Haskell SSE 128bit SIMD support in GHC 7.7.20130216 Compiled using ghc-7.7.20130216 -msse2 -O3 -fllvm -funbox-strict-fields -fforce-recomp -keep-llvm-files -optlo-O3 -optlc-O=3 -S simd.hs simd.ll and simd.s contain the 'most significant' parts of the compilation result.

simd.hs

{-# LANGUAGE MagicHash, UnboxedTuples, BangPatterns #-}
module Main where

import Data.Int

import System.Environment

import GHC.Int
import GHC.Prim

f :: Int32 -> Int32
f (I32# a#) =
    let !b# = packInt32X4# a# 1# 2# (0xdeadbeef#)
        !c# = plusInt32X4# b# b#
        !(# d1#, d2#, d3#, d4# #) = unpackInt32X4# c#
    in (I32# (d1# +# d2# +# d3# +# d4#))

main :: IO ()
main = return ()
{-main :: IO ()
main = do
    v <- (read . head) `fmap` getArgs
    print $ f v-}

simd.ll

; (snip)
c1o6:
  %ln1s2 = load i64* %ls1nc
  %ln1s3 = add i64 %ln1s2, 7
  %ln1s4 = inttoptr i64 %ln1s3 to i64*
  %ln1s5 = load i64* %ln1s4, !tbaa !5
  %ln1s6 = trunc i64 %ln1s5 to i32
  %ln1s7 = insertelement <4 x i32> < i32 0, i32 0, i32 0, i32 0 >, i32 %ln1s6, i32 0
  %ln1s8 = insertelement <4 x i32> %ln1s7, i32 1, i32 1
  %ln1s9 = insertelement <4 x i32> %ln1s8, i32 2, i32 2
  %ln1sa = insertelement <4 x i32> %ln1s9, i32 -559038737, i32 3
  %ln1sb = bitcast <4 x i32> %ln1sa to <4 x i32>
  store <4 x i32> %ln1sb, <4 x i32>* %ls1mP, align 1
  %ln1sc = ptrtoint [0 x i64]* @base_GHCziInt_I32zh_con_info to i64
  %ln1sd = load i64** %Hp_Var
  %ln1se = getelementptr inbounds i64* %ln1sd, i32 -1
  store i64 %ln1sc, i64* %ln1se, !tbaa !2
  %ln1sf = load <4 x i32>* %ls1mP, align 1
  %ln1sg = load <4 x i32>* %ls1mP, align 1
  %ln1sh = add <4 x i32> %ln1sf, %ln1sg
  %ln1si = bitcast <4 x i32> %ln1sh to <4 x i32>
  store <4 x i32> %ln1si, <4 x i32>* %ls1na, align 1
  %ln1sj = load <4 x i32>* %ls1na, align 1
  %ln1sk = extractelement <4 x i32> %ln1sj, i32 0
  %ln1sl = sext i32 %ln1sk to i64
  %ln1sm = load <4 x i32>* %ls1na, align 1
  %ln1sn = extractelement <4 x i32> %ln1sm, i32 1
  %ln1so = sext i32 %ln1sn to i64
  %ln1sp = load <4 x i32>* %ls1na, align 1
  %ln1sq = extractelement <4 x i32> %ln1sp, i32 2
  %ln1sr = sext i32 %ln1sq to i64
  %ln1ss = load <4 x i32>* %ls1na, align 1
  %ln1st = extractelement <4 x i32> %ln1ss, i32 3
  %ln1su = sext i32 %ln1st to i64
  %ln1sv = add i64 %ln1sr, %ln1su
  %ln1sw = add i64 %ln1so, %ln1sv
  %ln1sx = add i64 %ln1sl, %ln1sw
  %ln1sy = load i64** %Hp_Var
  %ln1sz = getelementptr inbounds i64* %ln1sy, i32 0
; (snip)

simd.s

# (snip)
.LBB0_6:                                # %c1o6.i
	movd	7(%r14), %xmm0
	movl	$1, %ecx
	pinsrd	$1, %ecx, %xmm0
	movl	$2, %ecx
	pinsrd	$2, %ecx, %xmm0
	movl	$-559038737, %ecx       # imm = 0xFFFFFFFFDEADBEEF
	pinsrd	$3, %ecx, %xmm0
	paddd	%xmm0, %xmm0
	movdqa	%xmm0, -24(%rsp)
	movq	$base_GHCziInt_I32zh_con_info, 8(%r12)
	leaq	-7(%rax), %rbx
	movslq	-16(%rsp), %rdx
	movslq	-12(%rsp), %rcx
	addq	%rdx, %rcx
	movslq	-24(%rsp), %rdx
	addq	%rcx, %rdx
	movslq	-20(%rsp), %rcx
	addq	%rdx, %rcx
	movq	%rcx, (%rax)
	movq	(%rbp), %rcx
	movq	%rax, %r12
	jmpq	*%rcx  # TAILCALL