diff --git a/src/hexagon.rs b/src/hexagon.rs
new file mode 100644
index 00000000..91cf91c3
--- /dev/null
+++ b/src/hexagon.rs
@@ -0,0 +1,55 @@
+#![cfg(not(feature = "no-asm"))]
+
+use core::arch::global_asm;
+
+global_asm!(include_str!("hexagon/func_macro.s"), options(raw));
+
+global_asm!(include_str!("hexagon/dfaddsub.s"), options(raw));
+
+global_asm!(include_str!("hexagon/dfdiv.s"), options(raw));
+
+global_asm!(include_str!("hexagon/dffma.s"), options(raw));
+
+global_asm!(include_str!("hexagon/dfminmax.s"), options(raw));
+
+global_asm!(include_str!("hexagon/dfmul.s"), options(raw));
+
+global_asm!(include_str!("hexagon/dfsqrt.s"), options(raw));
+
+global_asm!(include_str!("hexagon/divdi3.s"), options(raw));
+
+global_asm!(include_str!("hexagon/divsi3.s"), options(raw));
+
+global_asm!(include_str!("hexagon/fastmath2_dlib_asm.s"), options(raw));
+
+global_asm!(include_str!("hexagon/fastmath2_ldlib_asm.s"), options(raw));
+
+global_asm!(
+    include_str!("hexagon/memcpy_forward_vp4cp4n2.s"),
+    options(raw)
+);
+
+global_asm!(
+    include_str!("hexagon/memcpy_likely_aligned.s"),
+    options(raw)
+);
+
+global_asm!(include_str!("hexagon/moddi3.s"), options(raw));
+
+global_asm!(include_str!("hexagon/modsi3.s"), options(raw));
+
+global_asm!(include_str!("hexagon/sfdiv_opt.s"), options(raw));
+
+global_asm!(include_str!("hexagon/sfsqrt_opt.s"), options(raw));
+
+global_asm!(include_str!("hexagon/udivdi3.s"), options(raw));
+
+global_asm!(include_str!("hexagon/udivmoddi4.s"), options(raw));
+
+global_asm!(include_str!("hexagon/udivmodsi4.s"), options(raw));
+
+global_asm!(include_str!("hexagon/udivsi3.s"), options(raw));
+
+global_asm!(include_str!("hexagon/umoddi3.s"), options(raw));
+
+global_asm!(include_str!("hexagon/umodsi3.s"), options(raw));
diff --git a/src/hexagon/dfaddsub.s b/src/hexagon/dfaddsub.s
new file mode 100644
index 00000000..1f59e460
--- /dev/null
+++ b/src/hexagon/dfaddsub.s
@@ -0,0 +1,321 @@
+ .text
+ .global __hexagon_adddf3
+ .global __hexagon_subdf3
+ .type __hexagon_adddf3, @function
+ .type __hexagon_subdf3, @function
+
+.global __qdsp_adddf3 ; .set __qdsp_adddf3, __hexagon_adddf3
+.global __hexagon_fast_adddf3 ; .set __hexagon_fast_adddf3, __hexagon_adddf3
+.global __hexagon_fast2_adddf3 ; .set __hexagon_fast2_adddf3, __hexagon_adddf3
+.global __qdsp_subdf3 ; .set __qdsp_subdf3, __hexagon_subdf3
+.global __hexagon_fast_subdf3 ; .set __hexagon_fast_subdf3, __hexagon_subdf3
+.global __hexagon_fast2_subdf3 ; .set __hexagon_fast2_subdf3, __hexagon_subdf3
+
+ .p2align 5
+__hexagon_adddf3:
+ {
+  r4 = extractu(r1,#11,#20)
+  r5 = extractu(r3,#11,#20)
+  r13:12 = combine(##0x20000000,#0)
+ }
+ {
+  p3 = dfclass(r1:0,#2)
+  p3 = dfclass(r3:2,#2)
+  r9:8 = r13:12
+  p2 = cmp.gtu(r5,r4)
+ }
+ {
+  if (!p3) jump .Ladd_abnormal
+  if (p2) r1:0 = r3:2
+  if (p2) r3:2 = r1:0
+  if (p2) r5:4 = combine(r4,r5)
+ }
+ {
+  r13:12 = insert(r1:0,#52,#11 -2)
+  r9:8 = insert(r3:2,#52,#11 -2)
+  r15 = sub(r4,r5)
+  r7:6 = combine(#62,#1)
+ }
+
+
+
+
+
+.Ladd_continue:
+ {
+  r15 = min(r15,r7)
+
+  r11:10 = neg(r13:12)
+  p2 = cmp.gt(r1,#-1)
+  r14 = #0
+ }
+ {
+  if (!p2) r13:12 = r11:10
+  r11:10 = extractu(r9:8,r15:14)
+  r9:8 = ASR(r9:8,r15)
+
+
+
+
+  r15:14 = #0
+ }
+ {
+  p1 = cmp.eq(r11:10,r15:14)
+  if (!p1.new) r8 = or(r8,r6)
+  r5 = add(r4,#-1024 -60)
+  p3 = cmp.gt(r3,#-1)
+ }
+ {
+  r13:12 = add(r13:12,r9:8)
+  r11:10 = sub(r13:12,r9:8)
+  r7:6 = combine(#54,##2045)
+ }
+ {
+  p0 = cmp.gtu(r4,r7)
+  p0 = !cmp.gtu(r4,r6)
+  if (!p0.new) jump:nt .Ladd_ovf_unf
+  if (!p3) r13:12 = r11:10
+ }
+ {
+  r1:0 = convert_d2df(r13:12)
+  p0 = cmp.eq(r13,#0)
+  p0 = cmp.eq(r12,#0)
+  if (p0.new) jump:nt .Ladd_zero
+ }
+ {
+  r1 += asl(r5,#20)
+  jumpr r31
+ }
+ .falign
+__hexagon_subdf3:
+ {
+  r3 = togglebit(r3,#31)
+  jump __qdsp_adddf3
+ }
+
+
+ .falign
+.Ladd_zero:
+
+
+ {
+  r28 = USR
+  r1:0 = #0
+  r3 = #1
+ }
+ {
+  r28 = extractu(r28,#2,#22)
+  r3 = asl(r3,#31)
+ }
+ {
+  p0 = cmp.eq(r28,#2)
+  if (p0.new) r1 = xor(r1,r3)
+  jumpr r31
+ }
+ .falign
+.Ladd_ovf_unf:
+ {
+  r1:0 = convert_d2df(r13:12)
+  p0 = cmp.eq(r13,#0)
+  p0 = cmp.eq(r12,#0)
+  if (p0.new) jump:nt .Ladd_zero
+ }
+ {
+  r28 = extractu(r1,#11,#20)
+  r1 += asl(r5,#20)
+ }
+ {
+  r5 = add(r5,r28)
+  r3:2 = combine(##0x00100000,#0)
+ }
+ {
+  p0 = cmp.gt(r5,##1024 +1024 -2)
+  if (p0.new) jump:nt .Ladd_ovf
+ }
+ {
+  p0 = cmp.gt(r5,#0)
+  if (p0.new) jumpr:t r31
+  r28 = sub(#1,r5)
+ }
+ {
+  r3:2 = insert(r1:0,#52,#0)
+  r1:0 = r13:12
+ }
+ {
+  r3:2 = lsr(r3:2,r28)
+ }
+ {
+  r1:0 = insert(r3:2,#63,#0)
+  jumpr r31
+ }
+ .falign
+.Ladd_ovf:
+
+ {
+  r1:0 = r13:12
+  r28 = USR
+  r13:12 = combine(##0x7fefffff,#-1)
+ }
+ {
+  r5 = extractu(r28,#2,#22)
+  r28 = or(r28,#0x28)
+  r9:8 = combine(##0x7ff00000,#0)
+ }
+ {
+  USR = r28
+  r5 ^= lsr(r1,#31)
+  r28 = r5
+ }
+ {
+  p0 = !cmp.eq(r28,#1)
+  p0 = !cmp.eq(r5,#2)
+  if (p0.new) r13:12 = r9:8
+ }
+ {
+  r1:0 = insert(r13:12,#63,#0)
+ }
+ {
+  p0 = dfcmp.eq(r1:0,r1:0)
+  jumpr r31
+ }
+
+.Ladd_abnormal:
+ {
+  r13:12 = extractu(r1:0,#63,#0)
+  r9:8 = extractu(r3:2,#63,#0)
+ }
+ {
+  p3 = cmp.gtu(r13:12,r9:8)
+  if (!p3.new) r1:0 = r3:2
+  if (!p3.new) r3:2 = r1:0
+ }
+ {
+
+  p0 = dfclass(r1:0,#0x0f)
+  if (!p0.new) jump:nt .Linvalid_nan_add
+  if (!p3) r13:12 = r9:8
+  if (!p3) r9:8 = r13:12
+ }
+ {
+
+
+  p1 = dfclass(r1:0,#0x08)
+  if (p1.new) jump:nt .Linf_add
+ }
+ {
+  p2 = dfclass(r3:2,#0x01)
+  if (p2.new) jump:nt .LB_zero
+  r13:12 = #0
+ }
+
+ {
+  p0 = dfclass(r1:0,#4)
+  if (p0.new) jump:nt .Ladd_two_subnormal
+  r13:12 = combine(##0x20000000,#0)
+ }
+ {
+  r4 = extractu(r1,#11,#20)
+  r5 = #1
+
+  r9:8 = asl(r9:8,#11 -2)
+ }
+
+
+
+ {
+  r13:12 = insert(r1:0,#52,#11 -2)
+  r15 = sub(r4,r5)
+  r7:6 = combine(#62,#1)
+  jump .Ladd_continue
+ }
+
+.Ladd_two_subnormal:
+ {
+  r13:12 = extractu(r1:0,#63,#0)
+  r9:8 = extractu(r3:2,#63,#0)
+ }
+ {
+  r13:12 = neg(r13:12)
+  r9:8 = neg(r9:8)
+  p0 = cmp.gt(r1,#-1)
+  p1 = cmp.gt(r3,#-1)
+ }
+ {
+  if (p0) r13:12 = r1:0
+  if (p1) r9:8 = r3:2
+ }
+ {
+  r13:12 = add(r13:12,r9:8)
+ }
+ {
+  r9:8 = neg(r13:12)
+  p0 = cmp.gt(r13,#-1)
+  r3:2 = #0
+ }
+ {
+  if (!p0) r1:0 = r9:8
+  if (p0) r1:0 = r13:12
+  r3 = ##0x80000000
+ }
+ {
+  if (!p0) r1 = or(r1,r3)
+  p0 = dfcmp.eq(r1:0,r3:2)
+  if (p0.new) jump:nt .Lzero_plus_zero
+ }
+ {
+  jumpr r31
+ }
+
+.Linvalid_nan_add:
+ {
+  r28 = convert_df2sf(r1:0)
+  p0 = dfclass(r3:2,#0x0f)
+  if (p0.new) r3:2 = r1:0
+ }
+ {
+  r2 = convert_df2sf(r3:2)
+  r1:0 = #-1
+  jumpr r31
+ }
+ .falign
+.LB_zero:
+ {
+  p0 = dfcmp.eq(r13:12,r1:0)
+  if (!p0.new) jumpr:t r31
+ }
+
+
+
+
+.Lzero_plus_zero:
+ {
+  p0 = cmp.eq(r1:0,r3:2)
+  if (p0.new) jumpr:t r31
+ }
+ {
+  r28 = USR
+ }
+ {
+  r28 = extractu(r28,#2,#22)
+  r1:0 = #0
+ }
+ {
+  p0 = cmp.eq(r28,#2)
+  if (p0.new) r1 = ##0x80000000
+  jumpr r31
+ }
+.Linf_add:
+
+ {
+  p0 = !cmp.eq(r1,r3)
+  p0 = dfclass(r3:2,#8)
+  if (!p0.new) jumpr:t r31
+ }
+ {
+  r2 = ##0x7f800001
+ }
+ {
+  r1:0 = convert_sf2df(r2)
+  jumpr r31
+ }
+.size __hexagon_adddf3,.-__hexagon_adddf3
diff --git a/src/hexagon/dfdiv.s b/src/hexagon/dfdiv.s
new file mode 100644
index 00000000..6d65dbfc
--- /dev/null
+++ b/src/hexagon/dfdiv.s
@@ -0,0 +1,372 @@
+ .text
+ .global __hexagon_divdf3
+ .type __hexagon_divdf3,@function
+ .global __qdsp_divdf3 ; .set __qdsp_divdf3, __hexagon_divdf3
+        .global __hexagon_fast_divdf3 ; .set __hexagon_fast_divdf3, __hexagon_divdf3
+        .global __hexagon_fast2_divdf3 ; .set __hexagon_fast2_divdf3, __hexagon_divdf3
+ .p2align 5
+__hexagon_divdf3:
+ {
+  p2 = dfclass(r1:0,#0x02)
+  p2 = dfclass(r3:2,#0x02)
+  r13:12 = combine(r3,r1)
+  r28 = xor(r1,r3)
+ }
+ {
+  if (!p2) jump .Ldiv_abnormal
+  r7:6 = extractu(r3:2,#23,#52 -23)
+  r8 = ##0x3f800001
+ }
+ {
+  r9 = or(r8,r6)
+  r13 = extractu(r13,#11,#52 -32)
+  r12 = extractu(r12,#11,#52 -32)
+  p3 = cmp.gt(r28,#-1)
+ }
+
+
+.Ldenorm_continue:
+ {
+  r11,p0 = sfrecipa(r8,r9)
+  r10 = and(r8,#-2)
+  r28 = #1
+  r12 = sub(r12,r13)
+ }
+
+
+ {
+  r10 -= sfmpy(r11,r9):lib
+  r1 = insert(r28,#11 +1,#52 -32)
+  r13 = ##0x00800000 << 3
+ }
+ {
+  r11 += sfmpy(r11,r10):lib
+  r3 = insert(r28,#11 +1,#52 -32)
+  r10 = and(r8,#-2)
+ }
+ {
+  r10 -= sfmpy(r11,r9):lib
+  r5 = #-0x3ff +1
+  r4 = #0x3ff -1
+ }
+ {
+  r11 += sfmpy(r11,r10):lib
+  p1 = cmp.gt(r12,r5)
+  p1 = !cmp.gt(r12,r4)
+ }
+ {
+  r13 = insert(r11,#23,#3)
+  r5:4 = #0
+  r12 = add(r12,#-61)
+ }
+
+
+
+
+ {
+  r13 = add(r13,#((-3) << 3))
+ }
+ { r7:6 = mpyu(r13,r1); r1:0 = asl(r1:0,# ( 15 )); }; { r6 = # 0; r1:0 -= mpyu(r7,r2); r15:14 = mpyu(r7,r3); }; { r5:4 += ASL(r7:6, # ( 14 )); r1:0 -= asl(r15:14, # 32); }
+ { r7:6 = mpyu(r13,r1); r1:0 = asl(r1:0,# ( 15 )); }; { r6 = # 0; r1:0 -= mpyu(r7,r2); r15:14 = mpyu(r7,r3); }; { r5:4 += ASR(r7:6, # ( 1 )); r1:0 -= asl(r15:14, # 32); }
+ { r7:6 = mpyu(r13,r1); r1:0 = asl(r1:0,# ( 15 )); }; { r6 = # 0; r1:0 -= mpyu(r7,r2); r15:14 = mpyu(r7,r3); }; { r5:4 += ASR(r7:6, # ( 16 )); r1:0 -= asl(r15:14, # 32); }
+ { r7:6 = mpyu(r13,r1); r1:0 = asl(r1:0,# ( 15 )); }; { r6 = # 0; r1:0 -= mpyu(r7,r2); r15:14 = mpyu(r7,r3); }; { r5:4 += ASR(r7:6, # ( 31 )); r1:0 -= asl(r15:14, # 32); r7:6=# ( 0 ); }
+
+
+
+
+
+
+
+ {
+
+  r15:14 = sub(r1:0,r3:2)
+  p0 = cmp.gtu(r3:2,r1:0)
+
+  if (!p0.new) r6 = #2
+ }
+ {
+  r5:4 = add(r5:4,r7:6)
+  if (!p0) r1:0 = r15:14
+  r15:14 = #0
+ }
+ {
+  p0 = cmp.eq(r1:0,r15:14)
+  if (!p0.new) r4 = or(r4,r28)
+ }
+ {
+  r7:6 = neg(r5:4)
+ }
+ {
+  if (!p3) r5:4 = r7:6
+ }
+ {
+  r1:0 = convert_d2df(r5:4)
+  if (!p1) jump .Ldiv_ovf_unf
+ }
+ {
+  r1 += asl(r12,#52 -32)
+  jumpr r31
+ }
+
+.Ldiv_ovf_unf:
+ {
+  r1 += asl(r12,#52 -32)
+  r13 = extractu(r1,#11,#52 -32)
+ }
+ {
+  r7:6 = abs(r5:4)
+  r12 = add(r12,r13)
+ }
+ {
+  p0 = cmp.gt(r12,##0x3ff +0x3ff)
+  if (p0.new) jump:nt .Ldiv_ovf
+ }
+ {
+  p0 = cmp.gt(r12,#0)
+  if (p0.new) jump:nt .Lpossible_unf2
+ }
+ {
+  r13 = add(clb(r7:6),#-1)
+  r12 = sub(#7,r12)
+  r10 = USR
+  r11 = #63
+ }
+ {
+  r13 = min(r12,r11)
+  r11 = or(r10,#0x030)
+  r7:6 = asl(r7:6,r13)
+  r12 = #0
+ }
+ {
+  r15:14 = extractu(r7:6,r13:12)
+  r7:6 = lsr(r7:6,r13)
+  r3:2 = #1
+ }
+ {
+  p0 = cmp.gtu(r3:2,r15:14)
+  if (!p0.new) r6 = or(r2,r6)
+  r7 = setbit(r7,#52 -32+4)
+ }
+ {
+  r5:4 = neg(r7:6)
+  p0 = bitsclr(r6,#(1<<4)-1)
+  if (!p0.new) r10 = r11
+ }
+ {
+  USR = r10
+  if (p3) r5:4 = r7:6
+  r10 = #-0x3ff -(52 +4)
+ }
+ {
+  r1:0 = convert_d2df(r5:4)
+ }
+ {
+  r1 += asl(r10,#52 -32)
+  jumpr r31
+ }
+
+
+.Lpossible_unf2:
+
+
+ {
+  r3:2 = extractu(r1:0,#63,#0)
+  r15:14 = combine(##0x00100000,#0)
+  r10 = #0x7FFF
+ }
+ {
+  p0 = dfcmp.eq(r15:14,r3:2)
+  p0 = bitsset(r7,r10)
+ }
+
+
+
+
+
+
+ {
+  if (!p0) jumpr r31
+  r10 = USR
+ }
+
+ {
+  r10 = or(r10,#0x30)
+ }
+ {
+  USR = r10
+ }
+ {
+  p0 = dfcmp.eq(r1:0,r1:0)
+  jumpr r31
+ }
+
+.Ldiv_ovf:
+
+
+
+ {
+  r10 = USR
+  r3:2 = combine(##0x7fefffff,#-1)
+  r1 = mux(p3,#0,#-1)
+ }
+ {
+  r7:6 = combine(##0x7ff00000,#0)
+  r5 = extractu(r10,#2,#22)
+  r10 = or(r10,#0x28)
+ }
+ {
+  USR = r10
+  r5 ^= lsr(r1,#31)
+  r4 = r5
+ }
+ {
+  p0 = !cmp.eq(r4,#1)
+  p0 = !cmp.eq(r5,#2)
+  if (p0.new) r3:2 = r7:6
+  p0 = dfcmp.eq(r3:2,r3:2)
+ }
+ {
+  r1:0 = insert(r3:2,#63,#0)
+  jumpr r31
+ }
+
+
+
+
+
+
+
+.Ldiv_abnormal:
+ {
+  p0 = dfclass(r1:0,#0x0F)
+  p0 = dfclass(r3:2,#0x0F)
+  p3 = cmp.gt(r28,#-1)
+ }
+ {
+  p1 = dfclass(r1:0,#0x08)
+  p1 = dfclass(r3:2,#0x08)
+ }
+ {
+  p2 = dfclass(r1:0,#0x01)
+  p2 = dfclass(r3:2,#0x01)
+ }
+ {
+  if (!p0) jump .Ldiv_nan
+  if (p1) jump .Ldiv_invalid
+ }
+ {
+  if (p2) jump .Ldiv_invalid
+ }
+ {
+  p2 = dfclass(r1:0,#(0x0F ^ 0x01))
+  p2 = dfclass(r3:2,#(0x0F ^ 0x08))
+ }
+ {
+  p1 = dfclass(r1:0,#(0x0F ^ 0x08))
+  p1 = dfclass(r3:2,#(0x0F ^ 0x01))
+ }
+ {
+  if (!p2) jump .Ldiv_zero_result
+  if (!p1) jump .Ldiv_inf_result
+ }
+
+
+
+
+
+ {
+  p0 = dfclass(r1:0,#0x02)
+  p1 = dfclass(r3:2,#0x02)
+  r10 = ##0x00100000
+ }
+ {
+  r13:12 = combine(r3,r1)
+  r1 = insert(r10,#11 +1,#52 -32)
+  r3 = insert(r10,#11 +1,#52 -32)
+ }
+ {
+  if (p0) r1 = or(r1,r10)
+  if (p1) r3 = or(r3,r10)
+ }
+ {
+  r5 = add(clb(r1:0),#-11)
+  r4 = add(clb(r3:2),#-11)
+  r10 = #1
+ }
+ {
+  r12 = extractu(r12,#11,#52 -32)
+  r13 = extractu(r13,#11,#52 -32)
+ }
+ {
+  r1:0 = asl(r1:0,r5)
+  r3:2 = asl(r3:2,r4)
+  if (!p0) r12 = sub(r10,r5)
+  if (!p1) r13 = sub(r10,r4)
+ }
+ {
+  r7:6 = extractu(r3:2,#23,#52 -23)
+ }
+ {
+  r9 = or(r8,r6)
+  jump .Ldenorm_continue
+ }
+
+.Ldiv_zero_result:
+ {
+  r1 = xor(r1,r3)
+  r3:2 = #0
+ }
+ {
+  r1:0 = insert(r3:2,#63,#0)
+  jumpr r31
+ }
+.Ldiv_inf_result:
+ {
+  p2 = dfclass(r3:2,#0x01)
+  p2 = dfclass(r1:0,#(0x0F ^ 0x08))
+ }
+ {
+  r10 = USR
+  if (!p2) jump 1f
+  r1 = xor(r1,r3)
+ }
+ {
+  r10 = or(r10,#0x04)
+ }
+ {
+  USR = r10
+ }
+1:
+ {
+  r3:2 = combine(##0x7ff00000,#0)
+  p0 = dfcmp.uo(r3:2,r3:2)
+ }
+ {
+  r1:0 = insert(r3:2,#63,#0)
+  jumpr r31
+ }
+.Ldiv_nan:
+ {
+  p0 = dfclass(r1:0,#0x10)
+  p1 = dfclass(r3:2,#0x10)
+  if (!p0.new) r1:0 = r3:2
+  if (!p1.new) r3:2 = r1:0
+ }
+ {
+  r5 = convert_df2sf(r1:0)
+  r4 = convert_df2sf(r3:2)
+ }
+ {
+  r1:0 = #-1
+  jumpr r31
+ }
+
+.Ldiv_invalid:
+ {
+  r10 = ##0x7f800001
+ }
+ {
+  r1:0 = convert_sf2df(r10)
+  jumpr r31
+ }
+.size __hexagon_divdf3,.-__hexagon_divdf3
diff --git a/src/hexagon/dffma.s b/src/hexagon/dffma.s
new file mode 100644
index 00000000..043a1d29
--- /dev/null
+++ b/src/hexagon/dffma.s
@@ -0,0 +1,536 @@
+ .text
+ .global __hexagon_fmadf4
+        .type __hexagon_fmadf4,@function
+ .global __hexagon_fmadf5
+        .type __hexagon_fmadf5,@function
+ .global fma
+ .type fma,@function
+ .global __qdsp_fmadf5 ; .set __qdsp_fmadf5, __hexagon_fmadf5
+ .p2align 5
+__hexagon_fmadf4:
+__hexagon_fmadf5:
+fma:
+ {
+  p0 = dfclass(r1:0,#2)
+  p0 = dfclass(r3:2,#2)
+  r13:12 = #0
+  r15:14 = #0
+ }
+ {
+  r13:12 = insert(r1:0,#52,#11 -3)
+  r15:14 = insert(r3:2,#52,#11 -3)
+  r7 = ##0x10000000
+  allocframe(#32)
+ }
+ {
+  r9:8 = mpyu(r12,r14)
+  if (!p0) jump .Lfma_abnormal_ab
+  r13 = or(r13,r7)
+  r15 = or(r15,r7)
+ }
+ {
+  p0 = dfclass(r5:4,#2)
