Add support for 256-bit-wide vectors.

parent 9d47e583
......@@ -47,6 +47,7 @@ data ArgRep = P -- GC Ptr
| F -- Float
| D -- Double
| V16 -- 16-byte (128-bit) vectors of Float/Double/Int8/Word32/etc.
| V32 -- 32-byte (256-bit) vectors of Float/Double/Int8/Word32/etc.
instance Outputable ArgRep where ppr = text . argRepString
argRepString :: ArgRep -> String
......@@ -57,6 +58,7 @@ argRepString V = "V"
argRepString F = "F"
argRepString D = "D"
argRepString V16 = "V16"
argRepString V32 = "V32"
toArgRep :: PrimRep -> ArgRep
toArgRep VoidRep = V
......@@ -68,9 +70,10 @@ toArgRep Int64Rep = L
toArgRep Word64Rep = L
toArgRep FloatRep = F
toArgRep DoubleRep = D
toArgRep (VecRep len elem)
| len*primElemRepSizeB elem == 16 = V16
| otherwise = error "toArgRep: bad vector primrep"
toArgRep (VecRep len elem) = case len*primElemRepSizeB elem of
16 -> V16
32 -> V32
_ -> error "toArgRep: bad vector primrep"
isNonV :: ArgRep -> Bool
isNonV V = False
......@@ -84,6 +87,7 @@ argRepSizeW dflags L = wORD64_SIZE `quot` wORD_SIZE dflags
argRepSizeW dflags D = dOUBLE_SIZE dflags `quot` wORD_SIZE dflags
argRepSizeW _ V = 0
argRepSizeW dflags V16 = 16 `quot` wORD_SIZE dflags
argRepSizeW dflags V32 = 32 `quot` wORD_SIZE dflags
idArgRep :: Id -> ArgRep
idArgRep = toArgRep . idPrimRep
......@@ -132,4 +136,5 @@ slowCallPattern (F: _) = (fsLit "stg_ap_f", 1)
slowCallPattern (D: _) = (fsLit "stg_ap_d", 1)
slowCallPattern (L: _) = (fsLit "stg_ap_l", 1)
slowCallPattern (V16: _) = (fsLit "stg_ap_v16", 1)
slowCallPattern (V32: _) = (fsLit "stg_ap_v32", 1)
slowCallPattern [] = (fsLit "stg_ap_0", 0)
......@@ -385,6 +385,7 @@ stdPattern reps
[D] -> Just ARG_D
[L] -> Just ARG_L
[V16] -> Just ARG_V16
[V32] -> Just ARG_V32
[N,N] -> Just ARG_NN
[N,P] -> Just ARG_NP
......
......@@ -462,6 +462,7 @@ push_alts L = bci_PUSH_ALTS_L
push_alts F = bci_PUSH_ALTS_F
push_alts D = bci_PUSH_ALTS_D
push_alts V16 = error "push_alts: vector"
push_alts V32 = error "push_alts: vector"
return_ubx :: ArgRep -> Word16
return_ubx V = bci_RETURN_V
......@@ -471,6 +472,7 @@ return_ubx L = bci_RETURN_L
return_ubx F = bci_RETURN_F
return_ubx D = bci_RETURN_D
return_ubx V16 = error "return_ubx: vector"
return_ubx V32 = error "return_ubx: vector"
-- Make lists of host-sized words for literals, so that when the
-- words are placed in memory at increasing addresses, the
......
......@@ -99,6 +99,7 @@
#define D_ float64
#define L_ bits64
#define V16_ bits128
#define V32_ bits256
#define SIZEOF_StgDouble 8
#define SIZEOF_StgWord64 8
......
