Pass 256-bit-wide vectors in registers.

compiler/cmm/CmmCallConv.hs

@@ -67,6 +67,7 @@ assignArgumentsPos dflags off conv arg_ty reps = (stk_off, assignments)
                                     | otherwise      = int
         where vec = case (w, regs) of
                       (W128, (vs, fs, ds, ls, s:ss)) -> k (RegisterParam (XmmReg s), (vs, fs, ds, ls, ss))
+                      (W256, (vs, fs, ds, ls, s:ss)) -> k (RegisterParam (YmmReg s), (vs, fs, ds, ls, ss))
                       _ -> (assts, (r:rs))
               float = case (w, regs) of
                         (W32, (vs, fs, ds, ls, s:ss))

compiler/cmm/CmmExpr.hs

@@ -343,6 +343,9 @@ data GlobalReg
   | XmmReg                      -- 128-bit SIMD vector register
         {-# UNPACK #-} !Int     -- its number
 
+  | YmmReg                      -- 256-bit SIMD vector register 
+        {-# UNPACK #-} !Int     -- its number
+
   -- STG registers
   | Sp                  -- Stack ptr; points to last occupied stack location.
   | SpLim               -- Stack limit
@@ -379,6 +382,7 @@ instance Eq GlobalReg where
    DoubleReg i == DoubleReg j = i==j
    LongReg i == LongReg j = i==j
    XmmReg i == XmmReg j = i==j
+   YmmReg i == YmmReg j = i==j
    Sp == Sp = True
    SpLim == SpLim = True
    Hp == Hp = True
@@ -401,6 +405,7 @@ instance Ord GlobalReg where
    compare (DoubleReg i) (DoubleReg j) = compare i j
    compare (LongReg i)   (LongReg   j) = compare i j
    compare (XmmReg i)    (XmmReg    j) = compare i j
+   compare (YmmReg i)    (YmmReg    j) = compare i j
    compare Sp Sp = EQ
    compare SpLim SpLim = EQ
    compare Hp Hp = EQ
@@ -424,6 +429,8 @@ instance Ord GlobalReg where
    compare _ (LongReg _)      = GT
    compare (XmmReg _) _       = LT
    compare _ (XmmReg _)       = GT
+   compare (YmmReg _) _       = LT
+   compare _ (YmmReg _)       = GT
    compare Sp _ = LT
    compare _ Sp = GT
    compare SpLim _ = LT
@@ -467,6 +474,7 @@ globalRegType _      (FloatReg _)      = cmmFloat W32
 globalRegType _      (DoubleReg _)     = cmmFloat W64
 globalRegType _      (LongReg _)       = cmmBits W64
 globalRegType _      (XmmReg _)        = cmmVec 4 (cmmBits W32)
+globalRegType _      (YmmReg _)        = cmmVec 8 (cmmBits W32)
 
 globalRegType dflags Hp                = gcWord dflags
                                             -- The initialiser for all
@@ -479,4 +487,5 @@ isArgReg (FloatReg {})   = True
 isArgReg (DoubleReg {})  = True
 isArgReg (LongReg {})    = True
 isArgReg (XmmReg {})     = True
+isArgReg (YmmReg {})     = True
 isArgReg _               = False

compiler/cmm/PprCmmExpr.hs

@@ -256,6 +256,7 @@ pprGlobalReg gr
         DoubleReg  n   -> char 'D' <> int n
         LongReg    n   -> char 'L' <> int n
         XmmReg     n   -> ptext (sLit "XMM") <> int n
+        YmmReg     n   -> ptext (sLit "YMM") <> int n
         Sp             -> ptext (sLit "Sp")
         SpLim          -> ptext (sLit "SpLim")
         Hp             -> ptext (sLit "Hp")

