diff --git a/libclc/generic/lib/SOURCES b/libclc/generic/lib/SOURCES
index c3cc350140ee..6457c250a089 100644
--- a/libclc/generic/lib/SOURCES
+++ b/libclc/generic/lib/SOURCES
@@ -73,8 +73,6 @@ integer/mul_hi.cl
 integer/rhadd.cl
 integer/rotate.cl
 integer/sub_sat.cl
-integer/sub_sat_if.ll
-integer/sub_sat_impl.ll
 integer/upsample.cl
 math/acos.cl
 math/acosh.cl
diff --git a/libclc/generic/lib/integer/sub_sat.cl b/libclc/generic/lib/integer/sub_sat.cl
index 6b42cc86a74c..2fbc31664e71 100644
--- a/libclc/generic/lib/integer/sub_sat.cl
+++ b/libclc/generic/lib/integer/sub_sat.cl
@@ -1,46 +1,54 @@
 #include <clc/clc.h>
 #include "../clcmacro.h"
 
-// From sub_sat.ll
-_CLC_DECL char   __clc_sub_sat_s8(char, char);
-_CLC_DECL uchar  __clc_sub_sat_u8(uchar, uchar);
-_CLC_DECL short  __clc_sub_sat_s16(short, short);
-_CLC_DECL ushort __clc_sub_sat_u16(ushort, ushort);
-_CLC_DECL int    __clc_sub_sat_s32(int, int);
-_CLC_DECL uint   __clc_sub_sat_u32(uint, uint);
-_CLC_DECL long   __clc_sub_sat_s64(long, long);
-_CLC_DECL ulong  __clc_sub_sat_u64(ulong, ulong);
-
 _CLC_OVERLOAD _CLC_DEF char sub_sat(char x, char y) {
-  return __clc_sub_sat_s8(x, y);
+  short r = x - y;
+  return convert_char_sat(r);
 }
 
 _CLC_OVERLOAD _CLC_DEF uchar sub_sat(uchar x, uchar y) {
-  return __clc_sub_sat_u8(x, y);
+  short r = x - y;
+  return convert_uchar_sat(r);
 }
 
 _CLC_OVERLOAD _CLC_DEF short sub_sat(short x, short y) {
-  return __clc_sub_sat_s16(x, y);
+  int r = x - y;
+  return convert_short_sat(r);
 }
 
 _CLC_OVERLOAD _CLC_DEF ushort sub_sat(ushort x, ushort y) {
-  return __clc_sub_sat_u16(x, y);
+  int r = x - y;
+  return convert_ushort_sat(r);
 }
 
 _CLC_OVERLOAD _CLC_DEF int sub_sat(int x, int y) {
-  return __clc_sub_sat_s32(x, y);
+  int r;
+  if (__builtin_ssub_overflow(x, y, &r))
+    // The oveflow can only occur in the direction of the first operand
+    return x > 0 ? INT_MAX : INT_MIN;
+  return r;
 }
 
 _CLC_OVERLOAD _CLC_DEF uint sub_sat(uint x, uint y) {
-  return __clc_sub_sat_u32(x, y);
+  uint r;
+  if (__builtin_usub_overflow(x, y, &r))
+	return 0;
+  return r;
 }
 
 _CLC_OVERLOAD _CLC_DEF long sub_sat(long x, long y) {
-  return __clc_sub_sat_s64(x, y);
+  long r;
+  if (__builtin_ssubl_overflow(x, y, &r))
+    // The oveflow can only occur in the direction of the first operand
+    return x > 0 ? LONG_MAX : LONG_MIN;
+  return r;
 }
 
 _CLC_OVERLOAD _CLC_DEF ulong sub_sat(ulong x, ulong y) {
-  return __clc_sub_sat_u64(x, y);
+  ulong r;
+  if (__builtin_usubl_overflow(x, y, &r))
+	return 0;
+  return r;
 }
 