......@@ -34,22 +34,23 @@
#define ARG_D 7
#define ARG_L 8
#define ARG_V16 9
#define ARG_NN 10
#define ARG_NP 11
#define ARG_PN 12
#define ARG_PP 13
#define ARG_NNN 14
#define ARG_NNP 15
#define ARG_NPN 16
#define ARG_NPP 17
#define ARG_PNN 18
#define ARG_PNP 19
#define ARG_PPN 20
#define ARG_PPP 21
#define ARG_PPPP 22
#define ARG_PPPPP 23
#define ARG_PPPPPP 24
#define ARG_PPPPPPP 25
#define ARG_PPPPPPPP 26
#define ARG_V32 10
#define ARG_NN 11
#define ARG_NP 12
#define ARG_PN 13
#define ARG_PP 14
#define ARG_NNN 15
#define ARG_NNP 16
#define ARG_NPN 17
#define ARG_NPP 18
#define ARG_PNN 19
#define ARG_PNP 20
#define ARG_PPN 21
#define ARG_PPP 22
#define ARG_PPPP 23
#define ARG_PPPPP 24
#define ARG_PPPPPP 25
#define ARG_PPPPPPP 26
#define ARG_PPPPPPPP 27
#endif /* RTS_STORAGE_FUNTYPES_H */
......@@ -225,6 +225,7 @@ RTS_RET(stg_ap_f);
RTS_RET(stg_ap_d);
RTS_RET(stg_ap_l);
RTS_RET(stg_ap_v16);
RTS_RET(stg_ap_v32);
RTS_RET(stg_ap_n);
RTS_RET(stg_ap_p);
RTS_RET(stg_ap_pv);
......@@ -242,6 +243,7 @@ RTS_FUN_DECL(stg_ap_f_fast);
RTS_FUN_DECL(stg_ap_d_fast);
RTS_FUN_DECL(stg_ap_l_fast);
RTS_FUN_DECL(stg_ap_v16_fast);
RTS_FUN_DECL(stg_ap_v32_fast);
RTS_FUN_DECL(stg_ap_n_fast);
RTS_FUN_DECL(stg_ap_p_fast);
RTS_FUN_DECL(stg_ap_pv_fast);
......
......@@ -883,6 +883,7 @@ typedef struct _RtsSymbolVal {
SymI_HasProto(stg_ap_d_ret) \
SymI_HasProto(stg_ap_l_ret) \
SymI_HasProto(stg_ap_v16_ret) \
SymI_HasProto(stg_ap_v32_ret) \
SymI_HasProto(stg_ap_n_ret) \
SymI_HasProto(stg_ap_p_ret) \
SymI_HasProto(stg_ap_pv_ret) \
......@@ -1252,6 +1253,7 @@ typedef struct _RtsSymbolVal {
SymI_HasProto(stg_ap_d_info) \
SymI_HasProto(stg_ap_l_info) \
SymI_HasProto(stg_ap_v16_info) \
SymI_HasProto(stg_ap_v32_info) \
SymI_HasProto(stg_ap_n_info) \
SymI_HasProto(stg_ap_p_info) \
SymI_HasProto(stg_ap_pv_info) \
......@@ -1268,6 +1270,7 @@ typedef struct _RtsSymbolVal {
SymI_HasProto(stg_ap_d_fast) \
SymI_HasProto(stg_ap_l_fast) \
SymI_HasProto(stg_ap_v16_fast) \
SymI_HasProto(stg_ap_v32_fast) \
SymI_HasProto(stg_ap_n_fast) \
SymI_HasProto(stg_ap_p_fast) \
SymI_HasProto(stg_ap_pv_fast) \
......
......@@ -33,6 +33,7 @@ data ArgRep
| D -- double
| L -- long (64-bit)
| V16 -- 16-byte (128-bit) vectors
| V32 -- 32-byte (256-bit) vectors
-- size of a value in *words*
argSize :: ArgRep -> Int
......@@ -43,6 +44,7 @@ argSize F = 1
argSize D = (SIZEOF_DOUBLE `quot` SIZEOF_VOID_P :: Int)
argSize L = (8 `quot` SIZEOF_VOID_P :: Int)
argSize V16 = (16 `quot` SIZEOF_VOID_P :: Int)
argSize V32 = (32 `quot` SIZEOF_VOID_P :: Int)
showArg :: ArgRep -> String
showArg N = "n"
......@@ -52,6 +54,7 @@ showArg F = "f"
showArg D = "d"
showArg L = "l"
showArg V16 = "v16"
showArg V32 = "v32"
-- is a value a pointer?
isPtr :: ArgRep -> Bool
......@@ -504,6 +507,7 @@ argRep D = text "D_"
argRep L = text "L_"
argRep P = text "gcptr"
argRep V16 = text "V16_"
argRep V32 = text "V32_"
argRep _ = text "W_"
genApply regstatus args =
......@@ -854,6 +858,7 @@ applyTypes = [
[D],
[L],
[V16],
[V32],
[N],
[P],
[P,V],
......@@ -882,6 +887,7 @@ stackApplyTypes = [
[D],
[L],
[V16],
[V32],
[N,N],
[N,P],
[P,N],
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment