, 128) / elemsize;`
`890`	`890`	`}`
`891`	`891`
`892`	`892`	`NCNN_FORCEINLINE Mat::Mat(int _w, int _h, int _d, int _c, void* _data, size_t _elemsize, int _elempack, Allocator* _allocator)`
`893`	`893`	`: data(_data), refcount(0), elemsize(_elemsize), elempack(_elempack), allocator(_allocator), dims(4), w(_w), h(_h), d(_d), c(_c)`
`894`	`894`	`{`
`895`		`- cstep = alignSize((size_t)w * h * d * elemsize, 16) / elemsize;`
	`895`	`+ cstep = alignSize((size_t)w * h * d * elemsize, 128) / elemsize;`
`896`	`896`	`}`
`897`	`897`
`898`	`898`	`NCNN_FORCEINLINE Mat::~Mat()`
`@@ -1446,13 +1446,13 @@ NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, VkBufferMemory* _data, size_t _ele`
`1446`	`1446`	`NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, int _c, VkBufferMemory* _data, size_t _elemsize, VkAllocator* _allocator)`
`1447`	`1447`	`: data(_data), refcount(0), elemsize(_elemsize), elempack(1), allocator(_allocator), dims(3), w(_w), h(_h), d(1), c(_c)`
`1448`	`1448`	`{`
`1449`		`- cstep = alignSize(w * h * elemsize, 16) / elemsize;`
	`1449`	`+ cstep = alignSize(w * h * elemsize, 128) / elemsize;`
`1450`	`1450`	`}`
`1451`	`1451`
`1452`	`1452`	`NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, int _d, int _c, VkBufferMemory* _data, size_t _elemsize, VkAllocator* _allocator)`
`1453`	`1453`	`: data(_data), refcount(0), elemsize(_elemsize), elempack(1), allocator(_allocator), dims(4), w(_w), h(_h), d(_d), c(_c)`
`1454`	`1454`	`{`
`1455`		`- cstep = alignSize(w * h * d * elemsize, 16) / elemsize;`
	`1455`	`+ cstep = alignSize(w * h * d * elemsize, 128) / elemsize;`
`1456`	`1456`	`}`
`1457`	`1457`
`1458`	`1458`	`NCNN_FORCEINLINE VkMat::VkMat(int _w, VkBufferMemory* _data, size_t _elemsize, int _elempack, VkAllocator* _allocator)`
`@@ -1470,13 +1470,13 @@ NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, VkBufferMemory* _data, size_t _ele`
`1470`	`1470`	`NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, int _c, VkBufferMemory* _data, size_t _elemsize, int _elempack, VkAllocator* _allocator)`
`1471`	`1471`	`: data(_data), refcount(0), elemsize(_elemsize), elempack(_elempack), allocator(_allocator), dims(3), w(_w), h(_h), d(1), c(_c)`
`1472`	`1472`	`{`
`1473`		`- cstep = alignSize(w * h * elemsize, 16) / elemsize;`
	`1473`	`+ cstep = alignSize(w * h * elemsize, 128) / elemsize;`
`1474`	`1474`	`}`
`1475`	`1475`
`1476`	`1476`	`NCNN_FORCEINLINE VkMat::VkMat(int _w, int _h, int _d, int _c, VkBufferMemory* _data, size_t _elemsize, int _elempack, VkAllocator* _allocator)`
`1477`	`1477`	`: data(_data), refcount(0), elemsize(_elemsize), elempack(_elempack), allocator(_allocator), dims(4), w(_w), h(_h), d(_d), c(_c)`
`1478`	`1478`	`{`
`1479`		`- cstep = alignSize(w * h * d * elemsize, 16) / elemsize;`
	`1479`	`+ cstep = alignSize(w * h * d * elemsize, 128) / elemsize;`
`1480`	`1480`	`}`
`1481`	`1481`
`1482`	`1482`	`NCNN_FORCEINLINE VkMat::~VkMat()`