wiki:GcnInstrsDs

Version 21 (modified by trac, 8 years ago) (diff)

--

Back to Table of content

GCN ISA DS instructions

These instructions access to local or global data share (LDS/GDS) memory.

List of fields for the DS encoding:

Bits Name Description
0-7 OFFSET0 8-bit offset 0 for VDATA0
8-15 OFFSET1 8-bit offset 1 for VDATA1
0-15 OFFSET 16-bit offset for single DATA
17 GDS Access to GDS instead LDS
18-25 OPCODE Operation code (GCN 1.0/1.1)
17-24 OPCODE Operation code (GCN 1.2)
26-31 ENCODING Encoding type. Must be 0b110110
32-39 ADDR Vector register that holds byte address
40-47 VDATA0 Source data 0 vector register
48-55 VDATA1 Source data 1 vector register
56-63 VDST Destination vector register

Syntax: INSTRUCTION ADDR, VDATA0 [offset:OFFSET] [GDS]
Syntax: INSTRUCTION ADDR, VDATA0, VDATA1 [offset0:OFFSET0] [offset1:OFFSET1] [GDS]
VDST syntax: INSTRUCTION VDST, ADDR [offset:OFFSET] [GDS]

NOTE: Any operation on LDS requires correctly set M0 register, prior to execution. The M0 register holds maximum size of a LDS memory, that can be accessed by wavefront. To set no limits, just set a M0 register to 0xffffffff.

NOTE: Any operation on GDS requires correctly set M0 register, prior to execution. The M0 register holds maximum size of a GDS memory in bits 0-15, that can be accessed by kernel, and a GDS base offset (in bytes) in bits 16-31.

Any DS instruction return data in order (including DS_SWIZZLE_B32) and increments LGKM_CNT. Any operation increments LGKM by one, and decremented by one if it will be finished.

List of the instructions by opcode:

Opcode GCN 1.0 GCN 1.1 Mnemonic (GCN 1.0/1.1) Mnemonic (GCN 1.2)
0 (0x0) DS_ADD_U32 DS_ADD_U32
1 (0x1) DS_SUB_U32 DS_SUB_U32
2 (0x2) DS_RSUB_U32 DS_RSUB_U32
3 (0x3) DS_INC_U32 DS_INC_U32
4 (0x4) DS_DEC_U32 DS_DEC_U32
5 (0x5) DS_MIN_I32 DS_MIN_I32
6 (0x6) DS_MAX_I32 DS_MAX_I32
7 (0x7) DS_MIN_U32 DS_MIN_U32
8 (0x8) DS_MAX_U32 DS_MAX_U32
9 (0x9) DS_AND_B32 DS_AND_B32
10 (0xa) DS_OR_B32 DS_OR_B32
11 (0xb) DS_XOR_B32 DS_XOR_B32
12 (0xc) DS_MSKOR_B32 DS_MSKOR_B32
13 (0xd) DS_WRITE_B32 DS_WRITE_B32
14 (0xe) DS_WRITE2_B32 DS_WRITE2_B32
15 (0xf) DS_WRITE2ST64_B32 DS_WRITE2ST64_B32
16 (0x10) DS_CMPST_B32 DS_CMPST_B32
17 (0x11) DS_CMPST_F32 DS_CMPST_F32
18 (0x12) DS_MIN_F32 DS_MIN_F32
19 (0x13) DS_MAX_F32 DS_MAX_F32
20 (0x14) DS_NOP DS_NOP
21 (0x15) -- DS_ADD_F32
24 (0x18) DS_GWS_SEMA_RELEASE_ALL --
25 (0x19) DS_GWS_INIT --
26 (0x1a) DS_GWS_SEMA_V --
27 (0x1b) DS_GWS_SEMA_BR --
28 (0x1c) DS_GWS_SEMA_P --
29 (0x1d) DS_GWS_BARRIER --
30 (0x1e) DS_WRITE_B8 DS_WRITE_B8
31 (0x1f) DS_WRITE_B16 DS_WRITE_B16
32 (0x20) DS_ADD_RTN_U32 DS_ADD_RTN_U32
33 (0x21) DS_SUB_RTN_U32 DS_SUB_RTN_U32
34 (0x22) DS_RSUB_RTN_U32 DS_RSUB_RTN_U32
35 (0x23) DS_INC_RTN_U32 DS_INC_RTN_U32
36 (0x24) DS_DEC_RTN_U32 DS_DEC_RTN_U32
37 (0x25) DS_MIN_RTN_I32 DS_MIN_RTN_I32
38 (0x26) DS_MAX_RTN_I32 DS_MAX_RTN_I32
39 (0x27) DS_MIN_RTN_U32 DS_MIN_RTN_U32
40 (0x28) DS_MAX_RTN_U32 DS_MAX_RTN_U32
41 (0x29) DS_AND_RTN_B32 DS_AND_RTN_B32
42 (0x2a) DS_OR_RTN_B32 DS_OR_RTN_B32
43 (0x2b) DS_XOR_RTN_B32 DS_XOR_RTN_B32
44 (0x2c) DS_MSKOR_RTN_B32 DS_MSKOR_RTN_B32
45 (0x2d) DS_WRXCHG_RTN_B32 DS_WRXCHG_RTN_B32
46 (0x2e) DS_WRXCHG2_RTN_B32 DS_WRXCHG2_RTN_B32
47 (0x2f) DS_WRXCHG2ST64_RTN_B32 DS_WRXCHG2ST64_RTN_B32
48 (0x30) DS_CMPST_RTN_B32 DS_CMPST_RTN_B32
49 (0x31) DS_CMPST_RTN_F32 DS_CMPST_RTN_F32
50 (0x32) DS_MIN_RTN_F32 DS_MIN_RTN_F32
51 (0x33) DS_MAX_RTN_F32 DS_MAX_RTN_F32
52 (0x34) DS_WRAP_RTN_B32 DS_WRAP_RTN_B32
53 (0x35) DS_SWIZZLE_B32 DS_ADD_RTN_F32
54 (0x36) DS_READ_B32 DS_READ_B32
55 (0x37) DS_READ2_B32 DS_READ2_B32
56 (0x38) DS_READ2ST64_B32 DS_READ2ST64_B32
57 (0x39) DS_READ_I8 DS_READ_I8
58 (0x3a) DS_READ_U8 DS_READ_U8
59 (0x3b) DS_READ_I16 DS_READ_I16
60 (0x3c) DS_READ_U16 DS_READ_U16
61 (0x3d) DS_CONSUME DS_SWIZZLE_B32
62 (0x3e) DS_APPEND DS_PERMUTE_B32
63 (0x3f) DS_ORDERED_COUNT (???) DS_BPERMUTE_B32
64 (0x40) DS_ADD_U64 DS_ADD_U64
65 (0x41) DS_SUB_U64 DS_SUB_U64
66 (0x42) DS_RSUB_U64 DS_RSUB_U64
67 (0x43) DS_INC_U64 DS_INC_U64
68 (0x44) DS_DEC_U64 DS_DEC_U64
69 (0x45) DS_MIN_I64 DS_MIN_I64
70 (0x46) DS_MAX_I64 DS_MAX_I64
71 (0x47) DS_MIN_U64 DS_MIN_U64
72 (0x48) DS_MAX_U64 DS_MAX_U64
73 (0x49) DS_AND_B64 DS_AND_B64
74 (0x4a) DS_OR_B64 DS_OR_B64
75 (0x4b) DS_XOR_B64 DS_XOR_B64
76 (0x4c) DS_MSKOR_B64 DS_MSKOR_B64
77 (0x4d) DS_WRITE_B64 DS_WRITE_B64
78 (0x4e) DS_WRITE2_B64 DS_WRITE2_B64
79 (0x4f) DS_WRITE2ST64_B64 DS_WRITE2ST64_B64
80 (0x50) DS_CMPST_B64 DS_CMPST_B64
81 (0x51) DS_CMPST_F64 DS_CMPST_F64
82 (0x52) DS_MIN_F64 DS_MIN_F64
83 (0x53) DS_MAX_F64 DS_MAX_F64
96 (0x60) DS_ADD_RTN_U64 DS_ADD_RTN_U64
97 (0x61) DS_SUB_RTN_U64 DS_SUB_RTN_U64
98 (0x62) DS_RSUB_RTN_U64 DS_RSUB_RTN_U64
99 (0x63) DS_INC_RTN_U64 DS_INC_RTN_U64
100 (0x64) DS_DEC_RTN_U64 DS_DEC_RTN_U64
101 (0x65) DS_MIN_RTN_I64 DS_MIN_RTN_I64
102 (0x66) DS_MAX_RTN_I64 DS_MAX_RTN_I64
103 (0x67) DS_MIN_RTN_U64 DS_MIN_RTN_U64
104 (0x68) DS_MAX_RTN_U64 DS_MAX_RTN_U64
105 (0x69) DS_AND_RTN_B64 DS_AND_RTN_B64
106 (0x6a) DS_OR_RTN_B64 DS_OR_RTN_B64
107 (0x6b) DS_XOR_RTN_B64 DS_XOR_RTN_B64
108 (0x6c) DS_MSKOR_RTN_B64 DS_MSKOR_RTN_B64
109 (0x6d) DS_WRXCHG_RTN_B64 DS_WRXCHG_RTN_B64
110 (0x6e) DS_WRXCHG2_RTN_B64 DS_WRXCHG2_RTN_B64
111 (0x6f) DS_WRXCHG2ST64_RTN_B64 DS_WRXCHG2ST64_RTN_B64
112 (0x70) DS_CMPST_RTN_B64 DS_CMPST_RTN_B64
113 (0x71) DS_CMPST_RTN_F64 DS_CMPST_RTN_F64
114 (0x72) DS_MIN_RTN_F64 DS_MIN_RTN_F64
115 (0x73) DS_MAX_RTN_F64 DS_MAX_RTN_F64
118 (0x76) DS_READ_B64 DS_READ_B64
119 (0x77) DS_READ2_B64 DS_READ2_B64
120 (0x78) DS_READ2ST64_B64 DS_READ2ST64_B64
126 (0x7e) DS_CONDXCHG32_RTN_B64 DS_CONDXCHG32_RTN_B64
128 (0x80) DS_ADD_SRC2_U32 DS_ADD_SRC2_U32
129 (0x81) DS_SUB_SRC2_U32 DS_SUB_SRC2_U32
130 (0x82) DS_RSUB_SRC2_U32 DS_RSUB_SRC2_U32
131 (0x83) DS_INC_SRC2_U32 DS_INC_SRC2_U32
132 (0x84) DS_DEC_SRC2_U32 DS_DEC_SRC2_U32
133 (0x85) DS_MIN_SRC2_I32 DS_MIN_SRC2_I32
134 (0x86) DS_MAX_SRC2_I32 DS_MAX_SRC2_I32
135 (0x87) DS_MIN_SRC2_U32 DS_MIN_SRC2_U32
136 (0x88) DS_MAX_SRC2_U32 DS_MAX_SRC2_U32
137 (0x89) DS_AND_SRC2_B32 DS_AND_SRC2_B32
138 (0x8a) DS_OR_SRC2_B32 DS_OR_SRC2_B32
139 (0x8b) DS_XOR_SRC2_B32 DS_XOR_SRC2_B32
141 (0x8d) DS_WRITE_SRC2_B32 DS_WRITE_SRC2_B32
146 (0x92) DS_MIN_SRC2_F32 DS_MIN_SRC2_F32
147 (0x93) DS_MAX_SRC2_F32 DS_MAX_SRC2_F32
149 (0x95) -- DS_ADD_SRC2_F32
152 (0x98) -- DS_GWS_SEMA_RELEASE_ALL
153 (0x99) -- DS_GWS_INIT
154 (0x9a) -- DS_GWS_SEMA_V
155 (0x9b) -- DS_GWS_SEMA_BR
156 (0x9c) -- DS_GWS_SEMA_P
157 (0x9d) -- DS_GWS_BARRIER
189 (0xbd) -- DS_CONSUME
190 (0xbe) -- DS_APPEND
191 (0xbf) -- DS_ORDERED_COUNT
192 (0xc0) DS_ADD_SRC2_U64 DS_ADD_SRC2_U64
193 (0xc1) DS_SUB_SRC2_U64 DS_SUB_SRC2_U64
194 (0xc2) DS_RSUB_SRC2_U64 DS_RSUB_SRC2_U64
195 (0xc3) DS_INC_SRC2_U64 DS_INC_SRC2_U64
196 (0xc4) DS_DEC_SRC2_U64 DS_DEC_SRC2_U64
197 (0xc5) DS_MIN_SRC2_I64 DS_MIN_SRC2_I64
198 (0xc6) DS_MAX_SRC2_I64 DS_MAX_SRC2_I64
199 (0xc7) DS_MIN_SRC2_U64 DS_MIN_SRC2_U64
200 (0xc8) DS_MAX_SRC2_U64 DS_MAX_SRC2_U64
201 (0xc9) DS_AND_SRC2_B64 DS_AND_SRC2_B64
202 (0xca) DS_OR_SRC2_B64 DS_OR_SRC2_B64
203 (0xcb) DS_XOR_SRC2_B64 DS_XOR_SRC2_B64
205 (0xcd) DS_WRITE_SRC2_B64 DS_WRITE_SRC2_B64
210 (0xd2) DS_MIN_SRC2_F64 DS_MIN_SRC2_F64
211 (0xd3) DS_MAX_SRC2_F64 DS_MAX_SRC2_F64
222 (0xde) DS_WRITE_B96 DS_WRITE_B96
223 (0xdf) DS_WRITE_B128 DS_WRITE_B128
253 (0xfd) DS_CONDXCHG32_RTN_B128 DS_CONDXCHG32_RTN_B128
254 (0xfe) DS_READ_B96 DS_READ_B96
255 (0xff) DS_READ_B128 DS_READ_B128

