1 | |
---|
2 | #include "SpuFakeDma.h" |
---|
3 | #include <LinearMath/btScalar.h> //for btAssert |
---|
4 | //Disabling memcpy sometimes helps debugging DMA |
---|
5 | |
---|
6 | #define USE_MEMCPY 1 |
---|
7 | #ifdef USE_MEMCPY |
---|
8 | |
---|
9 | #endif |
---|
10 | |
---|
11 | |
---|
12 | void* cellDmaLargeGetReadOnly(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid) |
---|
13 | { |
---|
14 | |
---|
15 | #if defined (__CELLOS_LV2__) || defined (USE_LIBSPE2) |
---|
16 | cellDmaLargeGet(ls,ea,size,tag,tid,rid); |
---|
17 | return ls; |
---|
18 | #else |
---|
19 | return (void*)(uint32_t)ea; |
---|
20 | #endif |
---|
21 | } |
---|
22 | |
---|
23 | void* cellDmaSmallGetReadOnly(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid) |
---|
24 | { |
---|
25 | #if defined (__SPU__) || defined (USE_LIBSPE2) |
---|
26 | mfc_get(ls,ea,size,tag,0,0); |
---|
27 | return ls; |
---|
28 | #else |
---|
29 | return (void*)(uint32_t)ea; |
---|
30 | #endif |
---|
31 | } |
---|
32 | |
---|
33 | |
---|
34 | |
---|
35 | |
---|
36 | void* cellDmaGetReadOnly(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid) |
---|
37 | { |
---|
38 | #if defined (__SPU__) || defined (USE_LIBSPE2) |
---|
39 | cellDmaGet(ls,ea,size,tag,tid,rid); |
---|
40 | return ls; |
---|
41 | #else |
---|
42 | return (void*)(uint32_t)ea; |
---|
43 | #endif |
---|
44 | } |
---|
45 | |
---|
46 | |
---|
47 | ///this unalignedDma should not be frequently used, only for small data. It handles alignment and performs check on size (<16 bytes) |
---|
48 | int stallingUnalignedDmaSmallGet(void *ls, uint64_t ea, uint32_t size) |
---|
49 | { |
---|
50 | |
---|
51 | btAssert(size<32); |
---|
52 | |
---|
53 | ATTRIBUTE_ALIGNED16(char tmpBuffer[32]); |
---|
54 | |
---|
55 | char* mainMem = (char*)ea; |
---|
56 | char* localStore = (char*)ls; |
---|
57 | uint32_t i; |
---|
58 | |
---|
59 | |
---|
60 | ///make sure last 4 bits are the same, for cellDmaSmallGet |
---|
61 | uint32_t last4BitsOffset = ea & 0x0f; |
---|
62 | char* tmpTarget = tmpBuffer + last4BitsOffset; |
---|
63 | |
---|
64 | #if defined (__SPU__) || defined (USE_LIBSPE2) |
---|
65 | |
---|
66 | int remainingSize = size; |
---|
67 | |
---|
68 | //#define FORCE_cellDmaUnalignedGet 1 |
---|
69 | #ifdef FORCE_cellDmaUnalignedGet |
---|
70 | cellDmaUnalignedGet(tmpTarget,ea,size,DMA_TAG(1),0,0); |
---|
71 | #else |
---|
72 | char* remainingTmpTarget = tmpTarget; |
---|
73 | uint64_t remainingEa = ea; |
---|
74 | |
---|
75 | while (remainingSize) |
---|
76 | { |
---|
77 | switch (remainingSize) |
---|
78 | { |
---|
79 | case 1: |
---|
80 | case 2: |
---|
81 | case 4: |
---|
82 | case 8: |
---|
83 | case 16: |
---|
84 | { |
---|
85 | mfc_get(remainingTmpTarget,remainingEa,remainingSize,DMA_TAG(1),0,0); |
---|
86 | remainingSize=0; |
---|
87 | break; |
---|
88 | } |
---|
89 | default: |
---|
90 | { |
---|
91 | //spu_printf("unaligned DMA with non-natural size:%d\n",remainingSize); |
---|
92 | int actualSize = 0; |
---|
93 | |
---|
94 | if (remainingSize > 16) |
---|
95 | actualSize = 16; |
---|
96 | else |
---|
97 | if (remainingSize >8) |
---|
98 | actualSize=8; |
---|
99 | else |
---|
100 | if (remainingSize >4) |
---|
101 | actualSize=4; |
---|
102 | else |
---|
103 | if (remainingSize >2) |
---|
104 | actualSize=2; |
---|
105 | mfc_get(remainingTmpTarget,remainingEa,actualSize,DMA_TAG(1),0,0); |
---|
106 | remainingSize-=actualSize; |
---|
107 | remainingTmpTarget+=actualSize; |
---|
108 | remainingEa += actualSize; |
---|
109 | } |
---|
110 | } |
---|
111 | } |
---|
112 | #endif//FORCE_cellDmaUnalignedGet |
---|
113 | |
---|
114 | #else |
---|
115 | //copy into final destination |
---|
116 | #ifdef USE_MEMCPY |
---|
117 | memcpy(tmpTarget,mainMem,size); |
---|
118 | #else |
---|
119 | for ( i=0;i<size;i++) |
---|
120 | { |
---|
121 | tmpTarget[i] = mainMem[i]; |
---|
122 | } |
---|
123 | #endif //USE_MEMCPY |
---|
124 | |
---|
125 | #endif |
---|
126 | |
---|
127 | cellDmaWaitTagStatusAll(DMA_MASK(1)); |
---|
128 | |
---|
129 | //this is slowish, perhaps memcpy on SPU is smarter? |
---|
130 | for (i=0; btLikely( i<size );i++) |
---|
131 | { |
---|
132 | localStore[i] = tmpTarget[i]; |
---|
133 | } |
---|
134 | |
---|
135 | return 0; |
---|
136 | } |
---|
137 | |
---|
138 | #if defined (__SPU__) || defined (USE_LIBSPE2) |
---|
139 | #else |
---|
140 | |
---|
141 | int cellDmaLargeGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid) |
---|
142 | { |
---|
143 | char* mainMem = (char*)ea; |
---|
144 | char* localStore = (char*)ls; |
---|
145 | |
---|
146 | #ifdef USE_MEMCPY |
---|
147 | memcpy(localStore,mainMem,size); |
---|
148 | #else |
---|
149 | for (uint32_t i=0;i<size;i++) |
---|
150 | { |
---|
151 | localStore[i] = mainMem[i]; |
---|
152 | } |
---|
153 | #endif |
---|
154 | return 0; |
---|
155 | } |
---|
156 | |
---|
157 | int cellDmaGet(void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid) |
---|
158 | { |
---|
159 | char* mainMem = (char*)ea; |
---|
160 | char* localStore = (char*)ls; |
---|
161 | #ifdef USE_MEMCPY |
---|
162 | memcpy(localStore,mainMem,size); |
---|
163 | #else |
---|
164 | for (uint32_t i=0;i<size;i++) |
---|
165 | { |
---|
166 | localStore[i] = mainMem[i]; |
---|
167 | } |
---|
168 | #endif //#ifdef USE_MEMCPY |
---|
169 | return 0; |
---|
170 | } |
---|
171 | |
---|
172 | int cellDmaLargePut(const void *ls, uint64_t ea, uint32_t size, uint32_t tag, uint32_t tid, uint32_t rid) |
---|
173 | { |
---|
174 | char* mainMem = (char*)ea; |
---|
175 | const char* localStore = (const char*)ls; |
---|
176 | #ifdef USE_MEMCPY |
---|
177 | memcpy(mainMem,localStore,size); |
---|
178 | #else |
---|
179 | for (uint32_t i=0;i<size;i++) |
---|
180 | { |
---|
181 | mainMem[i] = localStore[i]; |
---|
182 | } |
---|
183 | #endif //#ifdef USE_MEMCPY |
---|
184 | |
---|
185 | return 0; |
---|
186 | } |
---|
187 | |
---|
188 | |
---|
189 | |
---|
190 | void cellDmaWaitTagStatusAll(int ignore) |
---|
191 | { |
---|
192 | |
---|
193 | } |
---|
194 | |
---|
195 | #endif |
---|