 _CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, char, sub_sat, char, char)
diff --git a/libclc/generic/lib/integer/sub_sat_if.ll b/libclc/generic/lib/integer/sub_sat_if.ll
deleted file mode 100644
index 7252574b5b8e..000000000000
--- a/libclc/generic/lib/integer/sub_sat_if.ll
+++ /dev/null
@@ -1,55 +0,0 @@
-declare i8 @__clc_sub_sat_impl_s8(i8 %x, i8 %y)
-
-define i8 @__clc_sub_sat_s8(i8 %x, i8 %y) nounwind readnone alwaysinline {
-  %call = call i8 @__clc_sub_sat_impl_s8(i8 %x, i8 %y)
-  ret i8 %call
-}
-
-declare i8 @__clc_sub_sat_impl_u8(i8 %x, i8 %y)
-
-define i8 @__clc_sub_sat_u8(i8 %x, i8 %y) nounwind readnone alwaysinline {
-  %call = call i8 @__clc_sub_sat_impl_u8(i8 %x, i8 %y)
-  ret i8 %call
-}
-
-declare i16 @__clc_sub_sat_impl_s16(i16 %x, i16 %y)
-
-define i16 @__clc_sub_sat_s16(i16 %x, i16 %y) nounwind readnone alwaysinline {
-  %call = call i16 @__clc_sub_sat_impl_s16(i16 %x, i16 %y)
-  ret i16 %call
-}
-
-declare i16 @__clc_sub_sat_impl_u16(i16 %x, i16 %y)
-
-define i16 @__clc_sub_sat_u16(i16 %x, i16 %y) nounwind readnone alwaysinline {
-  %call = call i16 @__clc_sub_sat_impl_u16(i16 %x, i16 %y)
-  ret i16 %call
-}
-
-declare i32 @__clc_sub_sat_impl_s32(i32 %x, i32 %y)
-
-define i32 @__clc_sub_sat_s32(i32 %x, i32 %y) nounwind readnone alwaysinline {
-  %call = call i32 @__clc_sub_sat_impl_s32(i32 %x, i32 %y)
-  ret i32 %call
-}
-
-declare i32 @__clc_sub_sat_impl_u32(i32 %x, i32 %y)
-
-define i32 @__clc_sub_sat_u32(i32 %x, i32 %y) nounwind readnone alwaysinline {
-  %call = call i32 @__clc_sub_sat_impl_u32(i32 %x, i32 %y)
-  ret i32 %call
-}
-
-declare i64 @__clc_sub_sat_impl_s64(i64 %x, i64 %y)
-
-define i64 @__clc_sub_sat_s64(i64 %x, i64 %y) nounwind readnone alwaysinline {
-  %call = call i64 @__clc_sub_sat_impl_s64(i64 %x, i64 %y)
-  ret i64 %call
-}
-
-declare i64 @__clc_sub_sat_impl_u64(i64 %x, i64 %y)
-
-define i64 @__clc_sub_sat_u64(i64 %x, i64 %y) nounwind readnone alwaysinline {
-  %call = call i64 @__clc_sub_sat_impl_u64(i64 %x, i64 %y)
-  ret i64 %call
-}
diff --git a/libclc/generic/lib/integer/sub_sat_impl.ll b/libclc/generic/lib/integer/sub_sat_impl.ll
deleted file mode 100644
index e82b632f43b4..000000000000
--- a/libclc/generic/lib/integer/sub_sat_impl.ll
+++ /dev/null
@@ -1,83 +0,0 @@
-declare {i8, i1} @llvm.ssub.with.overflow.i8(i8, i8)
-declare {i8, i1} @llvm.usub.with.overflow.i8(i8, i8)
-
-define i8 @__clc_sub_sat_impl_s8(i8 %x, i8 %y) nounwind readnone alwaysinline {
-  %call = call {i8, i1} @llvm.ssub.with.overflow.i8(i8 %x, i8 %y)
-  %res = extractvalue {i8, i1} %call, 0
-  %over = extractvalue {i8, i1} %call, 1
-  %x.msb = ashr i8 %x, 7