Instruction set

Alphabetically sorted instruction list:

DS_ADD_RTN_U32

Opcode: 32 (0x20)
Syntax: DS_ADD_RTN_U32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Add unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result back to LDS/GDS at this address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = *V + VDATA0 // atomic operation

DS_ADD_RTN_U64

Opcode: 96 (0x60)
Syntax: DS_ADD_RTN_U64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Add unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result back to LDS/GDS at this address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = *V + VDATA0 // atomic operation

DS_ADD_SRC2_U32

Opcode: 128 (0x80)
Syntax: DS_ADD_SRC2_U32 ADDR [OFFSET:OFFSET]
Description: Add unsigned integer value from LDS/GDS at address A and B, and store result back to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = *V + *(UINT32*)(DS + B) // atomic operation

DS_ADD_SRC2_U64

Opcode: 192 (0xc0)
Syntax: DS_ADD_SRC2_U64 ADDR [OFFSET:OFFSET]
Description: Add unsigned 64-bit integer value from LDS/GDS at address A and B, and store result back to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = *V + *(UINT64*)(DS + B) // atomic operation

DS_ADD_U32

Opcode: 0 (0x0)
Syntax: DS_ADD_U32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Add unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result back to LDS/GDS at this address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = *V + VDATA0 // atomic operation

DS_ADD_U64

Opcode: 64 (0x40)
Syntax: DS_ADD_U64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Add unsigned integer 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result back to LDS/GDS at this address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = *V + VDATA0 // atomic operation

DS_AND_B32

Opcode: 9 (0x9)
Syntax: DS_AND_B32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Do bitwise AND operation on 32-bit value from LDS/GDS at address (ADDR+OFFSET) & ~3, and value from VDATA0; and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = *V & VDATA0 // atomic operation

DS_AND_B64

Opcode: 73 (0x49)
Syntax: DS_AND_B64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Do bitwise AND operation on 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and value from VDATA0; and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = *V & VDATA0 // atomic operation