+  if (!p0.new) jump:nt .Lfma_abnormal_c
+  r11:10 = combine(r7,#0)
+  r7:6 = combine(#0,r9)
+ }
+.Lfma_abnormal_c_restart:
+ {
+  r7:6 += mpyu(r14,r13)
+  r11:10 = insert(r5:4,#52,#11 -3)
+  memd(r29+#0) = r17:16
+  memd(r29+#8) = r19:18
+ }
+ {
+  r7:6 += mpyu(r12,r15)
+  r19:18 = neg(r11:10)
+  p0 = cmp.gt(r5,#-1)
+  r28 = xor(r1,r3)
+ }
+ {
+  r18 = extractu(r1,#11,#20)
+  r19 = extractu(r3,#11,#20)
+  r17:16 = combine(#0,r7)
+  if (!p0) r11:10 = r19:18
+ }
+ {
+  r17:16 += mpyu(r13,r15)
+  r9:8 = combine(r6,r8)
+  r18 = add(r18,r19)
+
+
+
+
+  r19 = extractu(r5,#11,#20)
+ }
+ {
+  r18 = add(r18,#-1023 +(4))
+  p3 = !cmp.gt(r28,#-1)
+  r7:6 = #0
+  r15:14 = #0
+ }
+ {
+  r7:6 = sub(r7:6,r9:8,p3):carry
+  p0 = !cmp.gt(r28,#-1)
+  p1 = cmp.gt(r19,r18)
+  if (p1.new) r19:18 = combine(r18,r19)
+ }
+ {
+  r15:14 = sub(r15:14,r17:16,p3):carry
+  if (p0) r9:8 = r7:6
+
+
+
+
+  r7:6 = #0
+  r19 = sub(r18,r19)
+ }
+ {
+  if (p0) r17:16 = r15:14
+  p0 = cmp.gt(r19,#63)
+  if (p1) r9:8 = r7:6
+  if (p1) r7:6 = r9:8
+ }
+
+
+
+
+
+
+
+ {
+  if (p1) r17:16 = r11:10
+  if (p1) r11:10 = r17:16
+  if (p0) r19 = add(r19,#-64)
+  r28 = #63
+ }
+ {
+
+  if (p0) r7:6 = r11:10
+  r28 = asr(r11,#31)
+  r13 = min(r19,r28)
+  r12 = #0
+ }
+
+
+
+
+
+
+ {
+  if (p0) r11:10 = combine(r28,r28)
+  r5:4 = extract(r7:6,r13:12)
+  r7:6 = lsr(r7:6,r13)
+  r12 = sub(#64,r13)
+ }
+ {
+  r15:14 = #0
+  r28 = #-2
+  r7:6 |= lsl(r11:10,r12)
+  r11:10 = asr(r11:10,r13)
+ }
+ {
+  p3 = cmp.gtu(r5:4,r15:14)
+  if (p3.new) r6 = and(r6,r28)
+
+
+
+  r15:14 = #1
+  r5:4 = #0
+ }
+ {
+  r9:8 = add(r7:6,r9:8,p3):carry
+ }
+ {
+  r17:16 = add(r11:10,r17:16,p3):carry
+  r28 = #62
+ }
+
+
+
+
+
+
+
+ {
+  r12 = add(clb(r17:16),#-2)
+  if (!cmp.eq(r12.new,r28)) jump:t 1f
+ }
+
+ {
+  r11:10 = extractu(r9:8,#62,#2)
+  r9:8 = asl(r9:8,#62)
+  r18 = add(r18,#-62)
+ }
+ {
+  r17:16 = insert(r11:10,#62,#0)
+ }
+ {
+  r12 = add(clb(r17:16),#-2)
+ }
+ .falign
+1:
+ {
+  r11:10 = asl(r17:16,r12)
+  r5:4 |= asl(r9:8,r12)
+  r13 = sub(#64,r12)
+  r18 = sub(r18,r12)
+ }
+ {
+  r11:10 |= lsr(r9:8,r13)
+  p2 = cmp.gtu(r15:14,r5:4)
+  r28 = #1023 +1023 -2
+ }
+ {
+  if (!p2) r10 = or(r10,r14)
+
+  p0 = !cmp.gt(r18,r28)
+  p0 = cmp.gt(r18,#1)
+  if (!p0.new) jump:nt .Lfma_ovf_unf
+ }
+ {
+
+  p0 = cmp.gtu(r15:14,r11:10)
+  r1:0 = convert_d2df(r11:10)
+  r18 = add(r18,#-1023 -60)
+  r17:16 = memd(r29+#0)
+ }
+ {
+  r1 += asl(r18,#20)
+  r19:18 = memd(r29+#8)
+  if (!p0) dealloc_return
+ }
+.Ladd_yields_zero:
+
+ {
+  r28 = USR
+  r1:0 = #0
+ }
+ {
+  r28 = extractu(r28,#2,#22)
+  r17:16 = memd(r29+#0)
+  r19:18 = memd(r29+#8)
+ }
+ {
+  p0 = cmp.eq(r28,#2)
+  if (p0.new) r1 = ##0x80000000
+  dealloc_return
+ }
+.Lfma_ovf_unf:
+ {
+  p0 = cmp.gtu(r15:14,r11:10)
+  if (p0.new) jump:nt .Ladd_yields_zero
+ }
+ {
+  r1:0 = convert_d2df(r11:10)
+  r18 = add(r18,#-1023 -60)
+  r28 = r18
+ }
+
+
+ {
+  r1 += asl(r18,#20)
+  r7 = extractu(r1,#11,#20)
+ }
+ {
+  r6 = add(r18,r7)
+  r17:16 = memd(r29+#0)
+  r19:18 = memd(r29+#8)
+  r9:8 = abs(r11:10)
+ }
+ {
+  p0 = cmp.gt(r6,##1023 +1023)
+  if (p0.new) jump:nt .Lfma_ovf
+ }
+ {
+  p0 = cmp.gt(r6,#0)
+  if (p0.new) jump:nt .Lpossible_unf0
+ }
+ {
+
+
+
+  r7 = add(clb(r9:8),#-2)
+  r6 = sub(#1+5,r28)
+  p3 = cmp.gt(r11,#-1)
+ }
+
+
+
+ {
+  r6 = add(r6,r7)
+  r9:8 = asl(r9:8,r7)
+  r1 = USR
+  r28 = #63
+ }
+ {
+  r7 = min(r6,r28)
+  r6 = #0
+  r0 = #0x0030
+ }
+ {
+  r3:2 = extractu(r9:8,r7:6)
+  r9:8 = asr(r9:8,r7)
+ }
+ {
+  p0 = cmp.gtu(r15:14,r3:2)
+  if (!p0.new) r8 = or(r8,r14)
+  r9 = setbit(r9,#20 +3)
+ }
+ {
+  r11:10 = neg(r9:8)
+  p1 = bitsclr(r8,#(1<<3)-1)
+  if (!p1.new) r1 = or(r1,r0)
+  r3:2 = #0
+ }
+ {
+  if (p3) r11:10 = r9:8
+  USR = r1
+  r28 = #-1023 -(52 +3)
+ }
+ {
+  r1:0 = convert_d2df(r11:10)
+ }
+ {
+  r1 += asl(r28,#20)
+  dealloc_return
+ }
+.Lpossible_unf0:
+ {
+  r28 = ##0x7fefffff
+  r9:8 = abs(r11:10)
+ }
+ {
+  p0 = cmp.eq(r0,#0)
+  p0 = bitsclr(r1,r28)
+  if (!p0.new) dealloc_return:t
+  r28 = #0x7fff
+ }
+ {
+  p0 = bitsset(r9,r28)
+  r3 = USR
+  r2 = #0x0030
+ }
+ {
+  if (p0) r3 = or(r3,r2)
+ }
+ {
+  USR = r3
+ }
+ {
+  p0 = dfcmp.eq(r1:0,r1:0)
+  dealloc_return
+ }
+.Lfma_ovf:
+ {
+  r28 = USR
+  r11:10 = combine(##0x7fefffff,#-1)
+  r1:0 = r11:10
+ }
+ {
+  r9:8 = combine(##0x7ff00000,#0)
+  r3 = extractu(r28,#2,#22)
+  r28 = or(r28,#0x28)
+ }
+ {
+  USR = r28
+  r3 ^= lsr(r1,#31)
+  r2 = r3
+ }
+ {
+  p0 = !cmp.eq(r2,#1)
+  p0 = !cmp.eq(r3,#2)
+ }
+ {
+  p0 = dfcmp.eq(r9:8,r9:8)
+  if (p0.new) r11:10 = r9:8
+ }
+ {
+  r1:0 = insert(r11:10,#63,#0)
+  dealloc_return
+ }
+.Lfma_abnormal_ab:
+ {
+  r9:8 = extractu(r1:0,#63,#0)
+  r11:10 = extractu(r3:2,#63,#0)
+  deallocframe
+ }
+ {
+  p3 = cmp.gtu(r9:8,r11:10)
+  if (!p3.new) r1:0 = r3:2
+  if (!p3.new) r3:2 = r1:0
+ }
+ {
+  p0 = dfclass(r1:0,#0x0f)
+  if (!p0.new) jump:nt .Lnan
+  if (!p3) r9:8 = r11:10
+  if (!p3) r11:10 = r9:8
+ }
+ {
+  p1 = dfclass(r1:0,#0x08)
+  p1 = dfclass(r3:2,#0x0e)
+ }
+ {
+  p0 = dfclass(r1:0,#0x08)
+  p0 = dfclass(r3:2,#0x01)
+ }
+ {
+  if (p1) jump .Lab_inf
+  p2 = dfclass(r3:2,#0x01)
+ }
+ {
+  if (p0) jump .Linvalid
+  if (p2) jump .Lab_true_zero
+  r28 = ##0x7c000000
+ }
+
+
+
+
+
+ {
+  p0 = bitsclr(r1,r28)
+  if (p0.new) jump:nt .Lfma_ab_tiny
+ }
+ {
+  r28 = add(clb(r11:10),#-11)
+ }
+ {
+  r11:10 = asl(r11:10,r28)
+ }
+ {
+  r3:2 = insert(r11:10,#63,#0)
+  r1 -= asl(r28,#20)
+ }
+ jump fma
+
+.Lfma_ab_tiny:
+ r9:8 = combine(##0x00100000,#0)
+ {
+  r1:0 = insert(r9:8,#63,#0)
+  r3:2 = insert(r9:8,#63,#0)
+ }
+ jump fma
+
+.Lab_inf:
+ {
+  r3:2 = lsr(r3:2,#63)
+  p0 = dfclass(r5:4,#0x10)
+ }
+ {
+  r1:0 ^= asl(r3:2,#63)
+  if (p0) jump .Lnan
+ }
+ {
+  p1 = dfclass(r5:4,#0x08)
+  if (p1.new) jump:nt .Lfma_inf_plus_inf
+ }
+
+ {
+  jumpr r31
+ }
+ .falign
+.Lfma_inf_plus_inf:
+ {
+  p0 = dfcmp.eq(r1:0,r5:4)
+  if (!p0.new) jump:nt .Linvalid
+ }
+ {
+  jumpr r31
+ }
+
+.Lnan:
+ {
+  p0 = dfclass(r3:2,#0x10)
+  p1 = dfclass(r5:4,#0x10)
+  if (!p0.new) r3:2 = r1:0
+  if (!p1.new) r5:4 = r1:0
+ }
+ {
+  r3 = convert_df2sf(r3:2)
+  r2 = convert_df2sf(r5:4)
+ }
+ {
+  r3 = convert_df2sf(r1:0)
+  r1:0 = #-1
+  jumpr r31
+ }
+
+.Linvalid:
+ {
+  r28 = ##0x7f800001
+ }
+ {
+  r1:0 = convert_sf2df(r28)
+  jumpr r31
+ }
+
+.Lab_true_zero:
+
+ {
+  p0 = dfclass(r5:4,#0x10)
+  if (p0.new) jump:nt .Lnan
+  if (p0.new) r1:0 = r5:4
+ }
+ {
+  p0 = dfcmp.eq(r3:2,r5:4)
+  r1 = lsr(r1,#31)
+ }
+ {
+  r3 ^= asl(r1,#31)
+  if (!p0) r1:0 = r5:4
+  if (!p0) jumpr r31
+ }
+
+ {
+  p0 = cmp.eq(r3:2,r5:4)
+  if (p0.new) jumpr:t r31
+  r1:0 = r3:2
+ }
+ {
+  r28 = USR
+ }
+ {
+  r28 = extractu(r28,#2,#22)
+  r1:0 = #0
+ }
+ {
+  p0 = cmp.eq(r28,#2)
+  if (p0.new) r1 = ##0x80000000
+  jumpr r31
+ }
+
+
+
+
+ .falign
+.Lfma_abnormal_c:
+
+
+ {
+  p0 = dfclass(r5:4,#0x10)
+  if (p0.new) jump:nt .Lnan
+  if (p0.new) r1:0 = r5:4
+  deallocframe
+ }
+ {
+  p0 = dfclass(r5:4,#0x08)
+  if (p0.new) r1:0 = r5:4
+  if (p0.new) jumpr:nt r31
+ }
+
+
+ {
+  p0 = dfclass(r5:4,#0x01)
+  if (p0.new) jump:nt __hexagon_muldf3
+  r28 = #1
+ }
