diff options
author | Geoffrey Mainland <gmainlan@microsoft.com> | 2012-10-19 09:06:17 +0100 |
---|---|---|
committer | Geoffrey Mainland <gmainlan@microsoft.com> | 2013-02-01 22:00:24 +0000 |
commit | 4af62075bbe9e96a3678fc90288496e0c4c7c17d (patch) | |
tree | 3df4fa03089310cd66678681a4ce78dd39bea25f /compiler/codeGen | |
parent | 6480a35c15717025c169980b1cc763a7e6f36056 (diff) | |
download | haskell-4af62075bbe9e96a3678fc90288496e0c4c7c17d.tar.gz |
Add the Float32X4# primitive type and associated primops.
This patch lays the groundwork needed for primop support for SIMD vectors. In
addition to the groundwork, we add support for the FloatX4# primitive type and
associated primops.
* Add the FloatX4# primitive type and associated primops.
* Add CodeGen support for Float vectors.
* Compile vector operations to LLVM vector operations in the LLVM code
generator.
* Make the x86 native backend fail gracefully when encountering vector primops.
* Only generate primop wrappers for vector primops when using LLVM.
Diffstat (limited to 'compiler/codeGen')
-rw-r--r-- | compiler/codeGen/StgCmmPrim.hs | 474 |
1 files changed, 337 insertions, 137 deletions
diff --git a/compiler/codeGen/StgCmmPrim.hs b/compiler/codeGen/StgCmmPrim.hs index 986286647b..9a583b8354 100644 --- a/compiler/codeGen/StgCmmPrim.hs +++ b/compiler/codeGen/StgCmmPrim.hs @@ -365,117 +365,129 @@ emitPrimOp dflags [res] SizeofMutableArrayArrayOp [arg] -- IndexXXXoffAddr -emitPrimOp dflags res IndexOffAddrOp_Char args = doIndexOffAddrOp (Just (mo_u_8ToWord dflags)) b8 res args -emitPrimOp dflags res IndexOffAddrOp_WideChar args = doIndexOffAddrOp (Just (mo_u_32ToWord dflags)) b32 res args -emitPrimOp dflags res IndexOffAddrOp_Int args = doIndexOffAddrOp Nothing (bWord dflags) res args -emitPrimOp dflags res IndexOffAddrOp_Word args = doIndexOffAddrOp Nothing (bWord dflags) res args -emitPrimOp dflags res IndexOffAddrOp_Addr args = doIndexOffAddrOp Nothing (bWord dflags) res args -emitPrimOp _ res IndexOffAddrOp_Float args = doIndexOffAddrOp Nothing f32 res args -emitPrimOp _ res IndexOffAddrOp_Double args = doIndexOffAddrOp Nothing f64 res args -emitPrimOp dflags res IndexOffAddrOp_StablePtr args = doIndexOffAddrOp Nothing (bWord dflags) res args -emitPrimOp dflags res IndexOffAddrOp_Int8 args = doIndexOffAddrOp (Just (mo_s_8ToWord dflags)) b8 res args -emitPrimOp dflags res IndexOffAddrOp_Int16 args = doIndexOffAddrOp (Just (mo_s_16ToWord dflags)) b16 res args -emitPrimOp dflags res IndexOffAddrOp_Int32 args = doIndexOffAddrOp (Just (mo_s_32ToWord dflags)) b32 res args -emitPrimOp _ res IndexOffAddrOp_Int64 args = doIndexOffAddrOp Nothing b64 res args -emitPrimOp dflags res IndexOffAddrOp_Word8 args = doIndexOffAddrOp (Just (mo_u_8ToWord dflags)) b8 res args -emitPrimOp dflags res IndexOffAddrOp_Word16 args = doIndexOffAddrOp (Just (mo_u_16ToWord dflags)) b16 res args -emitPrimOp dflags res IndexOffAddrOp_Word32 args = doIndexOffAddrOp (Just (mo_u_32ToWord dflags)) b32 res args -emitPrimOp _ res IndexOffAddrOp_Word64 args = doIndexOffAddrOp Nothing b64 res args +emitPrimOp dflags res IndexOffAddrOp_Char args = doIndexOffAddrOp (Just (mo_u_8ToWord dflags)) b8 res args +emitPrimOp dflags res IndexOffAddrOp_WideChar args = doIndexOffAddrOp (Just (mo_u_32ToWord dflags)) b32 res args +emitPrimOp dflags res IndexOffAddrOp_Int args = doIndexOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res IndexOffAddrOp_Word args = doIndexOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res IndexOffAddrOp_Addr args = doIndexOffAddrOp Nothing (bWord dflags) res args +emitPrimOp _ res IndexOffAddrOp_Float args = doIndexOffAddrOp Nothing f32 res args +emitPrimOp _ res IndexOffAddrOp_Double args = doIndexOffAddrOp Nothing f64 res args +emitPrimOp dflags res IndexOffAddrOp_StablePtr args = doIndexOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res IndexOffAddrOp_Int8 args = doIndexOffAddrOp (Just (mo_s_8ToWord dflags)) b8 res args +emitPrimOp dflags res IndexOffAddrOp_Int16 args = doIndexOffAddrOp (Just (mo_s_16ToWord dflags)) b16 res args +emitPrimOp dflags res IndexOffAddrOp_Int32 args = doIndexOffAddrOp (Just (mo_s_32ToWord dflags)) b32 res args +emitPrimOp _ res IndexOffAddrOp_Int64 