DS_AND_RTN_B32

Opcode: 41 (0x29)
Syntax: DS_AND_RTN_B32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Do bitwise AND operation on 32-bit value from LDS/GDS at address (ADDR+OFFSET) & ~3, and value from VDATA0; and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = *V & VDATA0 // atomic operation

DS_AND_RTN_B64

Opcode: 105 (0x69)
Syntax: DS_AND_RTN_B64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Do bitwise AND operation on 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and value from VDATA0; and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = *V & VDATA0 // atomic operation

DS_AND_SRC2_B32

Opcode: 137 (0x89)
Syntax: DS_AND_SRC2_B32 ADDR [OFFSET:OFFSET]
Description: Do bitwise AND operation on 32-bit value from LDS/GDS at address A, and at address B; and store result to LDS/GDS at address A. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = *V & *(UINT32*)(DS + B) // atomic operation

DS_AND_SRC2_B64

Opcode: 201 (0xc9)
Syntax: DS_AND_SRC2_B64 ADDR [OFFSET:OFFSET]
Description: Do bitwise AND operation on 64-bit value from LDS/GDS at address A, and at address B; and store result to LDS/GDS at address A. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = *V & *(UINT64*)(DS + B) // atomic operation

DS_APPEND

Opcode: 62 (0x3e) for GCN 1.0/1.1; 190 (0xbe) GCN 1.2
Syntax: DS_APPEND VDST [OFFSET:OFFSET]
Description: Append entries to buffer. This instruction increments 32-bit value in LDS/GDS at address OFFSET&~3 by number of the active threads, and store previous value from LDS/GDS at this same address into VDST.
Operation:
UINT32* V = (UINT32*)(DS + (OFFSET&~3)) VDST = *V // scalar operation *V += BITCOUNT(EXEC) // scalar operation

DS_CONSUME

Opcode: 61 (0x3d) for GCN 1.0/1.1; 189 (0xbd) GCN 1.2
Syntax: DS_CONSUME VDST [OFFSET:OFFSET]
Description: Consume entries to buffer. This instruction increments 32-bit value in LDS/GDS at address OFFSET&~3 by number of the active threads, and store previous value from LDS/GDS at this same address into VDST.
Operation:
UINT32* V = (UINT32*)(DS + (OFFSET&~3)) VDST = *V // scalar operation *V -= BITCOUNT(EXEC) // scalar operation

DS_CMPST_B32

Opcode: 16 (0x10)
Syntax: DS_CMPST_B32 ADDR, VDATA0, VDATA1 [OFFSET:OFFSET]
Description: Compare values from VDATA0 and from LDS/GDS at address (ADDR+OFFSET) & ~3. If values are equal, store VDATA1 to LDS/GDS at this same address, otherwise do nothing. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = (*V==VDATA0) ? VDATA1 : *V // atomic operation

DS_CMPST_B64

Opcode: 80 (0x50)
Syntax: DS_CMPST_B64 ADDR, VDATA0(2), VDATA1(2) [OFFSET:OFFSET]
Description: Compare 64-bit values from VDATA0 and from LDS/GDS at address (ADDR+OFFSET) & ~7. If values are equal, store VDATA1 to LDS/GDS at this same address, otherwise do nothing. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = (*V==VDATA0) ? VDATA1 : *V // atomic operation

DS_CMPST_F32

Opcode: 17 (0x11)
Syntax: DS_CMPST_F32 ADDR, VDATA0, VDATA1 [OFFSET:OFFSET]
Description: Compare single floating point values from VDATA0 and from LDS/GDS at address (ADDR+OFFSET) & ~3. If values are equal, store VDATA1 to LDS/GDS at this same address, otherwise do nothing. Operation is atomic.
Operation:
FLOAT* V = (FLOAT*)(DS + ((ADDR+OFFSET)&~3)) *V = (*V==ASFLOATR(VDATA0)) ? ASFLOAT(VDATA1) : *V // atomic operation

DS_CMPST_F64

Opcode: 81 (0x51)
Syntax: DS_CMPST_F64 ADDR, VDATA0(2), VDATA1(2) [OFFSET:OFFSET]
Description: Compare double floating point values from VDATA0 and from LDS/GDS at address (ADDR+OFFSET) & ~7. If values are equal, store VDATA1 to LDS/GDS at this same address, otherwise do nothing. Operation is atomic.
Operation:
DOUBLE* V = (DOUBLE*)(DS + ((ADDR+OFFSET)&~7)) *V = (*V==ASDOUBLE(VDATA0)) ? ASDOUBLE(VDATA1) : *V // atomic operation

DS_CMPST_RTN_B32

Opcode: 48 (0x30)
Syntax: DS_CMPST_RTN_B32 VDST, ADDR, VDATA0, VDATA1 [OFFSET:OFFSET]
Description: Compare values from VDATA0 and from LDS/GDS at address (ADDR+OFFSET) & ~3. If values are equal, store VDATA1 to LDS/GDS at this same address, otherwise do nothing. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = VDST; *V = (*V==VDATA0) ? VDATA1 : *V // atomic operation

DS_CMPST_RTN_B64

Opcode: 112 (0x70)
Syntax: DS_CMPST_RTN_B64 VDST(2), ADDR, VDATA0(2), VDATA1(2) [OFFSET:OFFSET]
Description: Compare 64-bit values from VDATA0 and from LDS/GDS at address (ADDR+OFFSET) & ~7. If values are equal, store VDATA1 to LDS/GDS at this same address, otherwise do nothing. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = VDST; *V = (*V==VDATA0) ? VDATA1 : *V // atomic operation

DS_CMPST_RTN_F32

Opcode: 49 (0x31)
Syntax: DS_CMPST_RTN_F32 VDST, ADDR, VDATA0, VDATA1 [OFFSET:OFFSET]
Description: Compare single floating point values from VDATA0 and from LDS/GDS at address (ADDR+OFFSET) & ~3. If values are equal, store VDATA1 to LDS/GDS at this same address, otherwise do nothing. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
FLOAT* V = (FLOAT*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = (*V==ASFLOAT(VDATA0)) ? ASFLOAT(VDATA1) : *V // atomic operation

DS_CMPST_RTN_F64

Opcode: 113 (0x71)
Syntax: DS_CMPST_RTN_F64 VDST(2), ADDR, VDATA0(2), VDATA1(2) [OFFSET:OFFSET]
Description: Compare double floating point values from VDATA0 and from LDS/GDS at address (ADDR+OFFSET) & ~7. If values are equal, store VDATA1 to LDS/GDS at this same address, otherwise do nothing. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
DOUBLE* V = (DOUBLE*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = (*V==ASDOUBLE(VDATA0)) ? ASDOUBLE(VDATA1) : *V // atomic operation

DS_DEC_RTN_U32

Opcode: 36 (0x24)
Syntax: DS_DEC_RTN_U32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Load unsigned value from LDS/GDS at address (ADDR+OFFSET) & ~3, and compare with unsigned value from VDATA0. If VDATA0 is greater or equal and loaded unsigned value is not zero, then increment value from LDS/GDS, otherwise store VDATA0 to LDS/GDS. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = (VDATA0 >= *V && *V!=0) ? *V-1 : VDATA0 // atomic operation

DS_DEC_RTN_U64

Opcode: 100 (0x64)
Syntax: DS_DEC_RTN_U64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Load unsigned 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and compare with unsigned value from VDATA0. If VDATA0 is greater or equal and loaded unsigned value is not zero, then decrement value from LDS/GDS, otherwise store VDATA0 to LDS/GDS. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = (VDATA0 >= *V && *V!=0) ? *V-1 : VDATA0 // atomic operation

DS_DEC_SRC2_U32

Opcode: 132 (0x84)
Syntax: DS_DEC_SRC2_U32 ADDR [OFFSET:OFFSET]
Description: Load unsigned value from LDS/GDS at address A, and compare with unsigned value at address B. If value at address B is greater or equal and loaded unsigned value is not zero, then decrement value from LDS/GDS, otherwise store value at address B to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) UINT32* VB = (UINT32*)(DS + B) *V = (*VB >= *V && *V!=0) ? *V-1 : *VB // atomic operation

DS_DEC_SRC2_U64

Opcode: 196 (0xc4)
Syntax: DS_DEC_SRC2_U64 ADDR [OFFSET:OFFSET]
Description: Load unsigned 64-bit value from LDS/GDS at address A, and compare with unsigned 64-bit value at address B. If value at address B is greater or equal and loaded unsigned value is not zero, then decrement value from LDS/GDS, otherwise store value at address B to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) UINT64* VB = (UINT64*)(DS + B) *V = (*VB >= *V && *V!=0) ? *V-1 : *VB // atomic operation

DS_DEC_U32

Opcode: 4 (0x4)
Syntax: DS_DEC_U32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Load unsigned value from LDS/GDS at address (ADDR+OFFSET) & ~3, and compare with unsigned value from VDATA0. If VDATA0 is greater or equal and loaded unsigned value is not zero, then decrement value from LDS/GDS, otherwise store VDATA0 to LDS/GDS. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = (VDATA0 >= *V && *V!=0) ? *V-1 : VDATA0 // atomic operation

DS_DEC_U64

Opcode: 68 (0x44)
Syntax: DS_DEC_U64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Load unsigned 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and compare with unsigned value from VDATA0. If VDATA0 is greater or equal and loaded unsigned value is not zero, then decrement value from LDS/GDS, otherwise store VDATA0 to LDS/GDS. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = (VDATA0 >= *V && *V!=0) ? *V-1 : VDATA0 // atomic operation

DS_INC_RTN_U32

Opcode: 35 (0x23)
Syntax: DS_INC_RTN_U32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Load unsigned value from LDS/GDS at address (ADDR+OFFSET) & ~3, and compare with unsigned value from VDATA0. If VDATA0 greater, then increment value from LDS/GDS, otherwise store 0 to LDS/GDS. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = (VDATA0 > *V) ? *V+1 : 0 // atomic operation

DS_INC_RTN_U64

Opcode: 99 (0x63)
Syntax: DS_INC_RTN_U64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Load unsigned 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and compare with unsigned value from VDATA0. If VDATA0 is greater, then increment value from LDS/GDS, otherwise store 0 to LDS/GDS. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = (VDATA0 > *V) ? *V+1 : 0 // atomic operation

DS_INC_SRC2_U32

Opcode: 131 (0x83)
Syntax: DS_INC_SRC2_U32 ADDR [OFFSET:OFFSET]
Description: Load unsigned value from LDS/GDS at address A, and compare with unsigned value at address B. If value at address B is greater, then increment value from LDS/GDS, otherwise store 0 to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = (*(UINT32*)(DS + B) > *V) ? *V+1 : 0 // atomic operation

DS_INC_SRC2_U64

Opcode: 195 (0xc3)
Syntax: DS_INC_SRC2_U64 ADDR [OFFSET:OFFSET]
Description: Load unsigned 64-bit value from LDS/GDS at address A, and compare with unsigned value at address B. If value at address B is greater, then increment value from LDS/GDS, otherwise store 0 to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = (*(UINT64*)(DS + B) > *V) ? *V+1 : 0 // atomic operation

DS_INC_U32

Opcode: 3 (0x3)
Syntax: DS_INC_U32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Load unsigned value from LDS/GDS at address (ADDR+OFFSET) & ~3, and compare with unsigned value from VDATA0. If VDATA0 is greater, then increment value from LDS/GDS, otherwise store 0 to LDS/GDS. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = (VDATA0 > *V) ? *V+1 : 0 // atomic operation

DS_INC_U64

Opcode: 67 (0x43)
Syntax: DS_INC_U64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Load unsigned 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and compare with unsigned value from VDATA0. If VDATA0 is greater, then increment value from LDS/GDS, otherwise store 0 to LDS/GDS. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = (VDATA0 > *V) ? *V+1 : 0 // atomic operation

DS_MAX_F32

Opcode: 19 (0x13)
Syntax: DS_MAX_F32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose greatest single floating point value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
FLOAT* V = (FLOAT*)(DS + ((ADDR+OFFSET)&~3)) *V = MAX(*V, ASFLOAT(VDATA0)) // atomic operation

DS_MAX_F64

Opcode: 83 (0x53)
Syntax: DS_MAX_F64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose greatest double floating point value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
DOUBLE* V = (DOUBLE*)(DS + ((ADDR+OFFSET)&~7)) *V = MAX(*V, ASDOUBLE(VDATA0)) // atomic operation

DS_MAX_I32

Opcode: 6 (0x6)
Syntax: DS_MAX_I32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose greatest signed integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
INT32* V = (INT32*)(DS + ((ADDR+OFFSET)&~3)) *V = MAX(*V, (INT32)VDATA0) // atomic operation

DS_MAX_I64

Opcode: 70 (0x46)
Syntax: DS_MAX_I64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose greatest signed 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
INT64* V = (INT64*)(DS + ((ADDR+OFFSET)&~7)) *V = MAX(*V, (INT64)VDATA0) // atomic operation

DS_MAX_RTN_F32

Opcode: 51 (0x33)
Syntax: DS_MAX_F32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose smallest single floating point value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
FLOAT* V = (FLOAT*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = MAX(*V, ASFLOAT(VDATA0)) // atomic operation

DS_MAX_RTN_F64

Opcode: 115 (0x73)
Syntax: DS_MAX_F64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose smallest double floating point value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
DOUBLE* V = (DOUBLE*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = MAX(*V, ASDOUBLE(VDATA0)) // atomic operation

DS_MAX_RTN_I32

Opcode: 38 (0x26)
Syntax: DS_MAX_RTN_I32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose greatest signed integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
INT32* V = (INT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = MAX(*V, (INT32)VDATA0) // atomic operation

DS_MAX_RTN_I64

Opcode: 102 (0x66)
Syntax: DS_MAX_RTN_I64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose greatest signed 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
INT64* V = (INT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = MAX(*V, (INT64)VDATA0) // atomic operation

DS_MAX_RTN_U32

Opcode: 40 (0x28)
Syntax: DS_MAX_RTN_U32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose greatest unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = MAX(*V, VDATA0) // atomic operation

DS_MAX_RTN_U64

Opcode: 104 (0x68)
Syntax: DS_MAX_RTN_U64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose greatest unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = MAX(*V, VDATA0) // atomic operation

DS_MAX_SRC2_F32

Opcode: 147 (0x93)
Syntax: DS_MAX_SRC2_F32 ADDR [OFFSET:OFFSET]
Description: Choose greatest single floating point value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 FLOAT* V = (FLOAT*)(DS + A) *V = MAX(*V, *(FLOAT*)(DS + B)) // atomic operation

DS_MAX_SRC2_F64

Opcode: 211 (0xd3)
Syntax: DS_MAX_SRC2_F64 ADDR [OFFSET:OFFSET]
Description: Choose greatest double floating point value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 DOUBLE* V = (DOUBLE*)(DS + A) *V = MAX(*V, *(DOUBLE*)(DS + B)) // atomic operation

DS_MAX_SRC2_I32

Opcode: 134 (0x86)
Syntax: DS_MAX_SRC2_I32 ADDR [OFFSET:OFFSET]
Description: Choose greatest signed integer value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 INT32* V = (INT32*)(DS + A) *V = MAX(*V, *(INT32*)(DS + B)) // atomic operation

DS_MAX_SRC2_I64

Opcode: 198 (0x86)
Syntax: DS_MAX_SRC2_I64 ADDR [OFFSET:OFFSET]
Description: Choose greatest signed 64-bit integer value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 INT64* V = (INT64*)(DS + A) *V = MAX(*V, *(INT64*)(DS + B)) // atomic operation

DS_MAX_SRC2_U32

Opcode: 136 (0x88)
Syntax: DS_MAX_SRC2_U32 ADDR [OFFSET:OFFSET]
Description: Choose greatest unsigned integer value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = MAX(*V, *(UINT32*)(DS + B)) // atomic operation