compiler/codeGen/CgUtils.hs

@@ -56,6 +56,13 @@ baseRegOffset dflags (XmmReg 4)          = oFFSET_StgRegTable_rXMM4 dflags
 baseRegOffset dflags (XmmReg 5)          = oFFSET_StgRegTable_rXMM5 dflags
 baseRegOffset dflags (XmmReg 6)          = oFFSET_StgRegTable_rXMM6 dflags
 baseRegOffset _      (XmmReg n)          = panic ("Registers above XMM6 are not supported (tried to use XMM" ++ show n ++ ")")
+baseRegOffset dflags (YmmReg 1)          = oFFSET_StgRegTable_rYMM1 dflags
+baseRegOffset dflags (YmmReg 2)          = oFFSET_StgRegTable_rYMM2 dflags
+baseRegOffset dflags (YmmReg 3)          = oFFSET_StgRegTable_rYMM3 dflags
+baseRegOffset dflags (YmmReg 4)          = oFFSET_StgRegTable_rYMM4 dflags
+baseRegOffset dflags (YmmReg 5)          = oFFSET_StgRegTable_rYMM5 dflags
+baseRegOffset dflags (YmmReg 6)          = oFFSET_StgRegTable_rYMM6 dflags
+baseRegOffset _      (YmmReg n)          = panic ("Registers above YMM6 are not supported (tried to use YMM" ++ show n ++ ")")
 baseRegOffset dflags Sp                  = oFFSET_StgRegTable_rSp dflags
 baseRegOffset dflags SpLim               = oFFSET_StgRegTable_rSpLim dflags
 baseRegOffset dflags (LongReg 1)         = oFFSET_StgRegTable_rL1 dflags

compiler/llvmGen/LlvmCodeGen/Base.hs

@@ -155,6 +155,7 @@ llvmFunArgs dflags live =
           isSSE (FloatReg _)  = True
           isSSE (DoubleReg _) = True
           isSSE (XmmReg _)    = True
+          isSSE (YmmReg _)    = True
           isSSE _             = False
 
 -- | Llvm standard fun attributes

compiler/llvmGen/LlvmCodeGen/CodeGen.hs

@@ -1533,6 +1533,7 @@ funEpilogue live = do
         isSSE (FloatReg _)  = True
         isSSE (DoubleReg _) = True
         isSSE (XmmReg _)    = True
+        isSSE (YmmReg _)    = True
         isSSE _             = False
 
     -- Set to value or "undef" depending on whether the register is

compiler/llvmGen/LlvmCodeGen/Regs.hs

@@ -62,6 +62,12 @@ lmGlobalReg dflags suf reg
         XmmReg 4       -> xmmGlobal $ "XMM4" ++ suf
         XmmReg 5       -> xmmGlobal $ "XMM5" ++ suf
         XmmReg 6       -> xmmGlobal $ "XMM6" ++ suf
+        YmmReg 1       -> ymmGlobal $ "YMM1" ++ suf
+        YmmReg 2       -> ymmGlobal $ "YMM2" ++ suf
+        YmmReg 3       -> ymmGlobal $ "YMM3" ++ suf
+        YmmReg 4       -> ymmGlobal $ "YMM4" ++ suf
+        YmmReg 5       -> ymmGlobal $ "YMM5" ++ suf
+        YmmReg 6       -> ymmGlobal $ "YMM6" ++ suf
         _other         -> panic $ "LlvmCodeGen.Reg: GlobalReg (" ++ (show reg)
                                 ++ ") not supported!"
         -- LongReg, HpLim, CCSS, CurrentTSO, CurrentNusery, HpAlloc
@@ -72,6 +78,7 @@ lmGlobalReg dflags suf reg
         floatGlobal  name = LMNLocalVar (fsLit name) LMFloat
         doubleGlobal name = LMNLocalVar (fsLit name) LMDouble
         xmmGlobal    name = LMNLocalVar (fsLit name) (LMVector 4 (LMInt 32))
+        ymmGlobal    name = LMNLocalVar (fsLit name) (LMVector 8 (LMInt 32))
 
 -- | A list of STG Registers that should always be considered alive
 alwaysLive :: [GlobalReg]

includes/CodeGen.Platform.hs

@@ -65,6 +65,23 @@ import Reg
 # define xmm14 38
 # define xmm15 39
 
+# define ymm0  40
+# define ymm1  41
+# define ymm2  42
+# define ymm3  43
+# define ymm4  44
+# define ymm5  45
+# define ymm6  46
+# define ymm7  47
+# define ymm8  48
+# define ymm9  49
+# define ymm10 50
+# define ymm11 51
+# define ymm12 52
+# define ymm13 53
+# define ymm14 54
+# define ymm15 55
+
 #elif MACHREGS_powerpc
 