-  %x.limit = xor i8 %x.msb, 127
-  %sat = select i1 %over, i8 %x.limit, i8 %res
-  ret i8 %sat
-}
-
-define i8 @__clc_sub_sat_impl_u8(i8 %x, i8 %y) nounwind readnone alwaysinline {
-  %call = call {i8, i1} @llvm.usub.with.overflow.i8(i8 %x, i8 %y)
-  %res = extractvalue {i8, i1} %call, 0
-  %over = extractvalue {i8, i1} %call, 1
-  %sat = select i1 %over, i8 0, i8 %res
-  ret i8 %sat
-}
-
-declare {i16, i1} @llvm.ssub.with.overflow.i16(i16, i16)
-declare {i16, i1} @llvm.usub.with.overflow.i16(i16, i16)
-
-define i16 @__clc_sub_sat_impl_s16(i16 %x, i16 %y) nounwind readnone alwaysinline {
-  %call = call {i16, i1} @llvm.ssub.with.overflow.i16(i16 %x, i16 %y)
-  %res = extractvalue {i16, i1} %call, 0
-  %over = extractvalue {i16, i1} %call, 1
-  %x.msb = ashr i16 %x, 15
-  %x.limit = xor i16 %x.msb, 32767
-  %sat = select i1 %over, i16 %x.limit, i16 %res
-  ret i16 %sat
-}
-
-define i16 @__clc_sub_sat_impl_u16(i16 %x, i16 %y) nounwind readnone alwaysinline {
-  %call = call {i16, i1} @llvm.usub.with.overflow.i16(i16 %x, i16 %y)
-  %res = extractvalue {i16, i1} %call, 0
-  %over = extractvalue {i16, i1} %call, 1
-  %sat = select i1 %over, i16 0, i16 %res
-  ret i16 %sat
-}
-
-declare {i32, i1} @llvm.ssub.with.overflow.i32(i32, i32)
-declare {i32, i1} @llvm.usub.with.overflow.i32(i32, i32)
-
-define i32 @__clc_sub_sat_impl_s32(i32 %x, i32 %y) nounwind readnone alwaysinline {
-  %call = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %x, i32 %y)
-  %res = extractvalue {i32, i1} %call, 0
-  %over = extractvalue {i32, i1} %call, 1
-  %x.msb = ashr i32 %x, 31
-  %x.limit = xor i32 %x.msb, 2147483647
-  %sat = select i1 %over, i32 %x.limit, i32 %res
-  ret i32 %sat
-}
-
-define i32 @__clc_sub_sat_impl_u32(i32 %x, i32 %y) nounwind readnone alwaysinline {
-  %call = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %x, i32 %y)
-  %res = extractvalue {i32, i1} %call, 0
-  %over = extractvalue {i32, i1} %call, 1
-  %sat = select i1 %over, i32 0, i32 %res
-  ret i32 %sat
-}
-
-declare {i64, i1} @llvm.ssub.with.overflow.i64(i64, i64)
-declare {i64, i1} @llvm.usub.with.overflow.i64(i64, i64)
-
-define i64 @__clc_sub_sat_impl_s64(i64 %x, i64 %y) nounwind readnone alwaysinline {
-  %call = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %x, i64 %y)
-  %res = extractvalue {i64, i1} %call, 0
-  %over = extractvalue {i64, i1} %call, 1
-  %x.msb = ashr i64 %x, 63
-  %x.limit = xor i64 %x.msb, 9223372036854775807
-  %sat = select i1 %over, i64 %x.limit, i64 %res
-  ret i64 %sat
-}
-
-define i64 @__clc_sub_sat_impl_u64(i64 %x, i64 %y) nounwind readnone alwaysinline {
-  %call = call {i64, i1} @llvm.usub.with.overflow.i64(i64 %x, i64 %y)
-  %res = extractvalue {i64, i1} %call, 0
-  %over = extractvalue {i64, i1} %call, 1
-  %sat = select i1 %over, i64 0, i64 %res
-  ret i64 %sat
-}
diff --git a/libclc/ptx/lib/OVERRIDES b/libclc/ptx/lib/OVERRIDES
deleted file mode 100644
index 565dcca63b5a..000000000000
--- a/libclc/ptx/lib/OVERRIDES
+++ /dev/null
@@ -1 +0,0 @@
-integer/sub_sat_if.ll
diff --git a/libclc/ptx/lib/SOURCES b/libclc/ptx/lib/SOURCES
deleted file mode 100644
index 913e2ea737b3..000000000000
--- a/libclc/ptx/lib/SOURCES
+++ /dev/null
@@ -1 +0,0 @@
-integer/sub_sat.ll
diff --git a/libclc/ptx/lib/integer/sub_sat.ll b/libclc/ptx/lib/integer/sub_sat.ll
deleted file mode 100644
index 1a66eb566b52..000000000000
--- a/libclc/ptx/lib/integer/sub_sat.ll
+++ /dev/null
@@ -1,55 +0,0 @@
-declare i8 @__clc_sub_sat_impl_s8(i8 %x, i8 %y)
-
-define ptx_device i8 @__clc_sub_sat_s8(i8 %x, i8 %y) nounwind readnone alwaysinline {
-  %call = call i8 @__clc_sub_sat_impl_s8(i8 %x, i8 %y)
-  ret i8 %call
-}
-
-declare i8 @__clc_sub_sat_impl_u8(i8 %x, i8 %y)
-
-define ptx_device i8 @__clc_sub_sat_u8(i8 %x, i8 %y) nounwind readnone alwaysinline {
-  %call = call i8 @__clc_sub_sat_impl_u8(i8 %x, i8 %y)
-  ret i8 %call
-}
-
-declare i16 @__clc_sub_sat_impl_s16(i16 %x, i16 %y)
-
-define ptx_device i16 @__clc_sub_sat_s16(i16 %x, i16 %y) nounwind readnone alwaysinline {
-  %call = call i16 @__clc_sub_sat_impl_s16(i16 %x, i16 %y)
-  ret i16 %call
-}
-
-declare i16 @__clc_sub_sat_impl_u16(i16 %x, i16 %y)
-
-define ptx_device i16 @__clc_sub_sat_u16(i16 %x, i16 %y) nounwind readnone alwaysinline {
-  %call = call i16 @__clc_sub_sat_impl_u16(i16 %x, i16 %y)
-  ret i16 %call
-}
-
-declare i32 @__clc_sub_sat_impl_s32(i32 %x, i32 %y)
-
-define ptx_device i32 @__clc_sub_sat_s32(i32 %x, i32 %y) nounwind readnone alwaysinline {
-  %call = call i32 @__clc_sub_sat_impl_s32(i32 %x, i32 %y)
-  ret i32 %call
-}
-
-declare i32 @__clc_sub_sat_impl_u32(i32 %x, i32 %y)
-
-define ptx_device i32 @__clc_sub_sat_u32(i32 %x, i32 %y) nounwind readnone alwaysinline {
-  %call = call i32 @__clc_sub_sat_impl_u32(i32 %x, i32 %y)
-  ret i32 %call
-}
-
-declare i64 @__clc_sub_sat_impl_s64(i64 %x, i64 %y)
-
-define ptx_device i64 @__clc_sub_sat_s64(i64 %x, i64 %y) nounwind readnone alwaysinline {
-  %call = call i64 @__clc_sub_sat_impl_s64(i64 %x, i64 %y)
-  ret i64 %call
-}
-
-declare i64 @__clc_sub_sat_impl_u64(i64 %x, i64 %y)
-
-define ptx_device i64 @__clc_sub_sat_u64(i64 %x, i64 %y) nounwind readnone alwaysinline {
-  %call = call i64 @__clc_sub_sat_impl_u64(i64 %x, i64 %y)
-  ret i64 %call
-}