#8033: add AVX register support to llvm calling convention
-------------------------------------+------------------------------------
Reporter: carter | Owner: carter
Type: task | Status: new
Priority: normal | Milestone:
Component: Compiler | Version: 7.7
Resolution: | Keywords:
Operating System: Unknown/Multiple | Architecture: Unknown/Multiple
Type of failure: None/Unknown | Difficulty: Unknown
Test Case: | Blocked By:
Blocking: | Related Tickets:
-------------------------------------+------------------------------------
Comment (by carter):
Ok, so we should change the hasSSE1 to hasSSE2 to simplify adding short
vector / simd support uniformly when its available (ie, for now NOT have
32bit ghc put floats and doubles in registers, thats something we can
consider at some future point perhaps).
So for now (a)
{{{
// Pass in STG registers for floats, doubles and 128bit simd vectors
CCIfType<[f32, f64, v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
CCIfSubtarget<"hasSSE2()",
CCAssignToReg<[XMM1, XMM2, XMM3, XMM4, XMM5, XMM6]>>>,
// Pass in STG registers for 256bit simd vectors
CCIfType<[v32i8, v16i16, v8i32, v4i64, v8f32, v4f64],
CCIfSubtarget<"hasAVX()",
CCAssignToReg<[ YMM1, YMM2, YMM3,
YMM4, YMM5, YMM6]>>>
}}}
for x86_64
and
{{{
// Pass in STG registers for floats, doubles and 128bit simd vectors
CCIfType<[ v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
CCIfSubtarget<"hasSSE2()",
CCAssignToReg<[XMM1, XMM2, XMM3, XMM4, XMM5, XMM6]>>>,
// Pass in STG registers for 256bit simd vectors
CCIfType<[v32i8, v16i16, v8i32, v4i64, v8f32, v4f64],
CCIfSubtarget<"hasAVX()",
CCAssignToReg<[ YMM1, YMM2, YMM3,
YMM4, YMM5, YMM6]>>>
}}}
for x86_32? (though we could use the same stanza )
or
(b)
{{{
// Pass in STG registers for floats, doubles and 128bit simd vectors
CCIfType<[f32, f64, v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
CCIfSubtarget<"hasSSE2()",
CCAssignToReg<[XMM1, XMM2, XMM3, XMM4, XMM5, XMM6]>>>,
// Pass in STG registers for 256bit simd vectors
CCIfType<[v32i8, v16i16, v8i32, v4i64, v8f32, v4f64],
CCIfSubtarget<"hasAVX()",
CCAssignToReg<[ YMM1, YMM2, YMM3,
YMM4, YMM5, YMM6]>>>
}}}
for BOTH? (but not use the registers for floats and doubles in x86_32 for
now)
either change will be compatible with past and current GHCs
--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8033#comment:26>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
ghc-tickets mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/ghc-tickets