 # define r0 0
@@ -391,6 +408,9 @@ activeStgRegs = [
 #ifdef REG_XMM1
     ,XmmReg 1
 #endif
+#ifdef REG_YMM1
+    ,YmmReg 1
+#endif
 #ifdef REG_F2
     ,FloatReg 2
 #endif
@@ -400,6 +420,9 @@ activeStgRegs = [
 #ifdef REG_XMM2
     ,XmmReg 2
 #endif
+#ifdef REG_YMM2
+    ,YmmReg 2
+#endif
 #ifdef REG_F3
     ,FloatReg 3
 #endif
@@ -409,6 +432,9 @@ activeStgRegs = [
 #ifdef REG_XMM3
     ,XmmReg 3
 #endif
+#ifdef REG_YMM3
+    ,YmmReg 3
+#endif
 #ifdef REG_F4
     ,FloatReg 4
 #endif
@@ -418,6 +444,9 @@ activeStgRegs = [
 #ifdef REG_XMM4
     ,XmmReg 4
 #endif
+#ifdef REG_YMM4
+    ,YmmReg 4
+#endif
 #ifdef REG_F5
     ,FloatReg 5
 #endif
@@ -427,6 +456,9 @@ activeStgRegs = [
 #ifdef REG_XMM5
     ,XmmReg 5
 #endif
+#ifdef REG_YMM5
+    ,YmmReg 5
+#endif
 #ifdef REG_F6
     ,FloatReg 6
 #endif
@@ -436,6 +468,9 @@ activeStgRegs = [
 #ifdef REG_XMM6
     ,XmmReg 6
 #endif
+#ifdef REG_YMM6
+    ,YmmReg 6
+#endif
 #else /* MAX_REAL_XMM_REG == 0 */
 #ifdef REG_F1
     ,FloatReg 1
@@ -607,6 +642,26 @@ globalRegMaybe (XmmReg 5)               = Just (RealRegSingle REG_XMM5)
 globalRegMaybe (XmmReg 6)               = Just (RealRegSingle REG_XMM6)
 #  endif
 # endif
+# if MAX_REAL_YMM_REG != 0
+#  ifdef REG_YMM1
+globalRegMaybe (YmmReg 1)               = Just (RealRegSingle REG_YMM1)
+#  endif
+#  ifdef REG_YMM2
+globalRegMaybe (YmmReg 2)               = Just (RealRegSingle REG_YMM2)
+#  endif
+#  ifdef REG_YMM3
+globalRegMaybe (YmmReg 3)               = Just (RealRegSingle REG_YMM3)
+#  endif
+#  ifdef REG_YMM4
+globalRegMaybe (YmmReg 4)               = Just (RealRegSingle REG_YMM4)
+#  endif
+#  ifdef REG_YMM5
+globalRegMaybe (YmmReg 5)               = Just (RealRegSingle REG_YMM5)
+#  endif
+#  ifdef REG_YMM6
+globalRegMaybe (YmmReg 6)               = Just (RealRegSingle REG_YMM6)
+#  endif
+# endif
 # ifdef REG_Sp
 globalRegMaybe Sp                       = Just (RealRegSingle REG_Sp)
 # endif

includes/stg/MachRegs.h

@@ -108,11 +108,17 @@
 #define REG_XMM3    xmm2
 #define REG_XMM4    xmm3
 
+#define REG_YMM1    ymm0
+#define REG_YMM2    ymm1
+#define REG_YMM3    ymm2
+#define REG_YMM4    ymm3
+
 #define MAX_REAL_VANILLA_REG 1  /* always, since it defines the entry conv */
 #define MAX_REAL_FLOAT_REG   0
 #define MAX_REAL_DOUBLE_REG  0
 #define MAX_REAL_LONG_REG    0
 #define MAX_REAL_XMM_REG     4
+#define MAX_REAL_YMM_REG     4
 
 /* -----------------------------------------------------------------------------
   The x86-64 register mapping
@@ -179,6 +185,13 @@
 #define REG_XMM5    xmm5
 #define REG_XMM6    xmm6
 
+#define REG_YMM1    ymm1
+#define REG_YMM2    ymm2
+#define REG_YMM3    ymm3
+#define REG_YMM4    ymm4
+#define REG_YMM5    ymm5
+#define REG_YMM6    ymm6
+
 #if !defined(mingw32_HOST_OS)
 #define CALLER_SAVES_R3
 #define CALLER_SAVES_R4
@@ -213,11 +226,21 @@
 #define CALLER_SAVES_XMM6
 #endif
 