args = doIndexOffAddrOp Nothing b64 res args +emitPrimOp dflags res IndexOffAddrOp_Word8 args = doIndexOffAddrOp (Just (mo_u_8ToWord dflags)) b8 res args +emitPrimOp dflags res IndexOffAddrOp_Word16 args = doIndexOffAddrOp (Just (mo_u_16ToWord dflags)) b16 res args +emitPrimOp dflags res IndexOffAddrOp_Word32 args = doIndexOffAddrOp (Just (mo_u_32ToWord dflags)) b32 res args +emitPrimOp _ res IndexOffAddrOp_Word64 args = doIndexOffAddrOp Nothing b64 res args +emitPrimOp _ res IndexOffAddrOp_FloatX4 args = doIndexOffAddrOp Nothing vec4f32 res args +emitPrimOp _ res IndexOffAddrOp_FloatAsFloatX4 args = doIndexOffAddrOpAs Nothing vec4f32 f32 res args -- ReadXXXoffAddr, which are identical, for our purposes, to IndexXXXoffAddr. -emitPrimOp dflags res ReadOffAddrOp_Char args = doIndexOffAddrOp (Just (mo_u_8ToWord dflags)) b8 res args -emitPrimOp dflags res ReadOffAddrOp_WideChar args = doIndexOffAddrOp (Just (mo_u_32ToWord dflags)) b32 res args -emitPrimOp dflags res ReadOffAddrOp_Int args = doIndexOffAddrOp Nothing (bWord dflags) res args -emitPrimOp dflags res ReadOffAddrOp_Word args = doIndexOffAddrOp Nothing (bWord dflags) res args -emitPrimOp dflags res ReadOffAddrOp_Addr args = doIndexOffAddrOp Nothing (bWord dflags) res args -emitPrimOp _ res ReadOffAddrOp_Float args = doIndexOffAddrOp Nothing f32 res args -emitPrimOp _ res ReadOffAddrOp_Double args = doIndexOffAddrOp Nothing f64 res args -emitPrimOp dflags res ReadOffAddrOp_StablePtr args = doIndexOffAddrOp Nothing (bWord dflags) res args -emitPrimOp dflags res ReadOffAddrOp_Int8 args = doIndexOffAddrOp (Just (mo_s_8ToWord dflags)) b8 res args -emitPrimOp dflags res ReadOffAddrOp_Int16 args = doIndexOffAddrOp (Just (mo_s_16ToWord dflags)) b16 res args -emitPrimOp dflags res ReadOffAddrOp_Int32 args = doIndexOffAddrOp (Just (mo_s_32ToWord dflags)) b32 res args -emitPrimOp _ res ReadOffAddrOp_Int64 args = doIndexOffAddrOp Nothing b64 res args -emitPrimOp dflags res ReadOffAddrOp_Word8 args = doIndexOffAddrOp (Just (mo_u_8ToWord dflags)) b8 res args -emitPrimOp dflags res ReadOffAddrOp_Word16 args = doIndexOffAddrOp (Just (mo_u_16ToWord dflags)) b16 res args -emitPrimOp dflags res ReadOffAddrOp_Word32 args = doIndexOffAddrOp (Just (mo_u_32ToWord dflags)) b32 res args -emitPrimOp _ res ReadOffAddrOp_Word64 args = doIndexOffAddrOp Nothing b64 res args +emitPrimOp dflags res ReadOffAddrOp_Char args = doIndexOffAddrOp (Just (mo_u_8ToWord dflags)) b8 res args +emitPrimOp dflags res ReadOffAddrOp_WideChar args = doIndexOffAddrOp (Just (mo_u_32ToWord dflags)) b32 res args +emitPrimOp dflags res ReadOffAddrOp_Int args = doIndexOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res ReadOffAddrOp_Word args = doIndexOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res ReadOffAddrOp_Addr args = doIndexOffAddrOp Nothing (bWord dflags) res args +emitPrimOp _ res ReadOffAddrOp_Float args = doIndexOffAddrOp Nothing f32 res args +emitPrimOp _ res ReadOffAddrOp_Double args = doIndexOffAddrOp Nothing f64 res args +emitPrimOp dflags res ReadOffAddrOp_StablePtr args = doIndexOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res ReadOffAddrOp_Int8 args = doIndexOffAddrOp (Just (mo_s_8ToWord dflags)) b8 res args +emitPrimOp dflags res ReadOffAddrOp_Int16 args = doIndexOffAddrOp (Just (mo_s_16ToWord dflags)) b16 res args +emitPrimOp dflags res ReadOffAddrOp_Int32 args = doIndexOffAddrOp (Just (mo_s_32ToWord dflags)) b32 res args +emitPrimOp _ res ReadOffAddrOp_Int64 args = doIndexOffAddrOp Nothing b64 res args +emitPrimOp dflags res ReadOffAddrOp_Word8 args = doIndexOffAddrOp (Just (mo_u_8ToWord dflags)) b8 res args +emitPrimOp dflags res ReadOffAddrOp_Word16 args = doIndexOffAddrOp (Just (mo_u_16ToWord dflags)) b16 res args +emitPrimOp dflags res ReadOffAddrOp_Word32 args = doIndexOffAddrOp (Just (mo_u_32ToWord dflags)) b32 res args +emitPrimOp _ res ReadOffAddrOp_Word64 args = doIndexOffAddrOp Nothing b64 res args +emitPrimOp _ res ReadOffAddrOp_FloatX4 args = doIndexOffAddrOp Nothing vec4f32 res args +emitPrimOp _ res ReadOffAddrOp_FloatAsFloatX4 args = doIndexOffAddrOpAs Nothing vec4f32 b32 res args -- IndexXXXArray -emitPrimOp dflags res IndexByteArrayOp_Char args = doIndexByteArrayOp (Just (mo_u_8ToWord dflags)) b8 res args -emitPrimOp dflags res IndexByteArrayOp_WideChar args = doIndexByteArrayOp (Just (mo_u_32ToWord dflags)) b32 res args -emitPrimOp dflags res IndexByteArrayOp_Int args = doIndexByteArrayOp Nothing (bWord dflags) res args -emitPrimOp dflags res IndexByteArrayOp_Word args = doIndexByteArrayOp Nothing (bWord dflags) res args -emitPrimOp dflags res IndexByteArrayOp_Addr args = doIndexByteArrayOp Nothing (bWord dflags) res args -emitPrimOp _ res IndexByteArrayOp_Float args = doIndexByteArrayOp Nothing f32 res args -emitPrimOp _ res IndexByteArrayOp_Double args = doIndexByteArrayOp Nothing f64 res args -emitPrimOp dflags res IndexByteArrayOp_StablePtr args = doIndexByteArrayOp Nothing (bWord dflags) res args -emitPrimOp dflags res IndexByteArrayOp_Int8 args = doIndexByteArrayOp (Just (mo_s_8ToWord dflags)) b8 res args -emitPrimOp dflags res IndexByteArrayOp_Int16 args = doIndexByteArrayOp (Just (mo_s_16ToWord dflags)) b16 res args -emitPrimOp dflags res IndexByteArrayOp_Int32 args = doIndexByteArrayOp (Just (mo_s_32ToWord dflags)) b32 res args -emitPrimOp _ res IndexByteArrayOp_Int64 args = doIndexByteArrayOp Nothing b64 res args -emitPrimOp dflags res IndexByteArrayOp_Word8 args = doIndexByteArrayOp (Just (mo_u_8ToWord dflags)) b8 res args -emitPrimOp dflags res IndexByteArrayOp_Word16 args = doIndexByteArrayOp (Just (mo_u_16ToWord dflags)) b16 res args -emitPrimOp dflags res IndexByteArrayOp_Word32 args = doIndexByteArrayOp (Just (mo_u_32ToWord dflags)) b32 res args -emitPrimOp _ res IndexByteArrayOp_Word64 args = doIndexByteArrayOp Nothing b64 res args +emitPrimOp dflags res IndexByteArrayOp_Char args = doIndexByteArrayOp (Just (mo_u_8ToWord dflags)) b8 res args +emitPrimOp dflags res IndexByteArrayOp_WideChar args = doIndexByteArrayOp (Just (mo_u_32ToWord dflags)) b32 res args +emitPrimOp dflags res IndexByteArrayOp_Int args = doIndexByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res IndexByteArrayOp_Word args = doIndexByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res IndexByteArrayOp_Addr args = doIndexByteArrayOp Nothing (bWord dflags) res args +emitPrimOp _ res IndexByteArrayOp_Float args = doIndexByteArrayOp Nothing f32 res args +emitPrimOp _ res IndexByteArrayOp_Double args = doIndexByteArrayOp Nothing f64 res args +emitPrimOp dflags res IndexByteArrayOp_StablePtr args = doIndexByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res IndexByteArrayOp_Int8 args = doIndexByteArrayOp (Just (mo_s_8ToWord dflags)) b8 res args +emitPrimOp dflags res IndexByteArrayOp_Int16 args = doIndexByteArrayOp (Just (mo_s_16ToWord dflags)) b16 res args +emitPrimOp dflags res IndexByteArrayOp_Int32 args = doIndexByteArrayOp (Just (mo_s_32ToWord dflags)) b32 res args +emitPrimOp _ res IndexByteArrayOp_Int64 args = doIndexByteArrayOp Nothing b64 res args +emitPrimOp dflags res IndexByteArrayOp_Word8 args = doIndexByteArrayOp (Just (mo_u_8ToWord dflags)) b8 res args +emitPrimOp dflags res IndexByteArrayOp_Word16 args = doIndexByteArrayOp (Just (mo_u_16ToWord dflags)) b16 res args +emitPrimOp dflags res IndexByteArrayOp_Word32 args = doIndexByteArrayOp (Just (mo_u_32ToWord dflags)) b32 res args +emitPrimOp _ res IndexByteArrayOp_Word64 args = doIndexByteArrayOp Nothing b64 res args +emitPrimOp _ res IndexByteArrayOp_FloatX4 args = doIndexByteArrayOp Nothing vec4f32 res args +emitPrimOp _ res IndexByteArrayOp_FloatAsFloatX4 args = doIndexByteArrayOpAs Nothing vec4f32 f32 res args -- ReadXXXArray, identical to IndexXXXArray. -emitPrimOp dflags res ReadByteArrayOp_Char args = doIndexByteArrayOp (Just (mo_u_8ToWord dflags)) b8 res args -emitPrimOp dflags res ReadByteArrayOp_WideChar args = doIndexByteArrayOp (Just (mo_u_32ToWord dflags)) b32 res args -emitPrimOp dflags res ReadByteArrayOp_Int args = doIndexByteArrayOp Nothing (bWord dflags) res args -emitPrimOp dflags res ReadByteArrayOp_Word args = doIndexByteArrayOp Nothing (bWord dflags) res args -emitPrimOp dflags res ReadByteArrayOp_Addr args = doIndexByteArrayOp Nothing (bWord dflags) res args -emitPrimOp _ res ReadByteArrayOp_Float args = doIndexByteArrayOp Nothing f32 res args -emitPrimOp _ res ReadByteArrayOp_Double args = doIndexByteArrayOp