DS_MAX_SRC2_U64

Opcode: 200 (0xc8)
Syntax: DS_MAX_SRC2_U64 ADDR [OFFSET:OFFSET]
Description: Choose greatest unsigned 64-bit integer value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = MAX(*V, *(UINT64*)(DS + B)) // atomic operation

DS_MAX_U32

Opcode: 8 (0x8)
Syntax: DS_MAX_U32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose greatest unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = MAX(*V, VDATA0) // atomic operation

DS_MAX_U64

Opcode: 72 (0x48)
Syntax: DS_MAX_U64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose greatest unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = MAX(*V, VDATA0) // atomic operation

DS_MIN_F32

Opcode: 18 (0x12)
Syntax: DS_MIN_F32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose smallest single floating point value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
FLOAT* V = (FLOAT*)(DS + ((ADDR+OFFSET)&~3)) *V = MIN(*V, ASFLOAT(VDATA0)) // atomic operation

DS_MIN_F64

Opcode: 82 (0x52)
Syntax: DS_MIN_F64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose smallest double floating point value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
DOUBLE* V = (DOUBLE*)(DS + ((ADDR+OFFSET)&~7)) *V = MIN(*V, ASDOUBLE(VDATA0)) // atomic operation

DS_MIN_I32

Opcode: 5 (0x5)
Syntax: DS_MIN_I32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose smallest signed integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
INT32* V = (INT32*)(DS + ((ADDR+OFFSET)&~3)) *V = MIN(*V, (INT32)VDATA0) // atomic operation

DS_MIN_I64

Opcode: 69 (0x45)
Syntax: DS_MIN_I64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose smallest signed 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
INT64* V = (INT64*)(DS + ((ADDR+OFFSET)&~7)) *V = MIN(*V, (INT64)VDATA0) // atomic operation

DS_MIN_RTN_F32

Opcode: 50 (0x32)
Syntax: DS_MIN_F32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose smallest single floating point value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
FLOAT* V = (FLOAT*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = MIN(*V, ASFLOAT(VDATA0)) // atomic operation

DS_MIN_RTN_F64

Opcode: 114 (0x72)
Syntax: DS_MIN_F64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose smallest double floating point value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
DOUBLE* V = (FLOAT*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = MIN(*V, ASDOUBLE(VDATA0)) // atomic operation

DS_MIN_RTN_I32

Opcode: 37 (0x25)
Syntax: DS_MIN_RTN_I32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose smallest signed integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
INT32* V = (INT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = MIN(*V, (INT32)VDATA0) // atomic operation

DS_MIN_RTN_I64

Opcode: 101 (0x65)
Syntax: DS_MIN_RTN_I64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose smallest signed 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
INT64 V = (INT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = MIN(*V, (INT64)VDATA0) // atomic operation

DS_MIN_RTN_U32

Opcode: 39 (0x27)
Syntax: DS_MIN_RTN_U32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose smallest unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = MIN(*V, VDATA0) // atomic operation

DS_MIN_RTN_U64

Opcode: 103 (0x67)
Syntax: DS_MIN_RTN_U64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose smallest unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = MIN(*V, VDATA0) // atomic operation

DS_MIN_SRC2_F32

Opcode: 146 (0x92)
Syntax: DS_MIN_SRC2_F32 ADDR [OFFSET:OFFSET]
Description: Choose smallest single floating point value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 FLOAT* V = (FLOAT*)(DS + A) *V = MIN(*V, *(FLOAT*)(DS + B)) // atomic operation

DS_MIN_SRC2_F64

Opcode: 210 (0xd2)
Syntax: DS_MIN_SRC2_F64 ADDR [OFFSET:OFFSET]
Description: Choose smallest double floating point value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 DOUBLE* V = (DOUBLE*)(DS + A) *V = MIN(*V, *(DOUBLE*)(DS + B)) // atomic operation

DS_MIN_SRC2_I32

Opcode: 133 (0x85)
Syntax: DS_MIN_SRC2_I32 ADDR [OFFSET:OFFSET]
Description: Choose smallest signed integer value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 INT32* V = (INT32*)(DS + A) *V = MIN(*V, *(INT32*)(DS + B)) // atomic operation

DS_MIN_SRC2_I64

Opcode: 197 (0xc5)
Syntax: DS_MIN_SRC2_I64 ADDR [OFFSET:OFFSET]
Description: Choose smallest signed 64-bit integer value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = A + (((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 INT64* V = (INT64*)(DS + A) *V = MIN(*V, *(INT64*)(DS + B)) // atomic operation

DS_MIN_SRC2_U32

Opcode: 135 (0x87)
Syntax: DS_MIN_SRC2_U32 ADDR [OFFSET:OFFSET]
Description: Choose smallest unsigned integer value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = MIN(*V, *(UINT32*)(DS + B)) // atomic operation

DS_MIN_SRC2_U64

Opcode: 199 (0xc7)
Syntax: DS_MIN_SRC2_U64 ADDR [OFFSET:OFFSET]
Description: Choose smallest unsigned 64-bit integer value from LDS/GDS at address A and at address B, and store result to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = MIN(*V, *(UINT64*)(DS + B)) // atomic operation

DS_MIN_U32

Opcode: 7 (0x7)
Syntax: DS_MIN_U32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Choose smallest unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = MIN(*V, VDATA0) // atomic operation

DS_MIN_U64

Opcode: 71 (0x47)
Syntax: DS_MIN_U64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Choose smallest unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 and VDATA0, and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = MIN(*V, VDATA0) // atomic operation

DS_MSKOR_B32

Opcode: 12 (0xc)
Syntax: DS_MSKOR_U32 ADDR, VDATA0, VDATA1 [OFFSET:OFFSET]
Description: Load first argument from LDS/GDS at address (ADDR+OFFSET) & ~3. Second and third arguments are from VDATA0 and VDATA1. The instruction keeps all bits that is zeroed in second argument, and set all bits from third argument. Result is stored in LDS/GDS at this same address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = (*V & ~VDATA0) | VDATA1 // atomic operation

DS_MSKOR_B64

Opcode: 76 (0x4c)
Syntax: DS_MSKOR_U64 ADDR, VDATA0(2), VDATA1(2) [OFFSET:OFFSET]
Description: Load first argument from LDS/GDS at address (ADDR+OFFSET) & ~7. Second and third arguments are from VDATA0 and VDATA1. All three arguments are 64-bit. The instruction keeps all bits that is zeroed in second argument, and set all bits from third argument. Result is stored in LDS/GDS at this same address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = (*V & ~VDATA0) | VDATA1 // atomic operation

DS_MSKOR_RTN_B32

Opcode: 44 (0x2c)
Syntax: DS_MSKOR_RTN_U32 VDST, ADDR, VDATA0, VDATA1 [OFFSET:OFFSET]
Description: Load first argument from LDS/GDS at address (ADDR+OFFSET) & ~3. Second and third arguments are from VDATA0 and VDATA1. The instruction keeps all bits that is zeroed in second argument, and set all bits from third argument. Result is stored in LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = (*V & ~VDATA0) | VDATA1 // atomic operation

DS_MSKOR_RTN_B64

Opcode: 108 (0x6c)
Syntax: DS_MSKOR_RTN_U64 VDST(2), ADDR, VDATA0(2), VDATA1(2) [OFFSET:OFFSET]
Description: Load first argument from LDS/GDS at address (ADDR+OFFSET) & ~7. Second and third arguments are from VDATA0 and VDATA1. All three arguments are 64-bit. The instruction keeps all bits that is zeroed in second argument, and set all bits from third argument. Result is stored in LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = (*V & ~VDATA0) | VDATA1 // atomic operation

DS_NOP

Opcode: 20 (0x14) for GCN 1.1/1.2
Syntax: DS_NOP VADDR
Description: Do nothing.

