1 ; RUN: llc < %s -mtriple=i386-pc-linux-gnu -mattr=+sse2
2 ; RUN: llc < %s -mtriple=i386-pc-linux-gnu -mattr=+sse2 -relocation-model=pic -frame-pointer=all
5 define void @ff_flac_compute_autocorr_sse2(ptr %data, i32 %len, i32 %lag, ptr %autoc) nounwind {
7 %c = alloca double, align 8 ; <ptr> [#uses=2]
8 %0 = add i32 %len, 2 ; <i32> [#uses=1]
9 %1 = add i32 %0, %lag ; <i32> [#uses=1]
10 %2 = alloca double, i32 %1 ; <ptr> [#uses=2]
11 %3 = getelementptr double, ptr %2, i32 %lag ; <ptr> [#uses=2]
12 %4 = ptrtoint ptr %3 to i32 ; <i32> [#uses=1]
13 %5 = and i32 %4, 8 ; <i32> [#uses=1]
14 %6 = icmp eq i32 %5, 0 ; <i1> [#uses=1]
15 br i1 %6, label %bb19, label %bb
18 %.sum = add i32 %lag, 1 ; <i32> [#uses=1]
19 %7 = getelementptr double, ptr %2, i32 %.sum ; <ptr> [#uses=1]
22 bb19: ; preds = %bb, %entry
23 %data15.0 = phi ptr [ %7, %bb ], [ %3, %entry ] ; <ptr> [#uses=5]
24 %8 = sitofp i32 %len to double ; <double> [#uses=1]
25 %9 = fsub double %8, 1.000000e+00 ; <double> [#uses=1]
26 %10 = fdiv double 2.000000e+00, %9 ; <double> [#uses=1]
27 store double %10, ptr %c, align 8
28 %11 = ashr i32 %len, 1 ; <i32> [#uses=3]
29 %12 = mul i32 %11, -4 ; <i32> [#uses=2]
30 %13 = shl i32 %len, 1 ; <i32> [#uses=1]
31 %14 = and i32 %13, -4 ; <i32> [#uses=2]
32 call void asm sideeffect "movsd $0, %xmm7 \0A\09movapd ff_pd_1, %xmm6 \0A\09movapd ff_pd_2, %xmm5 \0A\09movlhps %xmm7, %xmm7 \0A\09subpd %xmm5, %xmm7 \0A\09addsd %xmm6, %xmm7 \0A\09", "*m,~{dirflag},~{fpsr},~{flags}"(ptr elementtype(double) %c) nounwind
33 %15 = and i32 %len, 1 ; <i32> [#uses=1]
34 %toBool = icmp eq i32 %15, 0 ; <i1> [#uses=1]
35 %16 = getelementptr double, ptr %data15.0, i32 %11 ; <ptr> [#uses=2]
36 %17 = getelementptr i32, ptr %data, i32 %11 ; <ptr> [#uses=2]
37 br i1 %toBool, label %bb22, label %bb20
40 %asmtmp = call { i32, i32 } asm sideeffect "1: \0A\09movapd %xmm7, %xmm1 \0A\09mulpd %xmm1, %xmm1 \0A\09movapd %xmm6, %xmm0 \0A\09subpd %xmm1, %xmm0 \0A\09pshufd $$0x4e, %xmm0, %xmm1 \0A\09cvtpi2pd ($3,$0), %xmm2 \0A\09cvtpi2pd -1*4($3,$1), %xmm3 \0A\09mulpd %xmm0, %xmm2 \0A\09mulpd %xmm1, %xmm3 \0A\09movapd %xmm2, ($2,$0,2) \0A\09movupd %xmm3, -1*8($2,$1,2) \0A\09subpd %xmm5, %xmm7 \0A\09sub $$8, $1 \0A\09add $$8, $0 \0A\09jl 1b \0A\09", "=&r,=&r,r,r,0,1,~{dirflag},~{fpsr},~{flags}"(ptr %16, ptr %17, i32 %12, i32 %14) nounwind ; <{ i32, i32 }> [#uses=0]
41 br label %bb28.preheader
44 %asmtmp23 = call { i32, i32 } asm sideeffect "1: \0A\09movapd %xmm7, %xmm1 \0A\09mulpd %xmm1, %xmm1 \0A\09movapd %xmm6, %xmm0 \0A\09subpd %xmm1, %xmm0 \0A\09pshufd $$0x4e, %xmm0, %xmm1 \0A\09cvtpi2pd ($3,$0), %xmm2 \0A\09cvtpi2pd -2*4($3,$1), %xmm3 \0A\09mulpd %xmm0, %xmm2 \0A\09mulpd %xmm1, %xmm3 \0A\09movapd %xmm2, ($2,$0,2) \0A\09movapd %xmm3, -2*8($2,$1,2) \0A\09subpd %xmm5, %xmm7 \0A\09sub $$8, $1 \0A\09add $$8, $0 \0A\09jl 1b \0A\09", "=&r,=&r,r,r,0,1,~{dirflag},~{fpsr},~{flags}"(ptr %16, ptr %17, i32 %12, i32 %14) nounwind ; <{ i32, i32 }> [#uses=0]
45 br label %bb28.preheader
47 bb28.preheader: ; preds = %bb22, %bb20
48 %18 = icmp sgt i32 %lag, 0 ; <i1> [#uses=2]
49 br i1 %18, label %bb27, label %bb29
51 bb27: ; preds = %bb27, %bb28.preheader
52 %j4.042 = phi i32 [ 0, %bb28.preheader ], [ %indvar.next45, %bb27 ] ; <i32> [#uses=2]
53 %19 = sub i32 %j4.042, %lag ; <i32> [#uses=1]