+
+
+ {
+  allocframe(#32)
+  r11:10 = #0
+  r5 = insert(r28,#11,#20)
+  jump .Lfma_abnormal_c_restart
+ }
+.size fma,.-fma
diff --git a/src/hexagon/dfminmax.s b/src/hexagon/dfminmax.s
new file mode 100644
index 00000000..3337a322
--- /dev/null
+++ b/src/hexagon/dfminmax.s
@@ -0,0 +1,51 @@
+ .text
+ .global __hexagon_mindf3
+ .global __hexagon_maxdf3
+ .global fmin
+ .type fmin,@function
+ .global fmax
+ .type fmax,@function
+ .type __hexagon_mindf3,@function
+ .type __hexagon_maxdf3,@function
+ .global __qdsp_mindf3 ; .set __qdsp_mindf3, __hexagon_mindf3
+ .global __qdsp_maxdf3 ; .set __qdsp_maxdf3, __hexagon_maxdf3
+ .p2align 5
+__hexagon_mindf3:
+fmin:
+ {
+  p0 = dfclass(r1:0,#0x10)
+  p1 = dfcmp.gt(r1:0,r3:2)
+  r5:4 = r1:0
+ }
+ {
+  if (p0) r1:0 = r3:2
+  if (p1) r1:0 = r3:2
+  p2 = dfcmp.eq(r1:0,r3:2)
+  if (!p2.new) jumpr:t r31
+ }
+
+ {
+  r1:0 = or(r5:4,r3:2)
+  jumpr r31
+ }
+.size __hexagon_mindf3,.-__hexagon_mindf3
+ .falign
+__hexagon_maxdf3:
+fmax:
+ {
+  p0 = dfclass(r1:0,#0x10)
+  p1 = dfcmp.gt(r3:2,r1:0)
+  r5:4 = r1:0
+ }
+ {
+  if (p0) r1:0 = r3:2
+  if (p1) r1:0 = r3:2
+  p2 = dfcmp.eq(r1:0,r3:2)
+  if (!p2.new) jumpr:t r31
+ }
+
+ {
+  r1:0 = and(r5:4,r3:2)
+  jumpr r31
+ }
+.size __hexagon_maxdf3,.-__hexagon_maxdf3
diff --git a/src/hexagon/dfmul.s b/src/hexagon/dfmul.s
new file mode 100644
index 00000000..32fc674f
--- /dev/null
+++ b/src/hexagon/dfmul.s
@@ -0,0 +1,309 @@
+ .text
+ .global __hexagon_muldf3
+ .type __hexagon_muldf3,@function
+ .global __qdsp_muldf3 ; .set __qdsp_muldf3, __hexagon_muldf3
+  .global __hexagon_fast_muldf3 ; .set __hexagon_fast_muldf3, __hexagon_muldf3
+  .global __hexagon_fast2_muldf3 ; .set __hexagon_fast2_muldf3, __hexagon_muldf3
+ .p2align 5
+__hexagon_muldf3:
+ {
+  p0 = dfclass(r1:0,#2)
+  p0 = dfclass(r3:2,#2)
+  r13:12 = combine(##0x40000000,#0)
+ }
+ {
+  r13:12 = insert(r1:0,#52,#11 -1)
+  r5:4 = asl(r3:2,#11 -1)
+  r28 = #-1024
+  r9:8 = #1
+ }
+ {
+  r7:6 = mpyu(r4,r13)
+  r5:4 = insert(r9:8,#2,#62)
+ }
+
+
+
+
+ {
+  r15:14 = mpyu(r12,r4)
+  r7:6 += mpyu(r12,r5)
+ }
+ {
+  r7:6 += lsr(r15:14,#32)
+  r11:10 = mpyu(r13,r5)
+  r5:4 = combine(##1024 +1024 -4,#0)
+ }
+ {
+  r11:10 += lsr(r7:6,#32)
+  if (!p0) jump .Lmul_abnormal
+  p1 = cmp.eq(r14,#0)
+  p1 = cmp.eq(r6,#0)
+ }
+ {
+  if (!p1) r10 = or(r10,r8)
+  r6 = extractu(r1,#11,#20)
+  r7 = extractu(r3,#11,#20)
+ }
+ {
+  r15:14 = neg(r11:10)
+  r6 += add(r28,r7)
+  r28 = xor(r1,r3)
+ }
+ {
+  if (!p2.new) r11:10 = r15:14
+  p2 = cmp.gt(r28,#-1)
+  p0 = !cmp.gt(r6,r5)
+  p0 = cmp.gt(r6,r4)
+  if (!p0.new) jump:nt .Lmul_ovf_unf
+ }
+ {
+  r1:0 = convert_d2df(r11:10)
+  r6 = add(r6,#-1024 -58)
+ }
+ {
+  r1 += asl(r6,#20)
+  jumpr r31
+ }
+
+ .falign
+.Lpossible_unf1:
+ {
+  p0 = cmp.eq(r0,#0)
+  p0 = bitsclr(r1,r4)
+  if (!p0.new) jumpr:t r31
+  r5 = #0x7fff
+ }
+ {
+  p0 = bitsset(r13,r5)
+  r4 = USR
+  r5 = #0x030
+ }
+ {
+  if (p0) r4 = or(r4,r5)
+ }
+ {
+  USR = r4
+ }
+ {
+  p0 = dfcmp.eq(r1:0,r1:0)
+  jumpr r31
+ }
+ .falign
+.Lmul_ovf_unf:
+ {
+  r1:0 = convert_d2df(r11:10)
+  r13:12 = abs(r11:10)
+  r7 = add(r6,#-1024 -58)
+ }
+ {
+  r1 += asl(r7,#20)
+  r7 = extractu(r1,#11,#20)
+  r4 = ##0x7FEFFFFF
+ }
+ {
+  r7 += add(r6,##-1024 -58)
+
+  r5 = #0
+ }
+ {
+  p0 = cmp.gt(r7,##1024 +1024 -2)
+  if (p0.new) jump:nt .Lmul_ovf
+ }
+ {
+  p0 = cmp.gt(r7,#0)
+  if (p0.new) jump:nt .Lpossible_unf1
+  r5 = sub(r6,r5)
+  r28 = #63
+ }
+ {
+  r4 = #0
+  r5 = sub(#5,r5)
+ }
+ {
+  p3 = cmp.gt(r11,#-1)
+  r5 = min(r5,r28)
+  r11:10 = r13:12
+ }
+ {
+  r28 = USR
+  r15:14 = extractu(r11:10,r5:4)
+ }
+ {
+  r11:10 = asr(r11:10,r5)
+  r4 = #0x0030
+  r1 = insert(r9,#11,#20)
+ }
+ {
+  p0 = cmp.gtu(r9:8,r15:14)
+  if (!p0.new) r10 = or(r10,r8)
+  r11 = setbit(r11,#20 +3)
+ }
+ {
+  r15:14 = neg(r11:10)
+  p1 = bitsclr(r10,#0x7)
+  if (!p1.new) r28 = or(r4,r28)
+ }
+ {
+  if (!p3) r11:10 = r15:14
+  USR = r28
+ }
+ {
+  r1:0 = convert_d2df(r11:10)
+  p0 = dfcmp.eq(r1:0,r1:0)
+ }
+ {
+  r1 = insert(r9,#11 -1,#20 +1)
+  jumpr r31
+ }
+ .falign
+.Lmul_ovf:
+
+ {
+  r28 = USR
+  r13:12 = combine(##0x7fefffff,#-1)
+  r1:0 = r11:10
+ }
+ {
+  r14 = extractu(r28,#2,#22)
+  r28 = or(r28,#0x28)
+  r5:4 = combine(##0x7ff00000,#0)
+ }
+ {
+  USR = r28
+  r14 ^= lsr(r1,#31)
+  r28 = r14
+ }
+ {
+  p0 = !cmp.eq(r28,#1)
+  p0 = !cmp.eq(r14,#2)
+  if (p0.new) r13:12 = r5:4
+  p0 = dfcmp.eq(r1:0,r1:0)
+ }
+ {
+  r1:0 = insert(r13:12,#63,#0)
+  jumpr r31
+ }
+
+.Lmul_abnormal:
+ {
+  r13:12 = extractu(r1:0,#63,#0)
+  r5:4 = extractu(r3:2,#63,#0)
+ }
+ {
+  p3 = cmp.gtu(r13:12,r5:4)
+  if (!p3.new) r1:0 = r3:2
+  if (!p3.new) r3:2 = r1:0
+ }
+ {
+
+  p0 = dfclass(r1:0,#0x0f)
+  if (!p0.new) jump:nt .Linvalid_nan
+  if (!p3) r13:12 = r5:4
+  if (!p3) r5:4 = r13:12
+ }
+ {
+
+  p1 = dfclass(r1:0,#0x08)
+  p1 = dfclass(r3:2,#0x0e)
+ }
+ {
+
+
+  p0 = dfclass(r1:0,#0x08)
+  p0 = dfclass(r3:2,#0x01)
+ }
+ {
+  if (p1) jump .Ltrue_inf
+  p2 = dfclass(r3:2,#0x01)
+ }
+ {
+  if (p0) jump .Linvalid_zeroinf
+  if (p2) jump .Ltrue_zero
+  r28 = ##0x7c000000
+ }
+
+
+
+
+
+ {
+  p0 = bitsclr(r1,r28)
+  if (p0.new) jump:nt .Lmul_tiny
+ }
+ {
+  r28 = cl0(r5:4)
+ }
+ {
+  r28 = add(r28,#-11)
+ }
+ {
+  r5:4 = asl(r5:4,r28)
+ }
+ {
+  r3:2 = insert(r5:4,#63,#0)
+  r1 -= asl(r28,#20)
+ }
+ jump __hexagon_muldf3
+.Lmul_tiny:
+ {
+  r28 = USR
+  r1:0 = xor(r1:0,r3:2)
+ }
+ {
+  r28 = or(r28,#0x30)
+  r1:0 = insert(r9:8,#63,#0)
+  r5 = extractu(r28,#2,#22)
+ }
+ {
+  USR = r28
+  p0 = cmp.gt(r5,#1)
+  if (!p0.new) r0 = #0
+  r5 ^= lsr(r1,#31)
+ }
+ {
+  p0 = cmp.eq(r5,#3)
+  if (!p0.new) r0 = #0
+  jumpr r31
+ }
+.Linvalid_zeroinf:
+ {
+  r28 = USR
+ }
+ {
+  r1:0 = #-1
+  r28 = or(r28,#2)
+ }
+ {
+  USR = r28
+ }
+ {
+  p0 = dfcmp.uo(r1:0,r1:0)
+  jumpr r31
+ }
+.Linvalid_nan:
+ {
+  p0 = dfclass(r3:2,#0x0f)
+  r28 = convert_df2sf(r1:0)
+  if (p0.new) r3:2 = r1:0
+ }
+ {
+  r2 = convert_df2sf(r3:2)
+  r1:0 = #-1
+  jumpr r31
+ }
+ .falign
+.Ltrue_zero:
+ {
+  r1:0 = r3:2
+  r3:2 = r1:0
+ }
+.Ltrue_inf:
+ {
+  r3 = extract(r3,#1,#31)
+ }
+ {
+  r1 ^= asl(r3,#31)
+  jumpr r31
+ }
+.size __hexagon_muldf3,.-__hexagon_muldf3
diff --git a/src/hexagon/dfsqrt.s b/src/hexagon/dfsqrt.s
new file mode 100644
index 00000000..14f584a1
--- /dev/null
+++ b/src/hexagon/dfsqrt.s
@@ -0,0 +1,277 @@
+ .text
+ .global __hexagon_sqrtdf2
+ .type __hexagon_sqrtdf2,@function
+ .global __hexagon_sqrt
+ .type __hexagon_sqrt,@function
+ .global __qdsp_sqrtdf2 ; .set __qdsp_sqrtdf2, __hexagon_sqrtdf2; .type __qdsp_sqrtdf2,@function
+ .global __qdsp_sqrt ; .set __qdsp_sqrt, __hexagon_sqrt; .type __qdsp_sqrt,@function
+ .global __hexagon_fast_sqrtdf2 ; .set __hexagon_fast_sqrtdf2, __hexagon_sqrtdf2; .type __hexagon_fast_sqrtdf2,@function
+ .global __hexagon_fast_sqrt ; .set __hexagon_fast_sqrt, __hexagon_sqrt; .type __hexagon_fast_sqrt,@function
+ .global __hexagon_fast2_sqrtdf2 ; .set __hexagon_fast2_sqrtdf2, __hexagon_sqrtdf2; .type __hexagon_fast2_sqrtdf2,@function
+ .global __hexagon_fast2_sqrt ; .set __hexagon_fast2_sqrt, __hexagon_sqrt; .type __hexagon_fast2_sqrt,@function
