Implement sub_sat builtin. Patch by Lei Mou!

git-svn-id: https://llvm.org/svn/llvm-project/libclc/trunk@161312 91177308-0d34-0410-b5e6-96231b3b80d8

10 files changed, 272 insertions, 0 deletions

diff --git a/generic/include/clc/clc.h b/generic/include/clc/clc.h
index b0cbd4a..8b41523 100644
--- a/generic/include/clc/clc.h
+++ b/generic/include/clc/clc.h
@@ -57,6 +57,7 @@
 #include <clc/integer/abs.h>
 #include <clc/integer/abs_diff.h>
 #include <clc/integer/add_sat.h>
+#include <clc/integer/sub_sat.h>
 
 /* 6.11.5 Geometric Functions */
 #include <clc/geometric/cross.h>
diff --git a/generic/include/clc/integer/sub_sat.h b/generic/include/clc/integer/sub_sat.h
new file mode 100644
index 0000000..942274d
--- /dev/null
+++ b/generic/include/clc/integer/sub_sat.h
@@ -0,0 +1,2 @@
+#define BODY <clc/integer/sub_sat.inc>
+#include <clc/integer/gentype.inc>
diff --git a/generic/include/clc/integer/sub_sat.inc b/generic/include/clc/integer/sub_sat.inc
new file mode 100644
index 0000000..3e0f8f9
--- /dev/null
+++ b/generic/include/clc/integer/sub_sat.inc
@@ -0,0 +1 @@
+_CLC_OVERLOAD _CLC_DECL GENTYPE sub_sat(GENTYPE x, GENTYPE y);
diff --git a/generic/lib/SOURCES b/generic/lib/SOURCES
index 0608116..344c865 100644
--- a/generic/lib/SOURCES
+++ b/generic/lib/SOURCES
@@ -7,5 +7,8 @@ integer/abs.cl
 integer/add_sat.cl
 integer/add_sat.ll
 integer/add_sat_impl.ll
+integer/sub_sat.cl
+integer/sub_sat.ll
+integer/sub_sat_impl.ll
 math/hypot.cl
 math/mad.cl
diff --git a/generic/lib/integer/sub_sat.cl b/generic/lib/integer/sub_sat.cl
new file mode 100644
index 0000000..9555b6d
--- /dev/null
+++ b/generic/lib/integer/sub_sat.cl
@@ -0,0 +1,52 @@
+#include <clc/clc.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);
+}
+
+_CLC_OVERLOAD _CLC_DEF uchar sub_sat(uchar x, uchar y) {
+  return __clc_sub_sat_u8(x, y);
+}
+
+_CLC_OVERLOAD _CLC_DEF short sub_sat(short x, short y) {
+  return __clc_sub_sat_s16(x, y);
+}
+
+_CLC_OVERLOAD _CLC_DEF ushort sub_sat(ushort x, ushort y) {
+  return __clc_sub_sat_u16(x, y);
+}
+
+_CLC_OVERLOAD _CLC_DEF int sub_sat(int x, int y) {
+  return __clc_sub_sat_s32(x, y);
+}
+
+_CLC_OVERLOAD _CLC_DEF uint sub_sat(uint x, uint y) {
+  return __clc_sub_sat_u32(x, y);
+}
+
+_CLC_OVERLOAD _CLC_DEF long sub_sat(long x, long y) {
+  return __clc_sub_sat_s64(x, y);
+}
+
+_CLC_OVERLOAD _CLC_DEF ulong sub_sat(ulong x, ulong y) {
+  return __clc_sub_sat_u64(x, y);
+}
+
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, char, sub_sat, char, char)
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, uchar, sub_sat, uchar, uchar)
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, short, sub_sat, short, short)
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, ushort, sub_sat, ushort, ushort)
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, int, sub_sat, int, int)
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, uint, sub_sat, uint, uint)
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, long, sub_sat, long, long)
+_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, ulong, sub_sat, ulong, ulong)
diff --git a/generic/lib/integer/sub_sat.ll b/generic/lib/integer/sub_sat.ll
new file mode 100644
index 0000000..f223a0e
--- /dev/null
+++ b/generic/lib/integer/sub_sat.ll
@@ -0,0 +1,55 @@
+declare i8 @__clc_sub_sat_impl_s8(i8 %x, i8 %y)
+
+define linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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/generic/lib/integer/sub_sat_impl.ll b/generic/lib/integer/sub_sat_impl.ll
new file mode 100644
index 0000000..99abbc3
--- /dev/null
+++ b/generic/lib/integer/sub_sat_impl.ll
@@ -0,0 +1,83 @@
+declare {i8, i1} @llvm.ssub.with.overflow.i8(i8, i8)
+declare {i8, i1} @llvm.usub.with.overflow.i8(i8, i8)
+
+define linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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/ptx/lib/SOURCES b/ptx/lib/SOURCES
index aab8e3f..fb6e17f 100644
--- a/ptx/lib/SOURCES
+++ b/ptx/lib/SOURCES
@@ -1 +1,2 @@
 integer/add_sat.ll
+integer/sub_sat.ll
\ No newline at end of file
diff --git a/ptx/lib/integer/sub_sat.ll b/ptx/lib/integer/sub_sat.ll
new file mode 100644
index 0000000..6a51a52
--- /dev/null
+++ b/ptx/lib/integer/sub_sat.ll
@@ -0,0 +1,55 @@
+declare i8 @__clc_sub_sat_impl_s8(i8 %x, i8 %y)
+
+define linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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 linkonce_odr 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
+}
diff --git a/test/subsat.cl b/test/subsat.cl
new file mode 100644
index 0000000..a83414b
--- /dev/null
+++ b/test/subsat.cl
@@ -0,0 +1,19 @@
+__kernel void test_subsat_char(char *a, char x, char y) {
+  *a = sub_sat(x, y);
+  return;
+}
+
+__kernel void test_subsat_uchar(uchar *a, uchar x, uchar y) {
+  *a = sub_sat(x, y);
+  return;
+}
+
+__kernel void test_subsat_long(long *a, long x, long y) {
+  *a = sub_sat(x, y);
+  return;
+}
+
+__kernel void test_subsat_ulong(ulong *a, ulong x, ulong y) {
+  *a = sub_sat(x, y);
+  return;
+}
\ No newline at end of file