Nothing f64 res args -emitPrimOp dflags res ReadByteArrayOp_StablePtr args = doIndexByteArrayOp Nothing (bWord dflags) res args -emitPrimOp dflags res ReadByteArrayOp_Int8 args = doIndexByteArrayOp (Just (mo_s_8ToWord dflags)) b8 res args -emitPrimOp dflags res ReadByteArrayOp_Int16 args = doIndexByteArrayOp (Just (mo_s_16ToWord dflags)) b16 res args -emitPrimOp dflags res ReadByteArrayOp_Int32 args = doIndexByteArrayOp (Just (mo_s_32ToWord dflags)) b32 res args -emitPrimOp _ res ReadByteArrayOp_Int64 args = doIndexByteArrayOp Nothing b64 res args -emitPrimOp dflags res ReadByteArrayOp_Word8 args = doIndexByteArrayOp (Just (mo_u_8ToWord dflags)) b8 res args -emitPrimOp dflags res ReadByteArrayOp_Word16 args = doIndexByteArrayOp (Just (mo_u_16ToWord dflags)) b16 res args -emitPrimOp dflags res ReadByteArrayOp_Word32 args = doIndexByteArrayOp (Just (mo_u_32ToWord dflags)) b32 res args -emitPrimOp _ res ReadByteArrayOp_Word64 args = doIndexByteArrayOp Nothing b64 res args +emitPrimOp dflags res ReadByteArrayOp_Char args = doIndexByteArrayOp (Just (mo_u_8ToWord dflags)) b8 res args +emitPrimOp dflags res ReadByteArrayOp_WideChar args = doIndexByteArrayOp (Just (mo_u_32ToWord dflags)) b32 res args +emitPrimOp dflags res ReadByteArrayOp_Int args = doIndexByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res ReadByteArrayOp_Word args = doIndexByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res ReadByteArrayOp_Addr args = doIndexByteArrayOp Nothing (bWord dflags) res args +emitPrimOp _ res ReadByteArrayOp_Float args = doIndexByteArrayOp Nothing f32 res args +emitPrimOp _ res ReadByteArrayOp_Double args = doIndexByteArrayOp Nothing f64 res args +emitPrimOp dflags res ReadByteArrayOp_StablePtr args = doIndexByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res ReadByteArrayOp_Int8 args = doIndexByteArrayOp (Just (mo_s_8ToWord dflags)) b8 res args +emitPrimOp dflags res ReadByteArrayOp_Int16 args = doIndexByteArrayOp (Just (mo_s_16ToWord dflags)) b16 res args +emitPrimOp dflags res ReadByteArrayOp_Int32 args = doIndexByteArrayOp (Just (mo_s_32ToWord dflags)) b32 res args +emitPrimOp _ res ReadByteArrayOp_Int64 args = doIndexByteArrayOp Nothing b64 res args +emitPrimOp dflags res ReadByteArrayOp_Word8 args = doIndexByteArrayOp (Just (mo_u_8ToWord dflags)) b8 res args +emitPrimOp dflags res ReadByteArrayOp_Word16 args = doIndexByteArrayOp (Just (mo_u_16ToWord dflags)) b16 res args +emitPrimOp dflags res ReadByteArrayOp_Word32 args = doIndexByteArrayOp (Just (mo_u_32ToWord dflags)) b32 res args +emitPrimOp _ res ReadByteArrayOp_Word64 args = doIndexByteArrayOp Nothing b64 res args +emitPrimOp _ res ReadByteArrayOp_FloatX4 args = doIndexByteArrayOp Nothing vec4f32 res args +emitPrimOp _ res ReadByteArrayOp_FloatAsFloatX4 args = doIndexByteArrayOpAs Nothing vec4f32 f32 res args -- WriteXXXoffAddr -emitPrimOp dflags res WriteOffAddrOp_Char args = doWriteOffAddrOp (Just (mo_WordTo8 dflags)) res args -emitPrimOp dflags res WriteOffAddrOp_WideChar args = doWriteOffAddrOp (Just (mo_WordTo32 dflags)) res args -emitPrimOp _ res WriteOffAddrOp_Int args = doWriteOffAddrOp Nothing res args -emitPrimOp _ res WriteOffAddrOp_Word args = doWriteOffAddrOp Nothing res args -emitPrimOp _ res WriteOffAddrOp_Addr args = doWriteOffAddrOp Nothing res args -emitPrimOp _ res WriteOffAddrOp_Float args = doWriteOffAddrOp Nothing res args -emitPrimOp _ res WriteOffAddrOp_Double args = doWriteOffAddrOp Nothing res args -emitPrimOp _ res WriteOffAddrOp_StablePtr args = doWriteOffAddrOp Nothing res args -emitPrimOp dflags res WriteOffAddrOp_Int8 args = doWriteOffAddrOp (Just (mo_WordTo8 dflags)) res args -emitPrimOp dflags res WriteOffAddrOp_Int16 args = doWriteOffAddrOp (Just (mo_WordTo16 dflags)) res args -emitPrimOp dflags res WriteOffAddrOp_Int32 args = doWriteOffAddrOp (Just (mo_WordTo32 dflags)) res args -emitPrimOp _ res WriteOffAddrOp_Int64 args = doWriteOffAddrOp Nothing res args -emitPrimOp dflags res WriteOffAddrOp_Word8 args = doWriteOffAddrOp (Just (mo_WordTo8 dflags)) res args -emitPrimOp dflags res WriteOffAddrOp_Word16 args = doWriteOffAddrOp (Just (mo_WordTo16 dflags)) res args -emitPrimOp dflags res WriteOffAddrOp_Word32 args = doWriteOffAddrOp (Just (mo_WordTo32 dflags)) res args -emitPrimOp _ res WriteOffAddrOp_Word64 args = doWriteOffAddrOp Nothing res args +emitPrimOp dflags res WriteOffAddrOp_Char args = doWriteOffAddrOp (Just (mo_WordTo8 dflags)) b8 res args +emitPrimOp dflags res WriteOffAddrOp_WideChar args = doWriteOffAddrOp (Just (mo_WordTo32 dflags)) b32 res args +emitPrimOp dflags res WriteOffAddrOp_Int args = doWriteOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res WriteOffAddrOp_Word args = doWriteOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res WriteOffAddrOp_Addr args = doWriteOffAddrOp Nothing (bWord dflags) res args +emitPrimOp _ res WriteOffAddrOp_Float args = doWriteOffAddrOp Nothing f32 res args +emitPrimOp _ res WriteOffAddrOp_Double args = doWriteOffAddrOp Nothing f64 res args +emitPrimOp dflags res WriteOffAddrOp_StablePtr args = doWriteOffAddrOp Nothing (bWord dflags) res args +emitPrimOp dflags res WriteOffAddrOp_Int8 args = doWriteOffAddrOp (Just (mo_WordTo8 dflags)) b8 res args +emitPrimOp dflags res WriteOffAddrOp_Int16 args = doWriteOffAddrOp (Just (mo_WordTo16 dflags)) b16 res args +emitPrimOp dflags res WriteOffAddrOp_Int32 args = doWriteOffAddrOp (Just (mo_WordTo32 dflags)) b32 res args +emitPrimOp _ res WriteOffAddrOp_Int64 args = doWriteOffAddrOp Nothing b64 res args +emitPrimOp dflags res WriteOffAddrOp_Word8 args = doWriteOffAddrOp (Just (mo_WordTo8 dflags)) b8 res args +emitPrimOp dflags res WriteOffAddrOp_Word16 args = doWriteOffAddrOp (Just (mo_WordTo16 dflags)) b16 res args +emitPrimOp dflags res WriteOffAddrOp_Word32 args = doWriteOffAddrOp (Just (mo_WordTo32 dflags)) b32 res args +emitPrimOp _ res WriteOffAddrOp_Word64 args = doWriteOffAddrOp Nothing b64 res args +emitPrimOp _ res WriteOffAddrOp_FloatX4 args = doWriteOffAddrOp Nothing vec4f32 res args +emitPrimOp _ res WriteOffAddrOp_FloatAsFloatX4 args = doWriteOffAddrOp Nothing f32 res args -- WriteXXXArray -emitPrimOp dflags res WriteByteArrayOp_Char args = doWriteByteArrayOp (Just (mo_WordTo8 dflags)) res args -emitPrimOp dflags res WriteByteArrayOp_WideChar args = doWriteByteArrayOp (Just (mo_WordTo32 dflags)) res args -emitPrimOp _ res WriteByteArrayOp_Int args = doWriteByteArrayOp Nothing res args -emitPrimOp _ res WriteByteArrayOp_Word args = doWriteByteArrayOp Nothing res args -emitPrimOp _ res WriteByteArrayOp_Addr args = doWriteByteArrayOp Nothing res args -emitPrimOp _ res WriteByteArrayOp_Float args = doWriteByteArrayOp Nothing res args -emitPrimOp _ res WriteByteArrayOp_Double args = doWriteByteArrayOp Nothing res args -emitPrimOp _ res WriteByteArrayOp_StablePtr args = doWriteByteArrayOp Nothing res args -emitPrimOp dflags res WriteByteArrayOp_Int8 args = doWriteByteArrayOp (Just (mo_WordTo8 dflags)) res args -emitPrimOp dflags res WriteByteArrayOp_Int16 args = doWriteByteArrayOp (Just (mo_WordTo16 dflags)) res args -emitPrimOp dflags res WriteByteArrayOp_Int32 args = doWriteByteArrayOp (Just (mo_WordTo32 dflags)) res args -emitPrimOp _ res WriteByteArrayOp_Int64 args = doWriteByteArrayOp Nothing res args -emitPrimOp dflags res WriteByteArrayOp_Word8 args = doWriteByteArrayOp (Just (mo_WordTo8 dflags)) res args -emitPrimOp dflags res WriteByteArrayOp_Word16 args = doWriteByteArrayOp (Just (mo_WordTo16 dflags)) res args -emitPrimOp dflags res WriteByteArrayOp_Word32 args = doWriteByteArrayOp (Just (mo_WordTo32 dflags)) res args -emitPrimOp _ res WriteByteArrayOp_Word64 args = doWriteByteArrayOp Nothing res args +emitPrimOp dflags res WriteByteArrayOp_Char args = doWriteByteArrayOp (Just (mo_WordTo8 dflags)) b8 res args +emitPrimOp dflags res WriteByteArrayOp_WideChar args = doWriteByteArrayOp (Just (mo_WordTo32 dflags)) b32 res args +emitPrimOp dflags res WriteByteArrayOp_Int args = doWriteByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res WriteByteArrayOp_Word args = doWriteByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res WriteByteArrayOp_Addr args = doWriteByteArrayOp Nothing (bWord dflags) res args +emitPrimOp _ res WriteByteArrayOp_Float args = doWriteByteArrayOp Nothing f32 res args +emitPrimOp _ res WriteByteArrayOp_Double args = doWriteByteArrayOp Nothing f64 res args +emitPrimOp dflags res WriteByteArrayOp_StablePtr