+ .type sqrt,@function
+ .p2align 5
+__hexagon_sqrtdf2:
+__hexagon_sqrt:
+ {
+  r15:14 = extractu(r1:0,#23 +1,#52 -23)
+  r28 = extractu(r1,#11,#52 -32)
+  r5:4 = combine(##0x3f000004,#1)
+ }
+ {
+  p2 = dfclass(r1:0,#0x02)
+  p2 = cmp.gt(r1,#-1)
+  if (!p2.new) jump:nt .Lsqrt_abnormal
+  r9 = or(r5,r14)
+ }
+
+.Ldenormal_restart:
+ {
+  r11:10 = r1:0
+  r7,p0 = sfinvsqrta(r9)
+  r5 = and(r5,#-16)
+  r3:2 = #0
+ }
+ {
+  r3 += sfmpy(r7,r9):lib
+  r2 += sfmpy(r7,r5):lib
+  r6 = r5
+
+
+  r9 = and(r28,#1)
+ }
+ {
+  r6 -= sfmpy(r3,r2):lib
+  r11 = insert(r4,#11 +1,#52 -32)
+  p1 = cmp.gtu(r9,#0)
+ }
+ {
+  r3 += sfmpy(r3,r6):lib
+  r2 += sfmpy(r2,r6):lib
+  r6 = r5
+  r9 = mux(p1,#8,#9)
+ }
+ {
+  r6 -= sfmpy(r3,r2):lib
+  r11:10 = asl(r11:10,r9)
+  r9 = mux(p1,#3,#2)
+ }
+ {
+  r2 += sfmpy(r2,r6):lib
+
+  r15:14 = asl(r11:10,r9)
+ }
+ {
+  r2 = and(r2,##0x007fffff)
+ }
+ {
+  r2 = add(r2,##0x00800000 - 3)
+  r9 = mux(p1,#7,#8)
+ }
+ {
+  r8 = asl(r2,r9)
+  r9 = mux(p1,#15-(1+1),#15-(1+0))
+ }
+ {
+  r13:12 = mpyu(r8,r15)
+ }
+ {
+  r1:0 = asl(r11:10,#15)
+  r15:14 = mpyu(r13,r13)
+  p1 = cmp.eq(r0,r0)
+ }
+ {
+  r1:0 -= asl(r15:14,#15)
+  r15:14 = mpyu(r13,r12)
+  p2 = cmp.eq(r0,r0)
+ }
+ {
+  r1:0 -= lsr(r15:14,#16)
+  p3 = cmp.eq(r0,r0)
+ }
+ {
+  r1:0 = mpyu(r1,r8)
+ }
+ {
+  r13:12 += lsr(r1:0,r9)
+  r9 = add(r9,#16)
+  r1:0 = asl(r11:10,#31)
+ }
+
+ {
+  r15:14 = mpyu(r13,r13)
+  r1:0 -= mpyu(r13,r12)
+ }
+ {
+  r1:0 -= asl(r15:14,#31)
+  r15:14 = mpyu(r12,r12)
+ }
+ {
+  r1:0 -= lsr(r15:14,#33)
+ }
+ {
+  r1:0 = mpyu(r1,r8)
+ }
+ {
+  r13:12 += lsr(r1:0,r9)
+  r9 = add(r9,#16)
+  r1:0 = asl(r11:10,#47)
+ }
+
+ {
+  r15:14 = mpyu(r13,r13)
+ }
+ {
+  r1:0 -= asl(r15:14,#47)
+  r15:14 = mpyu(r13,r12)
+ }
+ {
+  r1:0 -= asl(r15:14,#16)
+  r15:14 = mpyu(r12,r12)
+ }
+ {
+  r1:0 -= lsr(r15:14,#17)
+ }
+ {
+  r1:0 = mpyu(r1,r8)
+ }
+ {
+  r13:12 += lsr(r1:0,r9)
+ }
+ {
+  r3:2 = mpyu(r13,r12)
+  r5:4 = mpyu(r12,r12)
+  r15:14 = #0
+  r1:0 = #0
+ }
+ {
+  r3:2 += lsr(r5:4,#33)
+  r5:4 += asl(r3:2,#33)
+  p1 = cmp.eq(r0,r0)
+ }
+ {
+  r7:6 = mpyu(r13,r13)
+  r1:0 = sub(r1:0,r5:4,p1):carry
+  r9:8 = #1
+ }
+ {
+  r7:6 += lsr(r3:2,#31)
+  r9:8 += asl(r13:12,#1)
+ }
+
+
+
+
+
+ {
+  r15:14 = sub(r11:10,r7:6,p1):carry
+  r5:4 = sub(r1:0,r9:8,p2):carry
+
+
+
+
+  r7:6 = #1
+  r11:10 = #0
+ }
+ {
+  r3:2 = sub(r15:14,r11:10,p2):carry
+  r7:6 = add(r13:12,r7:6)
+  r28 = add(r28,#-0x3ff)
+ }
+ {
+
+  if (p2) r13:12 = r7:6
+  if (p2) r1:0 = r5:4
+  if (p2) r15:14 = r3:2
+ }
+ {
+  r5:4 = sub(r1:0,r9:8,p3):carry
+  r7:6 = #1
+  r28 = asr(r28,#1)
+ }
+ {
+  r3:2 = sub(r15:14,r11:10,p3):carry
+  r7:6 = add(r13:12,r7:6)
+ }
+ {
+  if (p3) r13:12 = r7:6
+  if (p3) r1:0 = r5:4
+
+
+
+
+
+  r2 = #1
+ }
+ {
+  p0 = cmp.eq(r1:0,r11:10)
+  if (!p0.new) r12 = or(r12,r2)
+  r3 = cl0(r13:12)
+  r28 = add(r28,#-63)
+ }
+
+
+
+ {
+  r1:0 = convert_ud2df(r13:12)
+  r28 = add(r28,r3)
+ }
+ {
+  r1 += asl(r28,#52 -32)
+  jumpr r31
+ }
+.Lsqrt_abnormal:
+ {
+  p0 = dfclass(r1:0,#0x01)
+  if (p0.new) jumpr:t r31
+ }
+ {
+  p0 = dfclass(r1:0,#0x10)
+  if (p0.new) jump:nt .Lsqrt_nan
+ }
+ {
+  p0 = cmp.gt(r1,#-1)
+  if (!p0.new) jump:nt .Lsqrt_invalid_neg
+  if (!p0.new) r28 = ##0x7F800001
+ }
+ {
+  p0 = dfclass(r1:0,#0x08)
+  if (p0.new) jumpr:nt r31
+ }
+
+
+ {
+  r1:0 = extractu(r1:0,#52,#0)
+ }
+ {
+  r28 = add(clb(r1:0),#-11)
+ }
+ {
+  r1:0 = asl(r1:0,r28)
+  r28 = sub(#1,r28)
+ }
+ {
+  r1 = insert(r28,#1,#52 -32)
+ }
+ {
+  r3:2 = extractu(r1:0,#23 +1,#52 -23)
+  r5 = ##0x3f000004
+ }
+ {
+  r9 = or(r5,r2)
+  r5 = and(r5,#-16)
+  jump .Ldenormal_restart
+ }
+.Lsqrt_nan:
+ {
+  r28 = convert_df2sf(r1:0)
+  r1:0 = #-1
+  jumpr r31
+ }
+.Lsqrt_invalid_neg:
+ {
+  r1:0 = convert_sf2df(r28)
+  jumpr r31
+ }
+.size __hexagon_sqrt,.-__hexagon_sqrt
+.size __hexagon_sqrtdf2,.-__hexagon_sqrtdf2
diff --git a/src/hexagon/divdi3.s b/src/hexagon/divdi3.s
new file mode 100644
index 00000000..0fee6e70
--- /dev/null
+++ b/src/hexagon/divdi3.s
@@ -0,0 +1,64 @@
+
+FUNCTION_BEGIN __hexagon_divdi3
+ {
+  p2 = tstbit(r1,#31)
+  p3 = tstbit(r3,#31)
+ }
+ {
+  r1:0 = abs(r1:0)
+  r3:2 = abs(r3:2)
+ }
+ {
+  r6 = cl0(r1:0)
+  r7 = cl0(r3:2)
+  r5:4 = r3:2
+  r3:2 = r1:0
+ }
+ {
+  p3 = xor(p2,p3)
+  r10 = sub(r7,r6)
+  r1:0 = #0
+  r15:14 = #1
+ }
+ {
+  r11 = add(r10,#1)
+  r13:12 = lsl(r5:4,r10)
+  r15:14 = lsl(r15:14,r10)
+ }
+ {
+  p0 = cmp.gtu(r5:4,r3:2)
+  loop0(1f,r11)
+ }
+ {
+  if (p0) jump .hexagon_divdi3_return
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r13:12,r3:2)
+ }
+ {
+  r7:6 = sub(r3:2, r13:12)
+  r9:8 = add(r1:0, r15:14)
+ }
+ {
+  r1:0 = vmux(p0, r1:0, r9:8)
+  r3:2 = vmux(p0, r3:2, r7:6)
+ }
+ {
+  r15:14 = lsr(r15:14, #1)
+  r13:12 = lsr(r13:12, #1)
+ }:endloop0
+
+.hexagon_divdi3_return:
+ {
+  r3:2 = neg(r1:0)
+ }
+ {
+  r1:0 = vmux(p3,r3:2,r1:0)
+  jumpr r31
+ }
+FUNCTION_END __hexagon_divdi3
+
+  .globl __qdsp_divdi3
+  .set __qdsp_divdi3, __hexagon_divdi3
diff --git a/src/hexagon/divsi3.s b/src/hexagon/divsi3.s
new file mode 100644
index 00000000..fc957a43
--- /dev/null
+++ b/src/hexagon/divsi3.s
@@ -0,0 +1,53 @@
+
+FUNCTION_BEGIN __hexagon_divsi3
+ {
+  p0 = cmp.ge(r0,#0)
+  p1 = cmp.ge(r1,#0)
+  r1 = abs(r0)
+  r2 = abs(r1)
+ }
+ {
+  r3 = cl0(r1)
+  r4 = cl0(r2)
+  r5 = sub(r1,r2)
+  p2 = cmp.gtu(r2,r1)
+ }
+ {
+  r0 = #0
+  p1 = xor(p0,p1)
+  p0 = cmp.gtu(r2,r5)
+  if (p2) jumpr r31
+ }
+
+ {
+  r0 = mux(p1,#-1,#1)
+  if (p0) jumpr r31
+  r4 = sub(r4,r3)
+  r3 = #1
+ }
+ {
+  r0 = #0
+  r3:2 = vlslw(r3:2,r4)
+  loop0(1f,r4)
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r2,r1)
+  if (!p0.new) r1 = sub(r1,r2)
+  if (!p0.new) r0 = add(r0,r3)
+  r3:2 = vlsrw(r3:2,#1)
+ }:endloop0
+ {
+  p0 = cmp.gtu(r2,r1)
+  if (!p0.new) r0 = add(r0,r3)
+  if (!p1) jumpr r31
+ }
+ {
+  r0 = neg(r0)
+  jumpr r31
+ }
+FUNCTION_END __hexagon_divsi3
+
+  .globl __qdsp_divsi3
+  .set __qdsp_divsi3, __hexagon_divsi3
diff --git a/src/hexagon/fastmath2_dlib_asm.s b/src/hexagon/fastmath2_dlib_asm.s
new file mode 100644
index 00000000..15c38784
--- /dev/null
+++ b/src/hexagon/fastmath2_dlib_asm.s
@@ -0,0 +1,266 @@
+        .text
+        .global fast2_dadd_asm
+        .type fast2_dadd_asm, @function
+fast2_dadd_asm:
+        .falign
+      {
+        R7:6 = VABSDIFFH(R1:0, R3:2)
+        R9 = #62
+        R4 = SXTH(R0)
+        R5 = SXTH(R2)
+      } {
+        R6 = SXTH(R6)
+        P0 = CMP.GT(R4, R5);
+        if ( P0.new) R8 = add(R4, #1)
+        if (!P0.new) R8 = add(R5, #1)
+      } {
+        if ( P0) R4 = #1
+        if (!P0) R5 = #1
+        R0.L = #0
+        R6 = MIN(R6, R9)
+      } {
+        if (!P0) R4 = add(R6, #1)
+        if ( P0) R5 = add(R6, #1)
+        R2.L = #0
+        R11:10 = #0
+      } {
+        R1:0 = ASR(R1:0, R4)
+        R3:2 = ASR(R3:2, R5)
+      } {
+        R1:0 = add(R1:0, R3:2)
+        R10.L = #0x8001
+      } {
+        R4 = clb(R1:0)
+        R9 = #58
+      } {
+        R4 = add(R4, #-1)
+        p0 = cmp.gt(R4, R9)
+      } {
+        R1:0 = ASL(R1:0, R4)
+        R8 = SUB(R8, R4)