DS_OR_B32

Opcode: 10 (0xa)
Syntax: DS_OR_B32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Do bitwise OR operation on 32-bit value from LDS/GDS at address (ADDR+OFFSET) & ~3, and value from VDATA0; and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = *V | VDATA0 // atomic operation

DS_OR_B64

Opcode: 74 (0x4a)
Syntax: DS_OR_B64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Do bitwise OR operation on 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and value from VDATA0; and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = *V | VDATA0 // atomic operation

DS_OR_RTN_B32

Opcode: 42 (0x2a)
Syntax: DS_OR_RTN_B32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Do bitwise OR operation on 32-bit value from LDS/GDS at address (ADDR+OFFSET) & ~3, and value from VDATA0; and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = *V | VDATA0 // atomic operation

DS_OR_RTN_B64

Opcode: 106 (0x6a)
Syntax: DS_OR_RTN_B64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Do bitwise OR operation on 32-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and value from VDATA0; and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = *V | VDATA0 // atomic operation

DS_OR_SRC2_B32

Opcode: 138 (0x8a)
Syntax: DS_OR_SRC2_B32 ADDR [OFFSET:OFFSET]
Description: Do bitwise OR operation on 32-bit value from LDS/GDS at address A, and at address B; and store result to LDS/GDS at address A. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = *V | *(UINT32*)(DS + B) // atomic operation

DS_OR_SRC2_B64

Opcode: 202 (0xca)
Syntax: DS_OR_SRC2_B64 ADDR [OFFSET:OFFSET]
Description: Do bitwise OR operation on 64-bit value from LDS/GDS at address A, and at address B; and store result to LDS/GDS at address A. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = *V | *(UINT64*)(DS + B) // atomic operation

DS_READ_B128

Opcode: 255 (0xff) for GCN 1.1/1.2
Syntax: DS_READ_B128 VDST(4), ADDR [OFFSET:OFFSET]
Description: Read four dwords from LDS/GDS at address (ADDR+OFFSET) & ~15, store into VDST.
Operation:
VDST[0:1] = *(UINT64*)(DS + ((ADDR+OFFSET)&~15)) VDST[2:3] = *(UINT64*)(DS + ((ADDR+OFFSET)&~15) + 8)

DS_READ_B32

Opcode: 54 (0x36)
Syntax: DS_READ_B32 VDST, ADDR [OFFSET:OFFSET]
Description: Read dword from LDS/GDS at address (ADDR+OFFSET) & ~3, store into VDST.
Operation:
VDST = *(UINT32*)(DS + ((ADDR+OFFSET)&~3))

DS_READ_B64

Opcode: 118 (0x76)
Syntax: DS_READ_B64 VDST(2), ADDR [OFFSET:OFFSET]
Description: Read two dwords from LDS/GDS at address (ADDR+OFFSET) & ~7, store into VDST.
Operation:
VDST = *(UINT64*)(DS + ((ADDR+OFFSET)&~7))

DS_READ_B96

Opcode: 254 (0xfe) for GCN 1.1/1.2
Syntax: DS_READ_B96 VDST(3), ADDR [OFFSET:OFFSET]
Description: Read three dwords from LDS/GDS at address (ADDR+OFFSET) & ~15, store into VDST.
Operation:
VDST[0:1] = *(UINT64*)(DS + ((ADDR+OFFSET)&~15)) VDST[2] = *(UINT32*)(DS + ((ADDR+OFFSET)&~15) + 8)

DS_READ_I16

Opcode: 59 (0x3b)
Syntax: DS_READ_I16 VDST, ADDR [OFFSET:OFFSET]
Description: Read signed 16-bit word from LDS/GDS at address (ADDR+OFFSET) & ~1, store into VDST. The value's sign will be extended to higher bits.
Operation:
VDST = (INT32)*(INT16*)(DS + ((ADDR+OFFSET)&~1))

DS_READ_I8

Opcode: 57 (0x39)
Syntax: DS_READ_I8 VDST, ADDR [OFFSET:OFFSET]
Description: Read signed byte from LDS/GDS at address (ADDR+OFFSET), store into VDST. The value's sign will be extended to higher bits.
Operation:
VDST = (INT32)*(INT8*)(DS + (ADDR+OFFSET))

DS_READ_U16

Opcode: 60 (0x3c)
Syntax: DS_READ_U16 VDST, ADDR [OFFSET:OFFSET]
Description: Read unsigned 16-bit word from LDS/GDS at address (ADDR+OFFSET) & ~1, store into VDST.
Operation:
VDST = *(UINT16*)(DS + ((ADDR+OFFSET)&~1))

DS_READ_U8

Opcode: 58 (0x3a)
Syntax: DS_READ_U8 VDST, ADDR [OFFSET:OFFSET]
Description: Read unsigned byte from LDS/GDS at address (ADDR+OFFSET), store into VDST.
Operation:
VDST = *(UINT8*)(DS + (ADDR+OFFSET))

DS_READ2_B32

Opcode: 55 (0x37)
Syntax: DS_READ2_B32 VDST(2), ADDR [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Read dword from LDS/GDS at address (ADDR+OFFSET0*4) & ~3, and second dword at address (ADDR+OFFSET1*4) & ~3, and store these dwords into VDST.
Operation:
UINT32* V0 = (UINT32*)(DS + ((ADDR + OFFSET0*4)&~3)) UINT32* V1 = (UINT32*)(DS + ((ADDR + OFFSET1*4)&~3)) VDST = *V0 | (UINT64(*V1)<<32)

DS_READ2_B64

Opcode: 119 (0x77)
Syntax: DS_READ2_B64 VDST(4), ADDR [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Read 64-bit value from LDS/GDS at address (ADDR+OFFSET0*8) & ~7, and second value at address (ADDR+OFFSET1*8) & ~7, and store these values into VDST.
Operation:
UINT64* V0 = (UINT64*)(DS + (ADDR + OFFSET0*8)&~7) UINT64* V1 = (UINT64*)(DS + (ADDR + OFFSET1*8)&~7) VDST = *V0 | (UINT128(*V1)<<64)

DS_READ2ST64_B32

Opcode: 56 (0x38)
Syntax: DS_READ2ST64_B32 VDST(2), ADDR [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Read dword from LDS/GDS at address (ADDR+OFFSET0*256) & ~3, and second dword at address (ADDR+OFFSET1*256) & ~3, and store these values into VDST.
Operation:
UINT32* V0 = (UINT32*)(DS + ((ADDR + OFFSET0*256)&~3)) UINT32* V1 = (UINT32*)(DS + ((ADDR + OFFSET1*256)&~3)) VDST = *V0 | (UINT64(*V1)<<32)

DS_READ2ST64_B64

Opcode: 120 (0x78)
Syntax: DS_READ2ST64_B64 VDST(4), ADDR [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Read 64-bit value from LDS/GDS at address (ADDR+OFFSET0*512) & ~7, and second value at address (ADDR+OFFSET1*512) & ~7, and store these values into VDST.
Operation:
UINT64* V0 = (UINT64*)(DS + ((ADDR + OFFSET0*512)&~7)) UINT64* V1 = (UINT64*)(DS + ((ADDR + OFFSET1*512)&~7)) VDST = *V0 | (UINT128(*V1)<<64)

DS_RSUB_RTN_U32

Opcode: 34 (0x22)
Syntax: DS_RSUB_RTN_U32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Subtract unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 from value in VDATA0, and store result back to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST= *V; *V = VDATA0 - *V // atomic operation

DS_RSUB_RTN_U64

Opcode: 98 (0x62)
Syntax: DS_RSUB_RTN_U64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Subtract unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 from value in VDATA0, and store result back to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST= *V; *V = VDATA0 - *V // atomic operation

DS_RSUB_SRC2_U32

Opcode: 130 (0x82)
Syntax: DS_RSUB_SRC2_U32 ADDR [OFFSET:OFFSET]
Description: Subtract unsigned integer value from LDS/GDS at address A from value at address B, and store result back to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = *(UINT32*)(DS + B) - *V // atomic operation

DS_RSUB_SRC2_U64

Opcode: 194 (0xc2)
Syntax: DS_RSUB_SRC2_U64 ADDR [OFFSET:OFFSET]
Description: Subtract unsigned integer value from LDS/GDS at address A from value at address B, and store result back to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = *(UINT64*)(DS + B) - *V // atomic operation