+#define CALLER_SAVES_YMM1
+#define CALLER_SAVES_YMM2
+#define CALLER_SAVES_YMM3
+#define CALLER_SAVES_YMM4
+#define CALLER_SAVES_YMM5
+#if !defined(mingw32_HOST_OS)
+#define CALLER_SAVES_YMM6
+#endif
+
 #define MAX_REAL_VANILLA_REG 6
 #define MAX_REAL_FLOAT_REG   6
 #define MAX_REAL_DOUBLE_REG  6
 #define MAX_REAL_LONG_REG    0
 #define MAX_REAL_XMM_REG     6
+#define MAX_REAL_YMM_REG     6
 
 /* -----------------------------------------------------------------------------
    The PowerPC register mapping

includes/stg/Regs.h

@@ -87,6 +87,12 @@ typedef struct {
   StgWord128 	  rXMM4;
   StgWord128 	  rXMM5;
   StgWord128 	  rXMM6;
+  StgWord256 	  rYMM1;
+  StgWord256 	  rYMM2;
+  StgWord256 	  rYMM3;
+  StgWord256 	  rYMM4;
+  StgWord256 	  rYMM5;
+  StgWord256 	  rYMM6;
   StgWord64       rL1;
   StgPtr 	  rSp;
   StgPtr 	  rSpLim;
@@ -312,6 +318,42 @@ GLOBAL_REG_DECL(StgWord128,XMM6,REG_XMM6)
 #define XMM6 (BaseReg->rXMM6)
 #endif
 
+#if defined(REG_YMM1) && !defined(NO_GLOBAL_REG_DECLS)
+GLOBAL_REG_DECL(StgWord256,YMM1,REG_YMM1)
+#else
+#define YMM1 (BaseReg->rYMM1)
+#endif
+
+#if defined(REG_YMM2) && !defined(NO_GLOBAL_REG_DECLS)
+GLOBAL_REG_DECL(StgWord256,YMM2,REG_YMM2)
+#else
+#define YMM2 (BaseReg->rYMM2)
+#endif
+
+#if defined(REG_YMM3) && !defined(NO_GLOBAL_REG_DECLS)
+GLOBAL_REG_DECL(StgWord256,YMM3,REG_YMM3)
+#else
+#define YMM3 (BaseReg->rYMM3)
+#endif
+
+#if defined(REG_YMM4) && !defined(NO_GLOBAL_REG_DECLS)
+GLOBAL_REG_DECL(StgWord256,YMM4,REG_YMM4)
+#else
+#define YMM4 (BaseReg->rYMM4)
+#endif
+
+#if defined(REG_YMM5) && !defined(NO_GLOBAL_REG_DECLS)
+GLOBAL_REG_DECL(StgWord256,YMM5,REG_YMM5)
+#else
+#define YMM5 (BaseReg->rYMM5)
+#endif
+
+#if defined(REG_YMM6) && !defined(NO_GLOBAL_REG_DECLS)
+GLOBAL_REG_DECL(StgWord256,YMM6,REG_YMM6)
+#else
+#define YMM6 (BaseReg->rYMM6)
+#endif
+
 #if defined(REG_L1) && !defined(NO_GLOBAL_REG_DECLS)
 GLOBAL_REG_DECL(StgWord64,L1,REG_L1)
 #else

includes/stg/Types.h

@@ -85,6 +85,8 @@ typedef unsigned long long int StgWord64;
 
 typedef struct { StgWord64 h; StgWord64 l; } StgWord128;
 
+typedef struct { StgWord128 h; StgWord128 l; } StgWord256;
+
 /*
  * Define the standard word size we'll use on this machine: make it
  * big enough to hold a pointer.

utils/deriveConstants/DeriveConstants.hs

@@ -313,6 +313,12 @@ wanteds = concat
           ,fieldOffset Both "StgRegTable" "rXMM4"
           ,fieldOffset Both "StgRegTable" "rXMM5"
           ,fieldOffset Both "StgRegTable" "rXMM6"
+          ,fieldOffset Both "StgRegTable" "rYMM1"
+          ,fieldOffset Both "StgRegTable" "rYMM2"
+          ,fieldOffset Both "StgRegTable" "rYMM3"
+          ,fieldOffset Both "StgRegTable" "rYMM4"
+          ,fieldOffset Both "StgRegTable" "rYMM5"
+          ,fieldOffset Both "StgRegTable" "rYMM6"
           ,fieldOffset Both "StgRegTable" "rL1"
           ,fieldOffset Both "StgRegTable" "rSp"
           ,fieldOffset Both "StgRegTable" "rSpLim"