+        if(p0) jump .Ldenorma
+      } {
+        R0 = insert(R8, #16, #0)
+        jumpr r31
+      }
+.Ldenorma:
+      {
+        R1:0 = R11:10
+        jumpr r31
+      }
+        .text
+        .global fast2_dsub_asm
+        .type fast2_dsub_asm, @function
+fast2_dsub_asm:
+        .falign
+      {
+        R7:6 = VABSDIFFH(R1:0, R3:2)
+        R9 = #62
+        R4 = SXTH(R0)
+        R5 = SXTH(R2)
+      } {
+        R6 = SXTH(R6)
+        P0 = CMP.GT(R4, R5);
+        if ( P0.new) R8 = add(R4, #1)
+        if (!P0.new) R8 = add(R5, #1)
+      } {
+        if ( P0) R4 = #1
+        if (!P0) R5 = #1
+        R0.L = #0
+        R6 = MIN(R6, R9)
+      } {
+        if (!P0) R4 = add(R6, #1)
+        if ( P0) R5 = add(R6, #1)
+        R2.L = #0
+        R11:10 = #0
+      } {
+        R1:0 = ASR(R1:0, R4)
+        R3:2 = ASR(R3:2, R5)
+      } {
+        R1:0 = sub(R1:0, R3:2)
+        R10.L = #0x8001
+      } {
+        R4 = clb(R1:0)
+        R9 = #58
+      } {
+        R4 = add(R4, #-1)
+        p0 = cmp.gt(R4, R9)
+      } {
+        R1:0 = ASL(R1:0, R4)
+        R8 = SUB(R8, R4)
+        if(p0) jump .Ldenorm
+      } {
+        R0 = insert(R8, #16, #0)
+        jumpr r31
+      }
+.Ldenorm:
+      {
+        R1:0 = R11:10
+        jumpr r31
+      }
+        .text
+        .global fast2_dmpy_asm
+        .type fast2_dmpy_asm, @function
+fast2_dmpy_asm:
+        .falign
+      {
+        R13= lsr(R2, #16)
+        R5 = sxth(R2)
+        R4 = sxth(R0)
+        R12= lsr(R0, #16)
+      }
+      {
+        R11:10 = mpy(R1, R3)
+        R7:6 = mpy(R1, R13)
+        R0.L = #0x0
+        R15:14 = #0
+      }
+      {
+        R11:10 = add(R11:10, R11:10)
+        R7:6 += mpy(R3, R12)
+        R2.L = #0x0
+        R15.H = #0x8000
+      }
+      {
+        R7:6 = asr(R7:6, #15)
+        R12.L = #0x8001
+        p1 = cmp.eq(R1:0, R3:2)
+      }
+      {
+        R7:6 = add(R7:6, R11:10)
+        R8 = add(R4, R5)
+        p2 = cmp.eq(R1:0, R15:14)
+      }
+      {
+        R9 = clb(R7:6)
+        R3:2 = abs(R7:6)
+        R11 = #58
+      }
+      {
+        p1 = and(p1, p2)
+        R8 = sub(R8, R9)
+        R9 = add(R9, #-1)
+ p0 = cmp.gt(R9, R11)
+      }
+      {
+        R8 = add(R8, #1)
+        R1:0 = asl(R7:6, R9)
+        if(p1) jump .Lsat
+      }
+      {
+        R0 = insert(R8,#16, #0)
+        if(!p0) jumpr r31
+      }
+      {
+        R0 = insert(R12,#16, #0)
+        jumpr r31
+      }
+.Lsat:
+      {
+        R1:0 = #-1
+      }
+      {
+        R1:0 = lsr(R1:0, #1)
+      }
+      {
+        R0 = insert(R8,#16, #0)
+        jumpr r31
+      }
+        .text
+        .global fast2_qd2f_asm
+        .type fast2_qd2f_asm, @function
+fast2_qd2f_asm:
+      .falign
+     {
+       R3 = abs(R1):sat
+       R4 = sxth(R0)
+       R5 = #0x40
+       R6.L = #0xffc0
+     }
+     {
+       R0 = extractu(R3, #8, #0)
+       p2 = cmp.gt(R4, #126)
+       p3 = cmp.ge(R4, #-126)
+       R6.H = #0x7fff
+     }
+     {
+       p1 = cmp.eq(R0,#0x40)
+       if(p1.new) R5 = #0
+       R4 = add(R4, #126)
+       if(!p3) jump .Lmin
+     }
+     {
+       p0 = bitsset(R3, R6)
+       R0.L = #0x0000
+       R2 = add(R3, R5)
+       R7 = lsr(R6, #8)
+     }
+     {
+       if(p0) R4 = add(R4, #1)
+       if(p0) R3 = #0
+       R2 = lsr(R2, #7)
+       R0.H = #0x8000
+     }
+     {
+       R0 = and(R0, R1)
+       R6 &= asl(R4, #23)
+       if(!p0) R3 = and(R2, R7)
+       if(p2) jump .Lmax
+     }
+     {
+       R0 += add(R6, R3)
+       jumpr r31
+     }
+.Lmax:
+     {
+       R0.L = #0xffff;
+     }
+     {
+       R0.H = #0x7f7f;
+       jumpr r31
+     }
+.Lmin:
+     {
+       R0 = #0x0
+       jumpr r31
+     }
+        .text
+        .global fast2_f2qd_asm
+        .type fast2_f2qd_asm, @function
+fast2_f2qd_asm:
+
+
+
+
+
+
+
+        .falign
+  {
+       R1 = asl(R0, #7)
+       p0 = tstbit(R0, #31)
+       R5:4 = #0
+       R3 = add(R0,R0)
+  }
+  {
+       R1 = setbit(R1, #30)
+       R0= extractu(R0,#8,#23)
+       R4.L = #0x8001
+       p1 = cmp.eq(R3, #0)
+  }
+  {
+       R1= extractu(R1, #31, #0)
+       R0= add(R0, #-126)
+       R2 = #0
+       if(p1) jump .Lminqd
+  }
+  {
+       R0 = zxth(R0)
+       if(p0) R1= sub(R2, R1)
+       jumpr r31
+  }
+.Lminqd:
+  {
+       R1:0 = R5:4
+       jumpr r31
+  }
diff --git a/src/hexagon/fastmath2_ldlib_asm.s b/src/hexagon/fastmath2_ldlib_asm.s
new file mode 100644
index 00000000..b72b7550
--- /dev/null
+++ b/src/hexagon/fastmath2_ldlib_asm.s
@@ -0,0 +1,187 @@
+        .text
+        .global fast2_ldadd_asm
+        .type fast2_ldadd_asm, @function
+fast2_ldadd_asm:
+        .falign
+      {
+        R4 = memw(r29+#8)
+        R5 = memw(r29+#24)
+        r7 = r0
+      }
+      {
+        R6 = sub(R4, R5):sat
+        P0 = CMP.GT(R4, R5);
+        if ( P0.new) R8 = add(R4, #1)
+        if (!P0.new) R8 = add(R5, #1)
+      } {
+        R6 = abs(R6):sat
+        if ( P0) R4 = #1
+        if (!P0) R5 = #1
+        R9 = #62
+      } {
+        R6 = MIN(R6, R9)
+        R1:0 = memd(r29+#0)
+        R3:2 = memd(r29+#16)
+      } {
+        if (!P0) R4 = add(R6, #1)
+        if ( P0) R5 = add(R6, #1)
+      } {
+        R1:0 = ASR(R1:0, R4)
+        R3:2 = ASR(R3:2, R5)
+      } {
+        R1:0 = add(R1:0, R3:2)
+        R3:2 = #0
+      } {
+        R4 = clb(R1:0)
+        R9.L =#0x0001
+      } {
+        R8 -= add(R4, #-1)
+        R4 = add(R4, #-1)
+        p0 = cmp.gt(R4, #58)
+        R9.H =#0x8000
+      } {
+        if(!p0)memw(r7+#8) = R8
+        R1:0 = ASL(R1:0, R4)
+        if(p0) jump .Ldenorma1
+      } {
+        memd(r7+#0) = R1:0
+        jumpr r31
+      }
+.Ldenorma1:
+        memd(r7+#0) = R3:2
+      {
+        memw(r7+#8) = R9
+        jumpr r31
+      }
+        .text
+        .global fast2_ldsub_asm
+        .type fast2_ldsub_asm, @function
+fast2_ldsub_asm:
+        .falign
+      {
+        R4 = memw(r29+#8)
+        R5 = memw(r29+#24)
+        r7 = r0
+      }
+      {
+        R6 = sub(R4, R5):sat
+        P0 = CMP.GT(R4, R5);
+        if ( P0.new) R8 = add(R4, #1)
+        if (!P0.new) R8 = add(R5, #1)
+      } {
+        R6 = abs(R6):sat
+        if ( P0) R4 = #1
+        if (!P0) R5 = #1
+        R9 = #62
+      } {
+        R6 = min(R6, R9)
+        R1:0 = memd(r29+#0)
+        R3:2 = memd(r29+#16)
+      } {
+        if (!P0) R4 = add(R6, #1)
+        if ( P0) R5 = add(R6, #1)
+      } {
+        R1:0 = ASR(R1:0, R4)
+        R3:2 = ASR(R3:2, R5)
+      } {
+        R1:0 = sub(R1:0, R3:2)
+        R3:2 = #0
+      } {
+        R4 = clb(R1:0)
+        R9.L =#0x0001
+      } {
+        R8 -= add(R4, #-1)
+        R4 = add(R4, #-1)
+        p0 = cmp.gt(R4, #58)
+        R9.H =#0x8000
+      } {
+        if(!p0)memw(r7+#8) = R8
+        R1:0 = asl(R1:0, R4)
+        if(p0) jump .Ldenorma_s
+      } {
+        memd(r7+#0) = R1:0
+        jumpr r31
+      }
+.Ldenorma_s:
+        memd(r7+#0) = R3:2
+      {
+        memw(r7+#8) = R9
+        jumpr r31
+      }
+        .text
+        .global fast2_ldmpy_asm