args = doWriteByteArrayOp Nothing (bWord dflags) res args +emitPrimOp dflags res WriteByteArrayOp_Int8 args = doWriteByteArrayOp (Just (mo_WordTo8 dflags)) b8 res args +emitPrimOp dflags res WriteByteArrayOp_Int16 args = doWriteByteArrayOp (Just (mo_WordTo16 dflags)) b16 res args +emitPrimOp dflags res WriteByteArrayOp_Int32 args = doWriteByteArrayOp (Just (mo_WordTo32 dflags)) b32 res args +emitPrimOp _ res WriteByteArrayOp_Int64 args = doWriteByteArrayOp Nothing b64 res args +emitPrimOp dflags res WriteByteArrayOp_Word8 args = doWriteByteArrayOp (Just (mo_WordTo8 dflags)) b8 res args +emitPrimOp dflags res WriteByteArrayOp_Word16 args = doWriteByteArrayOp (Just (mo_WordTo16 dflags)) b16 res args +emitPrimOp dflags res WriteByteArrayOp_Word32 args = doWriteByteArrayOp (Just (mo_WordTo32 dflags)) b32 res args +emitPrimOp _ res WriteByteArrayOp_Word64 args = doWriteByteArrayOp Nothing b64 res args +emitPrimOp _ res WriteByteArrayOp_FloatX4 args = doWriteByteArrayOp Nothing vec4f32 res args +emitPrimOp _ res WriteByteArrayOp_FloatAsFloatX4 args = doWriteByteArrayOp Nothing f32 res args -- Copying and setting byte arrays emitPrimOp _ [] CopyByteArrayOp [src,src_off,dst,dst_off,n] = @@ -498,6 +510,25 @@ emitPrimOp _ [res] Word2FloatOp [w] = emitPrimCall [res] emitPrimOp _ [res] Word2DoubleOp [w] = emitPrimCall [res] (MO_UF_Conv W64) [w] +-- SIMD vector packing and unpacking +emitPrimOp _ [res] FloatToFloatX4Op [e] = + doVecPackOp Nothing vec4f32 zero [e,e,e,e] res + where + zero :: CmmExpr + zero = CmmLit $ CmmVec (replicate 4 (CmmFloat 0 W32)) + +emitPrimOp _ [res] FloatX4PackOp es@[_,_,_,_] = + doVecPackOp Nothing vec4f32 zero es res + where + zero :: CmmExpr + zero = CmmLit $ CmmVec (replicate 4 (CmmFloat 0 W32)) + +emitPrimOp _ res@[_,_,_,_] FloatX4UnpackOp [arg] = + doVecUnpackOp Nothing vec4f32 arg res + +emitPrimOp _ [res] FloatX4InsertOp [v,e,i] = + doVecInsertOp Nothing vec4f32 v e i res + -- The rest just translate straightforwardly emitPrimOp dflags [res] op [arg] | nopOp op @@ -804,6 +835,14 @@ translateOp _ FloatMulOp = Just (MO_F_Mul W32) translateOp _ FloatDivOp = Just (MO_F_Quot W32) translateOp _ FloatNegOp = Just (MO_F_Neg W32) +-- Floating point vector ops + +translateOp _ FloatX4AddOp = Just (MO_VF_Add 4 W32) +translateOp _ FloatX4SubOp = Just (MO_VF_Sub 4 W32) +translateOp _ FloatX4MulOp = Just (MO_VF_Mul 4 W32) +translateOp _ FloatX4DivOp = Just (MO_VF_Quot 4 W32) +translateOp _ FloatX4NegOp = Just (MO_VF_Neg 4 W32) + -- Conversions translateOp dflags Int2DoubleOp = Just (MO_SF_Conv (wordWidth dflags) W64) @@ -864,42 +903,87 @@ callishOp _ = Nothing ------------------------------------------------------------------------------ -- Helpers for translating various minor variants of array indexing. -doIndexOffAddrOp :: Maybe MachOp -> CmmType -> [LocalReg] -> [CmmExpr] -> FCode () +doIndexOffAddrOp :: Maybe MachOp + -> CmmType + -> [LocalReg] + -> [CmmExpr] + -> FCode () doIndexOffAddrOp maybe_post_read_cast rep [res] [addr,idx] - = mkBasicIndexedRead 0 maybe_post_read_cast rep res addr idx + = mkBasicIndexedRead 0 maybe_post_read_cast rep res addr rep idx doIndexOffAddrOp _ _ _ _ - = panic "CgPrimOp: doIndexOffAddrOp" - -doIndexByteArrayOp :: Maybe MachOp -> CmmType -> [LocalReg] -> [CmmExpr] -> FCode () + = panic "StgCmmPrim: doIndexOffAddrOp" + +doIndexOffAddrOpAs :: Maybe MachOp + -> CmmType + -> CmmType + -> [LocalReg] + -> [CmmExpr] + -> FCode () +doIndexOffAddrOpAs maybe_post_read_cast rep idx_rep [res] [addr,idx] + = mkBasicIndexedRead 0 maybe_post_read_cast rep res addr idx_rep idx +doIndexOffAddrOpAs _ _ _ _ _ + = panic "StgCmmPrim: doIndexOffAddrOpAs" + +doIndexByteArrayOp :: Maybe MachOp + -> CmmType + -> [LocalReg] + -> [CmmExpr] + -> FCode () doIndexByteArrayOp maybe_post_read_cast rep [res] [addr,idx] = do dflags <- getDynFlags - mkBasicIndexedRead (arrWordsHdrSize dflags) maybe_post_read_cast rep res addr idx -doIndexByteArrayOp _ _ _ _ - = panic "CgPrimOp: doIndexByteArrayOp" + mkBasicIndexedRead (arrWordsHdrSize dflags) maybe_post_read_cast rep res addr rep idx +doIndexByteArrayOp _ _ _ _ + = panic "StgCmmPrim: doIndexByteArrayOp" + +doIndexByteArrayOpAs :: Maybe MachOp + -> CmmType + -> CmmType + -> [LocalReg] + -> [CmmExpr] + -> FCode () +doIndexByteArrayOpAs maybe_post_read_cast rep idx_rep [res] [addr,idx] + = do dflags <- getDynFlags + mkBasicIndexedRead (arrWordsHdrSize dflags) maybe_post_read_cast rep res addr idx_rep idx +doIndexByteArrayOpAs _ _ _ _ _ + = panic "StgCmmPrim: doIndexByteArrayOpAs" -doReadPtrArrayOp :: LocalReg -> CmmExpr -> CmmExpr -> FCode () +doReadPtrArrayOp :: LocalReg + -> CmmExpr + -> CmmExpr + -> FCode () doReadPtrArrayOp res addr idx = do dflags <- getDynFlags - mkBasicIndexedRead (arrPtrsHdrSize dflags) Nothing (gcWord dflags) res addr idx - - -doWriteOffAddrOp :: Maybe MachOp -> [LocalReg] -> [CmmExpr] -> FCode () -doWriteOffAddrOp maybe_pre_write_cast [] [addr,idx,val] - = mkBasicIndexedWrite 0 maybe_pre_write_cast addr idx val -doWriteOffAddrOp _ _ _ - = panic "CgPrimOp: doWriteOffAddrOp" + mkBasicIndexedRead (arrPtrsHdrSize dflags) Nothing (gcWord dflags) res addr (gcWord dflags) idx -doWriteByteArrayOp :: Maybe MachOp -> [LocalReg] -> [CmmExpr] -> FCode () -doWriteByteArrayOp maybe_pre_write_cast [] [addr,idx,val] +doWriteOffAddrOp :: Maybe MachOp + -> CmmType + -> [LocalReg] + -> [CmmExpr] + -> FCode () +doWriteOffAddrOp maybe_pre_write_cast idx_ty [] [addr,idx,val] + = mkBasicIndexedWrite 0 maybe_pre_write_cast addr idx_ty idx val +doWriteOffAddrOp _ _ _ _ + = panic "StgCmmPrim: doWriteOffAddrOp" + +doWriteByteArrayOp :: Maybe MachOp + -> CmmType + -> [LocalReg] + -> [CmmExpr] + -> FCode () +doWriteByteArrayOp maybe_pre_write_cast idx_ty [] [addr,idx,val] = do dflags <- getDynFlags - mkBasicIndexedWrite (arrWordsHdrSize dflags) maybe_pre_write_cast addr idx val -doWriteByteArrayOp _ _ _ - = panic "CgPrimOp: doWriteByteArrayOp" + mkBasicIndexedWrite (arrWordsHdrSize dflags) maybe_pre_write_cast addr idx_ty idx val +doWriteByteArrayOp _ _ _ _ + = panic "StgCmmPrim: doWriteByteArrayOp" -doWritePtrArrayOp :: CmmExpr -> CmmExpr -> CmmExpr -> FCode () +doWritePtrArrayOp :: CmmExpr + -> CmmExpr + -> CmmExpr + -> FCode () doWritePtrArrayOp addr idx val = do dflags <- getDynFlags - mkBasicIndexedWrite (arrPtrsHdrSize dflags) Nothing addr idx val + let ty = cmmExprType dflags val + mkBasicIndexedWrite (arrPtrsHdrSize dflags) Nothing addr ty idx val emit (setInfo addr (CmmLit (CmmLabel mkMAP_DIRTY_infoLabel))) -- the write barrier. We must write a byte into the mark table: -- bits8[a + header_size + StgMutArrPtrs_size(a) + x >> N] @@ -915,38 +999,154 @@ loadArrPtrsSize :: DynFlags -> CmmExpr -> CmmExpr loadArrPtrsSize dflags addr = CmmLoad (cmmOffsetB dflags addr off) (bWord dflags) where off = fixedHdrSize dflags * wORD_SIZE dflags + oFFSET_StgMutArrPtrs_ptrs dflags -mkBasicIndexedRead :: ByteOff -> Maybe MachOp -> CmmType - -> LocalReg -> CmmExpr -> CmmExpr -> FCode () -mkBasicIndexedRead off Nothing read_rep res base idx +mkBasicIndexedRead :: ByteOff -- Initial offset in bytes + -> Maybe MachOp -- Optional result cast + -> CmmType -- Type of element we are accessing + -> LocalReg -- Destination + -> CmmExpr -- Base address + -> CmmType -- Type of element by which we are indexing + -> CmmExpr -- Index + -> FCode () +mkBasicIndexedRead off Nothing ty res base idx_ty idx = do dflags <- getDynFlags - emitAssign (CmmLocal res) (cmmLoadIndexOffExpr dflags off read_rep base idx) -mkBasicIndexedRead off (Just cast) read_rep res base idx + emitAssign (CmmLocal res) (cmmLoadIndexOffExpr dflags off ty base idx_ty idx) +mkBasicIndexedRead off (Just cast) ty res base idx_ty idx = do dflags <- getDynFlags emitAssign (CmmLocal res) (CmmMachOp cast [ - cmmLoadIndexOffExpr dflags off read_rep base idx]) - -mkBasicIndexedWrite :: ByteOff -> Maybe MachOp - -> CmmExpr -> CmmExpr -> CmmExpr -> FCode () -mkBasicIndexedWrite off Nothing base idx val + cmmLoadIndexOffExpr dflags off ty base idx_ty idx]) + +mkBasicIndexedWrite :: ByteOff -- Initial offset in bytes + -> Maybe MachOp -- Optional value cast + -> CmmExpr -- Base address + -> CmmType -- Type of element by which we are indexing + -> CmmExpr -- Index + -> CmmExpr -- Value to write + -> FCode () +mkBasicIndexedWrite off Nothing base idx_ty idx val = do dflags <- getDynFlags - emitStore (cmmIndexOffExpr dflags off (typeWidth (cmmExprType dflags val)) base idx) val -mkBasicIndexedWrite off (Just cast) base idx val - = mkBasicIndexedWrite off Nothing base idx (CmmMachOp cast [val]) + emitStore (cmmIndexOffExpr dflags off (typeWidth idx_ty) base idx) val +mkBasicIndexedWrite off (Just cast) base idx_ty idx val + = mkBasicIndexedWrite off Nothing base idx_ty idx (CmmMachOp cast [val]) -- ---------------------------------------------------------------------------- -- Misc utils -cmmIndexOffExpr :: DynFlags -> ByteOff -> Width -> CmmExpr -> CmmExpr -> CmmExpr +cmmIndexOffExpr :: DynFlags + -> ByteOff -- Initial offset in bytes + -> Width -- Width of element by which we are indexing + -> CmmExpr -- Base address + -> CmmExpr -- Index + -> CmmExpr cmmIndexOffExpr dflags off width base idx = cmmIndexExpr dflags width (cmmOffsetB dflags base off) idx -cmmLoadIndexOffExpr :: DynFlags -> ByteOff -> CmmType -> CmmExpr -> CmmExpr -> CmmExpr -cmmLoadIndexOffExpr dflags off ty base idx - = CmmLoad (cmmIndexOffExpr dflags off (typeWidth ty) base idx) ty +cmmLoadIndexOffExpr :: DynFlags + -> ByteOff -- Initial offset in bytes + -> CmmType -- Type of element we are accessing + -> CmmExpr -- Base address + -> CmmType -- Type of element by which we are indexing + -> CmmExpr -- Index + -> CmmExpr +cmmLoadIndexOffExpr dflags off ty base idx_ty idx + = CmmLoad (cmmIndexOffExpr dflags off (typeWidth idx_ty) base idx) ty setInfo :: CmmExpr -> CmmExpr -> CmmAGraph setInfo closure_ptr info_ptr = mkStore closure_ptr info_ptr +------------------------------------------------------------------------------ +-- Helpers for translating vector packing and unpacking. + +doVecPackOp :: Maybe MachOp -- Cast from element to vector component + -> CmmType -- Type of vector + -> CmmExpr -- Initial vector + -> [CmmExpr] -- Elements + -> CmmFormal -- Destination for result + -> FCode () +doVecPackOp maybe_pre_write_cast ty z es res = do + dst <- newTemp ty + emitAssign (CmmLocal dst) z + vecPack dst es 0 + where + vecPack :: CmmFormal -> [CmmExpr] -> Int -> FCode () + vecPack src [] _ = + emitAssign (CmmLocal res) (CmmReg (CmmLocal src)) + + vecPack src (e : es) i = do + dst <- newTemp ty + emitAssign (CmmLocal dst) (CmmMachOp (MO_V_Insert len wid) + [CmmReg (CmmLocal src), cast e, iLit]) + vecPack dst es (i + 1) + where + -- vector indices are always 32-bits + iLit = CmmLit (CmmInt (toInteger i) W32) + + cast :: CmmExpr -> CmmExpr + cast val = case maybe_pre_write_cast of + Nothing -> val + Just cast -> CmmMachOp cast [val] + + len :: Length + len = vecLength ty + + wid :: Width + wid = typeWidth (vecElemType ty) + +doVecUnpackOp :: Maybe MachOp -- Cast from vector component to element result + -> CmmType -- Type of vector + -> CmmExpr -- Vector + -> [CmmFormal] -- Element results + -> FCode () +doVecUnpackOp maybe_post_read_cast ty e res = + vecUnpack res 0 + where + vecUnpack :: [CmmFormal] -> Int -> FCode () + vecUnpack [] _ = + return () + + vecUnpack (r : rs) i = do + emitAssign (CmmLocal r) (cast (CmmMachOp (MO_V_Extract len wid) + [e, iLit])) + vecUnpack rs (i + 1) + where + -- vector indices are always 32-bits + iLit = CmmLit (CmmInt (toInteger i) W32) + + cast :: CmmExpr -> CmmExpr + cast val = case maybe_post_read_cast of + Nothing -> val + Just cast -> CmmMachOp cast [val] + + len :: Length + len = vecLength ty + + wid :: Width + wid = typeWidth (vecElemType ty) + +doVecInsertOp :: Maybe MachOp -- Cast from element to vector component + -> CmmType -- Vector type + -> CmmExpr -- Source vector + -> CmmExpr -- Element + -> CmmExpr -- Index at which to insert element + -> CmmFormal -- Destination for result + -> FCode () +doVecInsertOp maybe_pre_write_cast ty src e idx res = do + dflags <- getDynFlags + -- vector indices are always 32-bits + let idx' :: CmmExpr + idx' = CmmMachOp (MO_SS_Conv (wordWidth dflags) W32) [idx] + emitAssign (CmmLocal res) (CmmMachOp (MO_V_Insert len wid) [src, cast e, idx']) + where + cast :: CmmExpr -> CmmExpr + cast val = case maybe_pre_write_cast of + Nothing -> val + Just cast -> CmmMachOp cast [val] + + len :: Length + len = vecLength ty + + wid :: Width + wid = typeWidth (vecElemType ty) + -- ---------------------------------------------------------------------------- -- Copying byte arrays |