removing x87 register support from native code gen

* simplifies registers to have GPR, Float and Double, by removing the SSE2 and X87 Constructors
* makes -msse2 assumed/default for x86 platforms, fixing a long standing nondeterminism in rounding
behavior in 32bit haskell code
* removes the 80bit floating point representation from the supported float sizes
* theres still 1 tiny bit of x87 support needed,
for handling float and double return values in FFI calls  wrt the C ABI on x86_32,
but this one piece does not leak into the rest of NCG.
* Lots of code thats not been touched in a long time got deleted as a
consequence of all of this

all in all, this change paves the way towards a lot of future further
improvements in how GHC handles floating point computations, along with
making the native code gen more accessible to a larger pool of contributors.

3 files changed, 12 insertions, 9 deletions

diff --git a/compiler/cmm/CmmCallConv.hs b/compiler/cmm/CmmCallConv.hs
index f338415adc..078390638d 100644
--- a/compiler/cmm/CmmCallConv.hs
+++ b/compiler/cmm/CmmCallConv.hs
@@ -81,7 +81,6 @@ assignArgumentsPos dflags off conv arg_ty reps = (stk_off, assignments)
                             | passFloatInXmm          -> k (RegisterParam (DoubleReg s), (vs, fs, ds, ls, ss))
                         (W64, (vs, fs, d:ds, ls, ss))
                             | not passFloatInXmm      -> k (RegisterParam d, (vs, fs, ds, ls, ss))
-                        (W80, _) -> panic "F80 unsupported register type"
                         _ -> (assts, (r:rs))
               int = case (w, regs) of
                       (W128, _) -> panic "W128 unsupported register type"
@@ -100,6 +99,7 @@ assignArgumentsPos dflags off conv arg_ty reps = (stk_off, assignments)
 passFloatArgsInXmm :: DynFlags -> Bool
 passFloatArgsInXmm dflags = case platformArch (targetPlatform dflags) of
                               ArchX86_64 -> True
+                              ArchX86    -> False
                               _          -> False
 
 -- We used to spill vector registers to the stack since the LLVM backend didn't
diff --git a/compiler/cmm/CmmExpr.hs b/compiler/cmm/CmmExpr.hs
index 601b1d9b85..dd4e777436 100644
--- a/compiler/cmm/CmmExpr.hs
+++ b/compiler/cmm/CmmExpr.hs
@@ -474,6 +474,9 @@ instance Eq GlobalReg where
    FloatReg i == FloatReg j = i==j
    DoubleReg i == DoubleReg j = i==j
    LongReg i == LongReg j = i==j
+   -- NOTE: XMM, YMM, ZMM registers actually are the same registers
+   -- at least with respect to store at YMM i and then read from XMM i
+   -- and similarly for ZMM etc.
    XmmReg i == XmmReg j = i==j
    YmmReg i == YmmReg j = i==j
    ZmmReg i == ZmmReg j = i==j
@@ -584,6 +587,9 @@ globalRegType dflags (VanillaReg _ VNonGcPtr) = bWord dflags
 globalRegType _      (FloatReg _)      = cmmFloat W32
 globalRegType _      (DoubleReg _)     = cmmFloat W64
 globalRegType _      (LongReg _)       = cmmBits W64
+-- TODO: improve the internal model of SIMD/vectorized registers
+-- the right design SHOULd improve handling of float and double code too.
+-- see remarks in "NOTE [SIMD Design for the future]"" in StgCmmPrim
 globalRegType _      (XmmReg _)        = cmmVec 4 (cmmBits W32)
 globalRegType _      (YmmReg _)        = cmmVec 8 (cmmBits W32)
 globalRegType _      (ZmmReg _)        = cmmVec 16 (cmmBits W32)
diff --git a/compiler/cmm/CmmType.hs b/compiler/cmm/CmmType.hs
index 0d6e770904..276fbff534 100644
--- a/compiler/cmm/CmmType.hs
+++ b/compiler/cmm/CmmType.hs
@@ -166,9 +166,6 @@ isFloat64 _other                 = False
 -----------------------------------------------------------------------------
 
 data Width   = W8 | W16 | W32 | W64
-             | W80      -- Extended double-precision float,
-                        -- used in x86 native codegen only.
-                        -- (we use Ord, so it'd better be in this order)
              | W128
              | W256
              | W512
@@ -185,7 +182,7 @@ mrStr W64  = sLit("W64")
 mrStr W128 = sLit("W128")
 mrStr W256 = sLit("W256")
 mrStr W512 = sLit("W512")
-mrStr W80  = sLit("W80")
+
 
 
 -------- Common Widths  ------------
@@ -222,7 +219,7 @@ widthInBits W64  = 64
 widthInBits W128 = 128
 widthInBits W256 = 256
 widthInBits W512 = 512
-widthInBits W80  = 80
+
 
 widthInBytes :: Width -> Int
 widthInBytes W8   = 1
@@ -232,7 +229,7 @@ widthInBytes W64  = 8
 widthInBytes W128 = 16
 widthInBytes W256 = 32
 widthInBytes W512 = 64
-widthInBytes W80  = 10
+
 
 widthFromBytes :: Int -> Width
 widthFromBytes 1  = W8
@@ -242,7 +239,7 @@ widthFromBytes 8  = W64
 widthFromBytes 16 = W128
 widthFromBytes 32 = W256
 widthFromBytes 64 = W512
-widthFromBytes 10 = W80
+
 widthFromBytes n  = pprPanic "no width for given number of bytes" (ppr n)
 
 -- log_2 of the width in bytes, useful for generating shifts.
@@ -254,7 +251,7 @@ widthInLog W64  = 3
 widthInLog W128 = 4
 widthInLog W256 = 5
 widthInLog W512 = 6
-widthInLog W80  = panic "widthInLog: F80"
+
 
 -- widening / narrowing