+        .type fast2_ldmpy_asm, @function
+fast2_ldmpy_asm:
+        .falign
+      {
+        R15:14 = memd(r29+#0)
+        R3:2 = memd(r29+#16)
+        R13:12 = #0
+      }
+      {
+        R8= extractu(R2, #31, #1)
+        R9= extractu(R14, #31, #1)
+        R13.H = #0x8000
+      }
+      {
+        R11:10 = mpy(R15, R3)
+        R7:6 = mpy(R15, R8)
+        R4 = memw(r29+#8)
+        R5 = memw(r29+#24)
+      }
+      {
+        R11:10 = add(R11:10, R11:10)
+        R7:6 += mpy(R3, R9)
+      }
+      {
+        R7:6 = asr(R7:6, #30)
+        R8.L = #0x0001
+        p1 = cmp.eq(R15:14, R3:2)
+      }
+      {
+        R7:6 = add(R7:6, R11:10)
+        R4= add(R4, R5)
+        p2 = cmp.eq(R3:2, R13:12)
+      }
+      {
+        R9 = clb(R7:6)
+        R8.H = #0x8000
+        p1 = and(p1, p2)
+      }
+      {
+        R4-= add(R9, #-1)
+        R9 = add(R9, #-1)
+        if(p1) jump .Lsat1
+      }
+      {
+        R7:6 = asl(R7:6, R9)
+        memw(R0+#8) = R4
+ p0 = cmp.gt(R9, #58)
+        if(p0.new) jump:NT .Ldenorm1
+      }
+      {
+        memd(R0+#0) = R7:6
+        jumpr r31
+      }
+.Lsat1:
+      {
+        R13:12 = #0
+        R4+= add(R9, #1)
+      }
+      {
+        R13.H = #0x4000
+        memw(R0+#8) = R4
+      }
+      {
+        memd(R0+#0) = R13:12
+        jumpr r31
+      }
+.Ldenorm1:
+      {
+        memw(R0+#8) = R8
+        R15:14 = #0
+      }
+      {
+        memd(R0+#0) = R15:14
+        jumpr r31
+      }
diff --git a/src/hexagon/func_macro.s b/src/hexagon/func_macro.s
new file mode 100644
index 00000000..9a1e11ae
--- /dev/null
+++ b/src/hexagon/func_macro.s
@@ -0,0 +1,12 @@
+ .macro FUNCTION_BEGIN name
+ .text
+ .p2align 5
+ .globl \name
+ .type \name, @function
+\name:
+ .endm
+
+ .macro FUNCTION_END name
+ .size \name, . - \name
+ .endm
+
diff --git a/src/hexagon/memcpy_forward_vp4cp4n2.s b/src/hexagon/memcpy_forward_vp4cp4n2.s
new file mode 100644
index 00000000..89f69010
--- /dev/null
+++ b/src/hexagon/memcpy_forward_vp4cp4n2.s
@@ -0,0 +1,91 @@
+  .text
+
+
+
+
+
+
+  .globl hexagon_memcpy_forward_vp4cp4n2
+  .balign 32
+  .type hexagon_memcpy_forward_vp4cp4n2,@function
+hexagon_memcpy_forward_vp4cp4n2:
+
+
+
+
+  {
+    r3 = sub(##4096, r1)
+    r5 = lsr(r2, #3)
+  }
+  {
+
+
+    r3 = extractu(r3, #10, #2)
+    r4 = extractu(r3, #7, #5)
+  }
+  {
+    r3 = minu(r2, r3)
+    r4 = minu(r5, r4)
+  }
+  {
+    r4 = or(r4, ##2105344)
+    p0 = cmp.eq(r3, #0)
+    if (p0.new) jump:nt .Lskipprolog
+  }
+    l2fetch(r1, r4)
+  {
+    loop0(.Lprolog, r3)
+    r2 = sub(r2, r3)
+  }
+  .falign
+.Lprolog:
+  {
+    r4 = memw(r1++#4)
+    memw(r0++#4) = r4.new
+  } :endloop0
+.Lskipprolog:
+  {
+
+    r3 = lsr(r2, #10)
+    if (cmp.eq(r3.new, #0)) jump:nt .Lskipmain
+  }
+  {
+    loop1(.Lout, r3)
+    r2 = extractu(r2, #10, #0)
+    r3 = ##2105472
+  }
+
+  .falign
+.Lout:
+
+    l2fetch(r1, r3)
+    loop0(.Lpage, #512)
+  .falign
+.Lpage:
+    r5:4 = memd(r1++#8)
+  {
+    memw(r0++#8) = r4
+    memw(r0+#4) = r5
+  } :endloop0:endloop1
+.Lskipmain:
+  {
+    r3 = ##2105344
+    r4 = lsr(r2, #3)
+    p0 = cmp.eq(r2, #0)
+    if (p0.new) jumpr:nt r31
+  }
+  {
+    r3 = or(r3, r4)
+    loop0(.Lepilog, r2)
+  }
+    l2fetch(r1, r3)
+  .falign
+.Lepilog:
+  {
+    r4 = memw(r1++#4)
+    memw(r0++#4) = r4.new
+  } :endloop0
+
+    jumpr r31
+
+.size hexagon_memcpy_forward_vp4cp4n2, . - hexagon_memcpy_forward_vp4cp4n2
diff --git a/src/hexagon/memcpy_likely_aligned.s b/src/hexagon/memcpy_likely_aligned.s
new file mode 100644
index 00000000..7e9b62f6
--- /dev/null
+++ b/src/hexagon/memcpy_likely_aligned.s
@@ -0,0 +1,42 @@
+
+FUNCTION_BEGIN __hexagon_memcpy_likely_aligned_min32bytes_mult8bytes
+ {
+  p0 = bitsclr(r1,#7)
+  p0 = bitsclr(r0,#7)
+  if (p0.new) r5:4 = memd(r1)
+  r3 = #-3
+ }
+ {
+  if (!p0) jump .Lmemcpy_call
+  if (p0) memd(r0++#8) = r5:4
+  if (p0) r5:4 = memd(r1+#8)
+  r3 += lsr(r2,#3)
+ }
+ {
+  memd(r0++#8) = r5:4
+  r5:4 = memd(r1+#16)
+  r1 = add(r1,#24)
+  loop0(1f,r3)
+ }
+ .falign
+1:
+ {
+  memd(r0++#8) = r5:4
+  r5:4 = memd(r1++#8)
+ }:endloop0
+ {
+  memd(r0) = r5:4
+  r0 -= add(r2,#-8)
+  jumpr r31
+ }
+FUNCTION_END __hexagon_memcpy_likely_aligned_min32bytes_mult8bytes
+
+.Lmemcpy_call:
+
+ jump memcpy@PLT
+
+
+
+
+  .globl __qdsp_memcpy_likely_aligned_min32bytes_mult8bytes
+  .set __qdsp_memcpy_likely_aligned_min32bytes_mult8bytes, __hexagon_memcpy_likely_aligned_min32bytes_mult8bytes
diff --git a/src/hexagon/moddi3.s b/src/hexagon/moddi3.s
new file mode 100644
index 00000000..53ea6d52
--- /dev/null
+++ b/src/hexagon/moddi3.s
@@ -0,0 +1,63 @@
+
+
+FUNCTION_BEGIN __hexagon_moddi3
+ {
+  p3 = tstbit(r1,#31)
+ }
+ {
+  r1:0 = abs(r1:0)
+  r3:2 = abs(r3:2)
+ }
+ {
+  r6 = cl0(r1:0)
+  r7 = cl0(r3:2)
+  r5:4 = r3:2
+  r3:2 = r1:0
+ }
+ {
+  r10 = sub(r7,r6)
+  r1:0 = #0
+  r15:14 = #1
+ }
+ {
+  r11 = add(r10,#1)
+  r13:12 = lsl(r5:4,r10)
+  r15:14 = lsl(r15:14,r10)
+ }
+ {
+  p0 = cmp.gtu(r5:4,r3:2)
+  loop0(1f,r11)
+ }
+ {
+  if (p0) jump .hexagon_moddi3_return
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r13:12,r3:2)
+ }
+ {
+  r7:6 = sub(r3:2, r13:12)
+  r9:8 = add(r1:0, r15:14)
+ }
+ {
+  r1:0 = vmux(p0, r1:0, r9:8)
+  r3:2 = vmux(p0, r3:2, r7:6)
+ }
+ {
+  r15:14 = lsr(r15:14, #1)
+  r13:12 = lsr(r13:12, #1)
+ }:endloop0
+
+.hexagon_moddi3_return:
+ {
+  r1:0 = neg(r3:2)
+ }
+ {
+  r1:0 = vmux(p3,r1:0,r3:2)
+  jumpr r31
+ }
+FUNCTION_END __hexagon_moddi3
+
+  .globl __qdsp_moddi3
+  .set __qdsp_moddi3, __hexagon_moddi3
diff --git a/src/hexagon/modsi3.s b/src/hexagon/modsi3.s
new file mode 100644
index 00000000..c4ae7e59
--- /dev/null
+++ b/src/hexagon/modsi3.s
@@ -0,0 +1,44 @@
+
+
+FUNCTION_BEGIN __hexagon_modsi3
+ {
+  p2 = cmp.ge(r0,#0)
+  r2 = abs(r0)
+  r1 = abs(r1)
+ }
+ {
+  r3 = cl0(r2)
+  r4 = cl0(r1)
+  p0 = cmp.gtu(r1,r2)
+ }
+ {
+  r3 = sub(r4,r3)
+  if (p0) jumpr r31
+ }
+ {
+  p1 = cmp.eq(r3,#0)
+  loop0(1f,r3)
+  r0 = r2
+  r2 = lsl(r1,r3)
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r2,r0)
+  if (!p0.new) r0 = sub(r0,r2)
+  r2 = lsr(r2,#1)
+  if (p1) r1 = #0
+ }:endloop0
+ {
+  p0 = cmp.gtu(r2,r0)
+  if (!p0.new) r0 = sub(r0,r1)
+  if (p2) jumpr r31
+ }
+ {
+  r0 = neg(r0)
+  jumpr r31
+ }
+FUNCTION_END __hexagon_modsi3
+
+  .globl __qdsp_modsi3
+  .set __qdsp_modsi3, __hexagon_modsi3
diff --git a/src/hexagon/sfdiv_opt.s b/src/hexagon/sfdiv_opt.s
new file mode 100644
index 00000000..26c91f15
--- /dev/null
+++ b/src/hexagon/sfdiv_opt.s
@@ -0,0 +1,42 @@
+
+FUNCTION_BEGIN __hexagon_divsf3
+  {
+    r2,p0 = sfrecipa(r0,r1)
+    r4 = sffixupd(r0,r1)
+    r3 = ##0x3f800000
+  }
+  {
+    r5 = sffixupn(r0,r1)
+    r3 -= sfmpy(r4,r2):lib
+    r6 = ##0x80000000
+    r7 = r3
+  }
+  {
+    r2 += sfmpy(r3,r2):lib
+    r3 = r7
+    r6 = r5
+    r0 = and(r6,r5)
+  }
+  {
+    r3 -= sfmpy(r4,r2):lib
+    r0 += sfmpy(r5,r2):lib
+  }
+  {
+    r2 += sfmpy(r3,r2):lib
+    r6 -= sfmpy(r0,r4):lib
+  }
+  {
+    r0 += sfmpy(r6,r2):lib
+  }
+  {
+    r5 -= sfmpy(r0,r4):lib
+  }
+  {
+    r0 += sfmpy(r5,r2,p0):scale
+    jumpr r31
+  }