DS_RSUB_U32

Opcode: 2 (0x2)
Syntax: DS_RSUB_U32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Subtract unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3 from value in VDATA0, and store result back to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = VDATA0 - *V // atomic operation

DS_RSUB_U64

Opcode: 66 (0x42)
Syntax: DS_RSUB_U64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Subtract unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7 from value in VDATA0, and store result back to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = VDATA0 - *V // atomic operation

DS_SUB_RTN_U32

Opcode: 33 (0x21)
Syntax: DS_SUB_RTN_U32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Subtract VDATA0 from unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3, and store result back to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = *V - VDATA0 // atomic operation

DS_SUB_RTN_U64

Opcode: 97 (0x61)
Syntax: DS_SUB_RTN_U64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Subtract VDATA0 from unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7, and store result back to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = *V - VDATA0 // atomic operation

DS_SUB_SRC2_U32

Opcode: 129 (0x81)
Syntax: DS_SUB_SRC2_U32 ADDR [OFFSET:OFFSET]
Description: Subtract unsigned integer value from LDS/GDS at address B from value at address A, and store result back to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = *V - *(UINT32*)(DS + B) // atomic operation

DS_SUB_SRC2_U64

Opcode: 193 (0xc1)
Syntax: DS_SUB_SRC2_U64 ADDR [OFFSET:OFFSET]
Description: Subtract unsigned 64-bit integer value from LDS/GDS at address B from value at address A, and store result back to LDS/GDS at address A. Refer to listing to learn about addressing. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = *V - *(UINT64*)(DS + B) // atomic operation

DS_SUB_U32

Opcode: 1 (0x1)
Syntax: DS_SUB_U32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Subtract VDATA0 from unsigned integer value from LDS/GDS at address (ADDR+OFFSET) & ~3, and store result back to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = *V - VDATA0 // atomic operation

DS_SUB_U64

Opcode: 65 (0x41)
Syntax: DS_SUB_U64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Subtract VDATA0 from unsigned 64-bit integer value from LDS/GDS at address (ADDR+OFFSET) & ~7, and store result back to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = *V - VDATA0 // atomic operation

DS_SWIZZLE_B32

Opcode: 53 (0x35) for GCN 1.0/1.1; 61 (0x3d) for GCN 1.2
Syntax: DS_SWIZZLE_B32 VDST, ADDR [OFFSET:OFFSET]
Description: Move between lanes ADDR value storing in VDST. Refer to operation's listing to learn about how this instruction calcutes input lane. If input lane is not valid then store zero to VDST[LANEID]. This instruction doesn't finish immediately and it increments LGKMCNT counter (you must use S_WAITCNT LGKMCNT(X) to wait for its results).
Operation:
BYTE INLANEID = 0 if (OFFSET&0x8000) INLANEID = (LANEID&0x3c) + ((OFFSET >> (2*(LANEID&3)) & 3) else { BYTE ANDMASK = (OFFSET&31) BYTE ORMASK = (OFFSET>>5)&31 BYTE XORMASK = OFFSET>>10 INLANEID = (LANEID&32) + ((((LANEID&31) & ANDMASK) | ORMASK) ^ XORMASK) } VDST[LANEID] = (EXEC & (1ULL<<INLANEID)) ? ADDR[INLANEID] : 0

DS_WRAP_RTN_B32

Opcode: 52 (0x34) for GCN 1.1/1.2
Syntax: DS_WRAP_RTN_B32 VDST, ADDR, VDATA0, VDATA1 [OFFSET:OFFSET]
Description: Compare unsigned 32-bit value from LDS/GDS at address A and D0. If value from LDS/GDS is not less than VDATA0 value, then subtract VDATA0 from this value and store result into LDS/GDS. Otherwise add VDATA1 to LDS/GDS value and store into LDS/GDS. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32 A = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = (*V >= VDATA0) ? (*V-VDATA0) : (*V+VDATA1) // atomic operation

DS_WRITE_B128

Opcode: 223 (0xdf)for GCN 1.1/1.2
Syntax: DS_WRITE_B128 ADDR, VDATA0(4) [OFFSET:OFFSET]
Description: Store four dwords value from VDATA0 into LDS/GDS at address (ADDR+OFFSET) & ~15.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~15)) *V = VDATA0[0:1] *(V+1) = VDATA0[2:3]

DS_WRITE_B16

Opcode: 31 (0x1f)
Syntax: DS_WRITE_B16 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Store low 16 bits from VDATA0 into LDS/GDS at address (ADDR+OFFSET) & 1.
Operation:
UINT16* V = (UINT16*)(DS + (ADDR+OFFSET)&~1) *V = VDATA0&0xffff

DS_WRITE_B32

Opcode: 13 (0xd)
Syntax: DS_WRITE_B32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Store value from VDATA0 into LDS/GDS at address (ADDR+OFFSET) & ~3.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = VDATA0

DS_WRITE_B64

Opcode: 77 (0x4d)
Syntax: DS_WRITE_B64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Store 64-bit value from VDATA0 into LDS/GDS at address (ADDR+OFFSET) & ~7.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = VDATA0

DS_WRITE_B8

Opcode: 30 (0x1e)
Syntax: DS_WRITE_B8 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Store low 8 bits from VDATA0 into LDS/GDS at address (ADDR+OFFSET).
Operation:
UINT8* V = (UINT8*)(DS + (ADDR+OFFSET)) *V = VDATA0&0xff

DS_WRITE_B96

Opcode: 222 (0xde) for GCN 1.1/1.2
Syntax: DS_WRITE_B96 ADDR, VDATA0(3) [OFFSET:OFFSET]
Description: Store three dwords value from VDATA0 into LDS/GDS at address (ADDR+OFFSET) & ~15.
Operation:
UINT32* V = (UINT64*)(DS + ((ADDR+OFFSET)&~15)) *(UINT64*)V = VDATA0[0:1] *(V+2) = VDATA0[2]

DS_WRITE_SRC2_B32

Opcode: 141 (0x8d)
Syntax: DS_WRITE_SRC2_B32 ADDR [OFFSET:OFFSET]
Description: Store value from LDS/GDS at address B into LDS/GDS at address A.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 *(UINT32*)(DS + A) = *(UINT32*)(DS + B)

DS_WRITE_SRC2_B64

Opcode: 205 (0xcd)
Syntax: DS_WRITE_SRC2_B64 ADDR [OFFSET:OFFSET]
Description: Store 64-bit value from LDS/GDS at address B into LDS/GDS at address A.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 *(UINT64*)(DS + A) = *(UINT64*)(DS + B)

DS_WRITE2_B32

Opcode: 14 (0xe)
Syntax: DS_WRITE2_B32 ADDR, VDATA0, VDATA1 [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Store one dword from VDATA0 into LDS/GDS at address (ADDR+OFFSET0*4) & ~3, and second into LDS/GDS at address (ADDR+OFFSET1*4) & ~3.
Operation:
UINT32* V0 = (UINT32*)(DS + ((ADDR + OFFSET0*4)&~3)) UINT32* V1 = (UINT32*)(DS + ((ADDR + OFFSET1*4)&~3)) *V0 = VDATA0 *V1 = VDATA1

DS_WRITE2_B64

Opcode: 78 (0x4e)
Syntax: DS_WRITE2_B64 ADDR, VDATA0(2), VDATA1(2) [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Store one 64-bit value from VDATA0 into LDS/GDS at address (ADDR+OFFSET0*8) & ~7, and second into LDS/GDS at address (ADDR+OFFSET1*8) & ~7.
Operation:
UINT64* V0 = (UINT64*)(DS + (ADDR + OFFSET0*8)&~7) UINT64* V1 = (UINT64*)(DS + (ADDR + OFFSET1*8)&~7) *V0 = VDATA0 *V1 = VDATA1