54 %20 = getelementptr double, ptr %data15.0, i32 %19 ; <ptr> [#uses=1]
55 store double 0.000000e+00, ptr %20, align 8
56 %indvar.next45 = add i32 %j4.042, 1 ; <i32> [#uses=2]
57 %exitcond = icmp eq i32 %indvar.next45, %lag ; <i1> [#uses=1]
58 br i1 %exitcond, label %bb29, label %bb27
60 bb29: ; preds = %bb27, %bb28.preheader
61 %21 = getelementptr double, ptr %data15.0, i32 %len ; <ptr> [#uses=3]
62 store double 0.000000e+00, ptr %21, align 8
63 br i1 %18, label %bb.nph, label %bb37
65 bb.nph: ; preds = %bb29
66 %22 = mul i32 %len, -8 ; <i32> [#uses=2]
67 %23 = add i32 %lag, -2 ; <i32> [#uses=1]
70 bb30: ; preds = %bb35, %bb.nph
71 %indvar = phi i32 [ 0, %bb.nph ], [ %indvar.next, %bb35 ] ; <i32> [#uses=2]
72 %j4.141 = shl i32 %indvar, 1 ; <i32> [#uses=8]
73 %24 = icmp eq i32 %23, %j4.141 ; <i1> [#uses=1]
74 %25 = or i32 %j4.141, 1 ; <i32> [#uses=2]
75 br i1 %24, label %bb31, label %bb33
78 %26 = add i32 %j4.141, 2 ; <i32> [#uses=2]
79 %.sum38 = sub i32 %len, %j4.141 ; <i32> [#uses=1]
80 %27 = getelementptr double, ptr %data15.0, i32 %.sum38 ; <ptr> [#uses=1]
81 %28 = getelementptr double, ptr %autoc, i32 %j4.141 ; <ptr> [#uses=1]
82 %29 = getelementptr double, ptr %autoc, i32 %25 ; <ptr> [#uses=1]
83 %30 = getelementptr double, ptr %autoc, i32 %26 ; <ptr> [#uses=1]
84 %asmtmp32 = call i32 asm sideeffect "movsd ff_pd_1, %xmm0 \0A\09movsd ff_pd_1, %xmm1 \0A\09movsd ff_pd_1, %xmm2 \0A\091: \0A\09movapd ($4,$0), %xmm3 \0A\09movupd -8($5,$0), %xmm4 \0A\09movapd ($5,$0), %xmm5 \0A\09mulpd %xmm3, %xmm4 \0A\09mulpd %xmm3, %xmm5 \0A\09mulpd -16($5,$0), %xmm3 \0A\09addpd %xmm4, %xmm1 \0A\09addpd %xmm5, %xmm0 \0A\09addpd %xmm3, %xmm2 \0A\09add $$16, $0 \0A\09jl 1b \0A\09movhlps %xmm0, %xmm3 \0A\09movhlps %xmm1, %xmm4 \0A\09movhlps %xmm2, %xmm5 \0A\09addsd %xmm3, %xmm0 \0A\09addsd %xmm4, %xmm1 \0A\09addsd %xmm5, %xmm2 \0A\09movsd %xmm0, $1 \0A\09movsd %xmm1, $2 \0A\09movsd %xmm2, $3 \0A\09", "=&r,=*m,=*m,=*m,r,r,0,~{dirflag},~{fpsr},~{flags}"(ptr elementtype(double) %28, ptr elementtype(double) %29, ptr elementtype(double) %30, ptr %21, ptr %27, i32 %22) nounwind ; <i32> [#uses=0]
88 %.sum39 = sub i32 %len, %j4.141 ; <i32> [#uses=1]
89 %31 = getelementptr double, ptr %data15.0, i32 %.sum39 ; <ptr> [#uses=1]
90 %32 = getelementptr double, ptr %autoc, i32 %j4.141 ; <ptr> [#uses=1]
91 %33 = getelementptr double, ptr %autoc, i32 %25 ; <ptr> [#uses=1]
92 %asmtmp34 = call i32 asm sideeffect "movsd ff_pd_1, %xmm0 \0A\09movsd ff_pd_1, %xmm1 \0A\091: \0A\09movapd ($3,$0), %xmm3 \0A\09movupd -8($4,$0), %xmm4 \0A\09mulpd %xmm3, %xmm4 \0A\09mulpd ($4,$0), %xmm3 \0A\09addpd %xmm4, %xmm1 \0A\09addpd %xmm3, %xmm0 \0A\09add $$16, $0 \0A\09jl 1b \0A\09movhlps %xmm0, %xmm3 \0A\09movhlps %xmm1, %xmm4 \0A\09addsd %xmm3, %xmm0 \0A\09addsd %xmm4, %xmm1 \0A\09movsd %xmm0, $1 \0A\09movsd %xmm1, $2 \0A\09", "=&r,=*m,=*m,r,r,0,~{dirflag},~{fpsr},~{flags}"(ptr elementtype(double) %32, ptr elementtype(double) %33, ptr %21, ptr %31, i32 %22) nounwind ; <i32> [#uses=0]
93 %.pre = add i32 %j4.141, 2 ; <i32> [#uses=1]
96 bb35: ; preds = %bb33, %bb31
97 %.pre-phi = phi i32 [ %.pre, %bb33 ], [ %26, %bb31 ] ; <i32> [#uses=1]
98 %34 = icmp slt i32 %.pre-phi, %lag ; <i1> [#uses=1]
99 %indvar.next = add i32 %indvar, 1 ; <i32> [#uses=1]
100 br i1 %34, label %bb30, label %bb37
102 bb37: ; preds = %bb35, %bb29