+FUNCTION_END __hexagon_divsf3
+
+.global __qdsp_divsf3 ; .set __qdsp_divsf3, __hexagon_divsf3
+.global __hexagon_fast_divsf3 ; .set __hexagon_fast_divsf3, __hexagon_divsf3
+.global __hexagon_fast2_divsf3 ; .set __hexagon_fast2_divsf3, __hexagon_divsf3
diff --git a/src/hexagon/sfsqrt_opt.s b/src/hexagon/sfsqrt_opt.s
new file mode 100644
index 00000000..c90af179
--- /dev/null
+++ b/src/hexagon/sfsqrt_opt.s
@@ -0,0 +1,49 @@
+FUNCTION_BEGIN __hexagon_sqrtf
+  {
+    r3,p0 = sfinvsqrta(r0)
+    r5 = sffixupr(r0)
+    r4 = ##0x3f000000
+    r1:0 = combine(#0,#0)
+  }
+  {
+    r0 += sfmpy(r3,r5):lib
+    r1 += sfmpy(r3,r4):lib
+    r2 = r4
+    r3 = r5
+  }
+  {
+    r2 -= sfmpy(r0,r1):lib
+    p1 = sfclass(r5,#1)
+
+  }
+  {
+    r0 += sfmpy(r0,r2):lib
+    r1 += sfmpy(r1,r2):lib
+    r2 = r4
+    r3 = r5
+  }
+  {
+    r2 -= sfmpy(r0,r1):lib
+    r3 -= sfmpy(r0,r0):lib
+  }
+  {
+    r0 += sfmpy(r1,r3):lib
+    r1 += sfmpy(r1,r2):lib
+    r2 = r4
+    r3 = r5
+  }
+  {
+
+    r3 -= sfmpy(r0,r0):lib
+    if (p1) r0 = or(r0,r5)
+  }
+  {
+    r0 += sfmpy(r1,r3,p0):scale
+    jumpr r31
+  }
+
+FUNCTION_END __hexagon_sqrtf
+
+.global __qdsp_sqrtf ; .set __qdsp_sqrtf, __hexagon_sqrtf
+.global __hexagon_fast_sqrtf ; .set __hexagon_fast_sqrtf, __hexagon_sqrtf
+.global __hexagon_fast2_sqrtf ; .set __hexagon_fast2_sqrtf, __hexagon_sqrtf
diff --git a/src/hexagon/udivdi3.s b/src/hexagon/udivdi3.s
new file mode 100644
index 00000000..f0fffc23
--- /dev/null
+++ b/src/hexagon/udivdi3.s
@@ -0,0 +1,50 @@
+
+
+FUNCTION_BEGIN __hexagon_udivdi3
+ {
+  r6 = cl0(r1:0)
+  r7 = cl0(r3:2)
+  r5:4 = r3:2
+  r3:2 = r1:0
+ }
+ {
+  r10 = sub(r7,r6)
+  r1:0 = #0
+  r15:14 = #1
+ }
+ {
+  r11 = add(r10,#1)
+  r13:12 = lsl(r5:4,r10)
+  r15:14 = lsl(r15:14,r10)
+ }
+ {
+  p0 = cmp.gtu(r5:4,r3:2)
+  loop0(1f,r11)
+ }
+ {
+  if (p0) jumpr r31
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r13:12,r3:2)
+ }
+ {
+  r7:6 = sub(r3:2, r13:12)
+  r9:8 = add(r1:0, r15:14)
+ }
+ {
+  r1:0 = vmux(p0, r1:0, r9:8)
+  r3:2 = vmux(p0, r3:2, r7:6)
+ }
+ {
+  r15:14 = lsr(r15:14, #1)
+  r13:12 = lsr(r13:12, #1)
+ }:endloop0
+ {
+  jumpr r31
+ }
+FUNCTION_END __hexagon_udivdi3
+
+  .globl __qdsp_udivdi3
+  .set __qdsp_udivdi3, __hexagon_udivdi3
diff --git a/src/hexagon/udivmoddi4.s b/src/hexagon/udivmoddi4.s
new file mode 100644
index 00000000..cbfb3987
--- /dev/null
+++ b/src/hexagon/udivmoddi4.s
@@ -0,0 +1,50 @@
+
+
+FUNCTION_BEGIN __hexagon_udivmoddi4
+ {
+  r6 = cl0(r1:0)
+  r7 = cl0(r3:2)
+  r5:4 = r3:2
+  r3:2 = r1:0
+ }
+ {
+  r10 = sub(r7,r6)
+  r1:0 = #0
+  r15:14 = #1
+ }
+ {
+  r11 = add(r10,#1)
+  r13:12 = lsl(r5:4,r10)
+  r15:14 = lsl(r15:14,r10)
+ }
+ {
+  p0 = cmp.gtu(r5:4,r3:2)
+  loop0(1f,r11)
+ }
+ {
+  if (p0) jumpr r31
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r13:12,r3:2)
+ }
+ {
+  r7:6 = sub(r3:2, r13:12)
+  r9:8 = add(r1:0, r15:14)
+ }
+ {
+  r1:0 = vmux(p0, r1:0, r9:8)
+  r3:2 = vmux(p0, r3:2, r7:6)
+ }
+ {
+  r15:14 = lsr(r15:14, #1)
+  r13:12 = lsr(r13:12, #1)
+ }:endloop0
+ {
+  jumpr r31
+ }
+FUNCTION_END __hexagon_udivmoddi4
+
+  .globl __qdsp_udivmoddi4
+  .set __qdsp_udivmoddi4, __hexagon_udivmoddi4
diff --git a/src/hexagon/udivmodsi4.s b/src/hexagon/udivmodsi4.s
new file mode 100644
index 00000000..83489c51
--- /dev/null
+++ b/src/hexagon/udivmodsi4.s
@@ -0,0 +1,39 @@
+
+
+FUNCTION_BEGIN __hexagon_udivmodsi4
+ {
+  r2 = cl0(r0)
+  r3 = cl0(r1)
+  r5:4 = combine(#1,#0)
+  p0 = cmp.gtu(r1,r0)
+ }
+ {
+  r6 = sub(r3,r2)
+  r4 = r1
+  r1:0 = combine(r0,r4)
+  if (p0) jumpr r31
+ }
+ {
+  r3:2 = vlslw(r5:4,r6)
+  loop0(1f,r6)
+  p0 = cmp.eq(r6,#0)
+  if (p0.new) r4 = #0
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r2,r1)
+  if (!p0.new) r1 = sub(r1,r2)
+  if (!p0.new) r0 = add(r0,r3)
+  r3:2 = vlsrw(r3:2,#1)
+ }:endloop0
+ {
+  p0 = cmp.gtu(r2,r1)
+  if (!p0.new) r1 = sub(r1,r4)
+  if (!p0.new) r0 = add(r0,r3)
+  jumpr r31
+ }
+FUNCTION_END __hexagon_udivmodsi4
+
+  .globl __qdsp_udivmodsi4
+  .set __qdsp_udivmodsi4, __hexagon_udivmodsi4
diff --git a/src/hexagon/udivsi3.s b/src/hexagon/udivsi3.s
new file mode 100644
index 00000000..e0b94aa9
--- /dev/null
+++ b/src/hexagon/udivsi3.s
@@ -0,0 +1,36 @@
+
+
+FUNCTION_BEGIN __hexagon_udivsi3
+ {
+  r2 = cl0(r0)
+  r3 = cl0(r1)
+  r5:4 = combine(#1,#0)
+  p0 = cmp.gtu(r1,r0)
+ }
+ {
+  r6 = sub(r3,r2)
+  r4 = r1
+  r1:0 = combine(r0,r4)
+  if (p0) jumpr r31
+ }
+ {
+  r3:2 = vlslw(r5:4,r6)
+  loop0(1f,r6)
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r2,r1)
+  if (!p0.new) r1 = sub(r1,r2)
+  if (!p0.new) r0 = add(r0,r3)
+  r3:2 = vlsrw(r3:2,#1)
+ }:endloop0
+ {
+  p0 = cmp.gtu(r2,r1)
+  if (!p0.new) r0 = add(r0,r3)
+  jumpr r31
+ }
+FUNCTION_END __hexagon_udivsi3
+
+  .globl __qdsp_udivsi3
+  .set __qdsp_udivsi3, __hexagon_udivsi3
diff --git a/src/hexagon/umoddi3.s b/src/hexagon/umoddi3.s
new file mode 100644
index 00000000..c76011c3
--- /dev/null
+++ b/src/hexagon/umoddi3.s
@@ -0,0 +1,53 @@
+
+
+FUNCTION_BEGIN __hexagon_umoddi3
+ {
+  r6 = cl0(r1:0)
+  r7 = cl0(r3:2)
+  r5:4 = r3:2
+  r3:2 = r1:0
+ }
+ {
+  r10 = sub(r7,r6)
+  r1:0 = #0
+  r15:14 = #1
+ }
+ {
+  r11 = add(r10,#1)
+  r13:12 = lsl(r5:4,r10)
+  r15:14 = lsl(r15:14,r10)
+ }
+ {
+  p0 = cmp.gtu(r5:4,r3:2)
+  loop0(1f,r11)
+ }
+ {
+  if (p0) jump .hexagon_umoddi3_return
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r13:12,r3:2)
+ }
+ {
+  r7:6 = sub(r3:2, r13:12)
+  r9:8 = add(r1:0, r15:14)
+ }
+ {
+  r1:0 = vmux(p0, r1:0, r9:8)
+  r3:2 = vmux(p0, r3:2, r7:6)
+ }
+ {
+  r15:14 = lsr(r15:14, #1)
+  r13:12 = lsr(r13:12, #1)
+ }:endloop0
+
+.hexagon_umoddi3_return:
+ {
+  r1:0 = r3:2
+  jumpr r31
+ }
+FUNCTION_END __hexagon_umoddi3
+
+  .globl __qdsp_umoddi3
+  .set __qdsp_umoddi3, __hexagon_umoddi3
diff --git a/src/hexagon/umodsi3.s b/src/hexagon/umodsi3.s
new file mode 100644
index 00000000..1b592a7c
--- /dev/null
+++ b/src/hexagon/umodsi3.s
@@ -0,0 +1,34 @@
+
+
+FUNCTION_BEGIN __hexagon_umodsi3
+ {
+  r2 = cl0(r0)
+  r3 = cl0(r1)
+  p0 = cmp.gtu(r1,r0)
+ }
+ {
+  r2 = sub(r3,r2)
+  if (p0) jumpr r31
+ }
+ {
+  loop0(1f,r2)
+  p1 = cmp.eq(r2,#0)
+  r2 = lsl(r1,r2)
+ }
+ .falign
+1:
+ {
+  p0 = cmp.gtu(r2,r0)
+  if (!p0.new) r0 = sub(r0,r2)
+  r2 = lsr(r2,#1)
+  if (p1) r1 = #0
+ }:endloop0
+ {
+  p0 = cmp.gtu(r2,r0)
+  if (!p0.new) r0 = sub(r0,r1)
+  jumpr r31
+ }
+FUNCTION_END __hexagon_umodsi3
+
+  .globl __qdsp_umodsi3
+  .set __qdsp_umodsi3, __hexagon_umodsi3
diff --git a/src/lib.rs b/src/lib.rs
index 3e549187..47aef540 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -1,6 +1,7 @@
 #![cfg_attr(feature = "compiler-builtins", compiler_builtins)]
 #![cfg_attr(not(feature = "no-asm"), feature(asm))]
 #![feature(abi_unadjusted)]
+#![feature(asm_experimental_arch)]
 #![cfg_attr(not(feature = "no-asm"), feature(global_asm))]
 #![feature(cfg_target_has_atomic)]
 #![feature(compiler_builtins)]
@@ -70,6 +71,9 @@ pub mod aarch64_linux;
 ))]
 pub mod arm_linux;
 
+#[cfg(target_arch = "hexagon")]
+pub mod hexagon;
+
 #[cfg(any(target_arch = "riscv32", target_arch = "riscv64"))]
 pub mod riscv;