DS_WRITE2ST64_B32

Opcode: 15 (0xf)
Syntax: DS_WRITE2ST64_B32 ADDR, VDATA0, VDATA1 [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Store one dword from VDATA0 into LDS/GDS at address (ADDR+OFFSET0*256) & ~3, and second into LDS/GDS at address (ADDR+OFFSET1*256) & ~3.
Operation:
UINT32* V0 = (UINT32*)(DS + ((ADDR + OFFSET0*256)&~3)) UINT32* V1 = (UINT32*)(DS + ((ADDR + OFFSET1*256)&~3)) *V0 = VDATA0 *V1 = VDATA1

DS_WRITE2ST64_B64

Opcode: 79 (0x4f)
Syntax: DS_WRITE2ST64_B64 ADDR, VDATA0(2), VDATA1(2) [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Store one 64-bit dword from VDATA0 into LDS/GDS at address (ADDR+OFFSET0*512) & ~7, and second into LDS/GDS at address (ADDR+OFFSET1*512) & ~7.
Operation:
UINT64* V0 = (UINT64*)(DS + ((ADDR + OFFSET0*512)&~7)) UINT64* V1 = (UINT64*)(DS + ((ADDR + OFFSET1*512)&~7)) *V0 = VDATA0 *V1 = VDATA1

DS_WRXCHG_RTN_B32

Opcode: 45 (0x2d)
Syntax: DS_WRXCHG_B32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Store value from VDATA0 into LDS/GDS at address (ADDR+OFFSET) & ~3. Previous value from LDS/GDS are stored in VDST.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = VDATA0 // atomic operation

DS_WRXCHG_RTN_B64

Opcode: 109 (0x6d)
Syntax: DS_WRXCHG_B64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Store 64-bit value from VDATA0 into LDS/GDS at address (ADDR+OFFSET) & ~7. Previous value from LDS/GDS are stored in VDST.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = VDATA0 // atomic operation

DS_WRXCHG2_RTN_B32

Opcode: 46 (0x2e)
Syntax: DS_WRXCHG2_RTN_B32 VDST(2), ADDR, VDATA0, VDATA1 [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Store one dword from VDATA0 into LDS/GDS at address (ADDR+OFFSET0*4) & ~3, and second into LDS/GDS at address (ADDR+OFFSET1*4) & ~3. Previous values from LDS/GDS are stored in VDST.
Operation:
UINT32* V0 = (UINT32*)(DS + ((ADDR + OFFSET0*4)&~3)) UINT32* V1 = (UINT32*)(DS + ((ADDR + OFFSET1*4)&~3)) VDST = (*V0) | (UINT64(*V1)<<32) *V0 = VDATA0 *V1 = VDATA1

DS_WRXCHG2_RTN_B64

Opcode: 110 (0x6e)
Syntax: DS_WRXCHG2_RTN_B64 VDST(4), ADDR, VDATA0(2), VDATA1(2) [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Store one 64-bit value from VDATA0 into LDS/GDS at address (ADDR+OFFSET0*8) & ~7, and second into LDS/GDS at address (ADDR+OFFSET1*8) & ~7. Previous values from LDS/GDS are stored in VDST.
Operation:
UINT64* V0 = (UINT64*)(DS + ((ADDR + OFFSET0*8)&~7)) UINT64* V1 = (UINT64*)(DS + ((ADDR + OFFSET1*8)&~7)) VDST = (*V0) | (UINT128(*V1)<<64) *V0 = VDATA0 *V1 = VDATA1

DS_WRXCHG2ST64_RTN_B32

Opcode: 47 (0x2f)
Syntax: DS_WRXCHG2ST64_RTN_B32 VDST(2), ADDR, VDATA0, VDATA1 [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Store one dword from VDATA0 into LDS/GDS at address (ADDR+OFFSET0*256) & ~3, and second into LDS/GDS at address (ADDR+OFFSET1*256) & ~3. Previous values from LDS/GDS are stored in VDST.
Operation:
UINT32* V0 = (UINT32*)(DS + ((ADDR + OFFSET0*256)&~3)) UINT32* V1 = (UINT32*)(DS + ((ADDR + OFFSET1*256)&~3)) VDST = (*V0) | (UINT64(*V1)<<32) *V0 = VDATA0 *V1 = VDATA1

DS_WRXCHG2ST64_RTN_B64

Opcode: 111 (0x6f)
Syntax: DS_WRXCHG2ST64_RTN_B64 VDST(4), ADDR, VDATA0(2), VDATA1(2) [OFFSET0:OFFSET0] [OFFSET1:OFFSET1]
Description: Store one 64-bit value from VDATA0 into LDS/GDS at address (ADDR+OFFSET0*512) & ~7, and second into LDS/GDS at address (ADDR+OFFSET1*512) & ~7. Previous values from LDS/GDS are stored in VDST.
Operation:
UINT64* V0 = (UINT64*)(DS + ((ADDR + OFFSET0*512)&~7)) UINT64* V1 = (UINT64*)(DS + ((ADDR + OFFSET1*512)&~7)) VDST = (*V0) | (UINT128(*V1)<<64) *V0 = VDATA0 *V1 = VDATA1

DS_XOR_B32

Opcode: 11 (0xb)
Syntax: DS_XOR_B32 ADDR, VDATA0 [OFFSET:OFFSET]
Description: Do bitwise XOR operation on 32-bit value from LDS/GDS at address (ADDR+OFFSET) & ~3, and value from VDATA0; and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) *V = *V ^ VDATA0 // atomic operation

DS_XOR_B64

Opcode: 75 (0x4b)
Syntax: DS_XOR_B64 ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Do bitwise XOR operation on 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and value from VDATA0; and store result to LDS/GDS at this same address. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) *V = *V ^ VDATA0 // atomic operation

DS_XOR_RTN_B32

Opcode: 43 (0x2b)
Syntax: DS_XOR_RTN_B32 VDST, ADDR, VDATA0 [OFFSET:OFFSET]
Description: Do bitwise XOR operation on 32-bit value from LDS/GDS at address (ADDR+OFFSET) & ~3, and value from VDATA0; and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT32* V = (UINT32*)(DS + ((ADDR+OFFSET)&~3)) VDST = *V; *V = *V ^ VDATA0 // atomic operation

DS_XOR_RTN_B64

Opcode: 107 (0x6b)
Syntax: DS_XOR_RTN_B64 VDST(2), ADDR, VDATA0(2) [OFFSET:OFFSET]
Description: Do bitwise XOR operation on 64-bit value from LDS/GDS at address (ADDR+OFFSET) & ~7, and value from VDATA0; and store result to LDS/GDS at this same address. Previous value from LDS/GDS are stored in VDST. Operation is atomic.
Operation:
UINT64* V = (UINT64*)(DS + ((ADDR+OFFSET)&~7)) VDST = *V; *V = *V ^ VDATA0 // atomic operation

DS_XOR_SRC2_B32

Opcode: 139 (0x8b)
Syntax: DS_XOR_SRC2_B32 ADDR [OFFSET:OFFSET]
Description: Do bitwise XOR operation on 32-bit value from LDS/GDS at address A, and at address B; and store result to LDS/GDS at address A. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fffc : ADDR&~3 UINT32 B = A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4 UINT32* V = (UINT32*)(DS + A) *V = *V ^ *(UINT32*)(DS + B) // atomic operation

DS_XOR_SRC2_B64

Opcode: 203 (0xcb)
Syntax: DS_XOR_SRC2_B64 ADDR [OFFSET:OFFSET]
Description: Do bitwise XOR operation on 64-bit value from LDS/GDS at address A, and at address B; and store result to LDS/GDS at address A. Operation is atomic.
Operation:
UINT32 A = (OFFSET&0x8000) ? ADDR&0x1fff8 : ADDR&~7 UINT32 B = (A + ((OFFSET&0x8000) ? \ ((ADDR>>17) | ((ADDR>>16)&0x8000)) : \ ((OFFSET&0x7fff) | (OFFSET<<1)&0x8000)) * 4)&~7 UINT64* V = (UINT64*)(DS + A) *V = *V ^ *(UINT64*)(DS + B) // atomic operation