integer/clz: Use clang builtin instead of llvm asm

The generated llvm IR mostly identical. char/uchar case is a bit worse.

reviewer: Tom Stellard

Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 314701

diff --git a/libclc/generic/lib/SOURCES b/libclc/generic/lib/SOURCES
index f919bc7..aa7f4aa 100644 (file)
--- a/libclc/generic/lib/SOURCES
+++ b/libclc/generic/lib/SOURCES
@@ -67,8 +67,6 @@ integer/add_sat.cl
 integer/add_sat_if.ll
 integer/add_sat_impl.ll
 integer/clz.cl
-integer/clz_if.ll
-integer/clz_impl.ll
 integer/hadd.cl
 integer/mad24.cl
 integer/mad_sat.cl

diff --git a/libclc/generic/lib/integer/clz.cl b/libclc/generic/lib/integer/clz.cl
index 17e3fe0..e2080b5 100644 (file)
--- a/libclc/generic/lib/integer/clz.cl
+++ b/libclc/generic/lib/integer/clz.cl
@@ -1,46 +1,36 @@
 #include <clc/clc.h>
 #include "../clcmacro.h"
 
-// From clz.ll
-_CLC_DECL char   __clc_clz_s8(char);
-_CLC_DECL uchar  __clc_clz_u8(uchar);
-_CLC_DECL short  __clc_clz_s16(short);
-_CLC_DECL ushort __clc_clz_u16(ushort);
-_CLC_DECL int    __clc_clz_s32(int);
-_CLC_DECL uint   __clc_clz_u32(uint);
-_CLC_DECL long   __clc_clz_s64(long);
-_CLC_DECL ulong  __clc_clz_u64(ulong);
-
 _CLC_OVERLOAD _CLC_DEF char clz(char x) {
-  return __clc_clz_s8(x);
+  return clz((ushort)(uchar)x) - 8;
 }
 
 _CLC_OVERLOAD _CLC_DEF uchar clz(uchar x) {
-  return __clc_clz_u8(x);
+  return clz((ushort)x) - 8;
 }
 
 _CLC_OVERLOAD _CLC_DEF short clz(short x) {
-  return __clc_clz_s16(x);
+  return x ? __builtin_clzs(x) : 16;
 }
 
 _CLC_OVERLOAD _CLC_DEF ushort clz(ushort x) {
-  return __clc_clz_u16(x);
+  return x ? __builtin_clzs(x) : 16;
 }
 
 _CLC_OVERLOAD _CLC_DEF int clz(int x) {
-  return __clc_clz_s32(x);
+  return x ? __builtin_clz(x) : 32;
 }
 
 _CLC_OVERLOAD _CLC_DEF uint clz(uint x) {
-  return __clc_clz_u32(x);
+  return x ? __builtin_clz(x) : 32;
 }
 
 _CLC_OVERLOAD _CLC_DEF long clz(long x) {
-  return __clc_clz_s64(x);
+  return x ? __builtin_clzl(x) : 64;
 }
 
 _CLC_OVERLOAD _CLC_DEF ulong clz(ulong x) {
-  return __clc_clz_u64(x);
+  return x ? __builtin_clzl(x) : 64;
 }
 
 _CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, char, clz, char)

diff --git a/libclc/generic/lib/integer/clz_if.ll b/libclc/generic/lib/integer/clz_if.ll
deleted file mode 100644 (file)
index 23dfc74..0000000
--- a/libclc/generic/lib/integer/clz_if.ll
+++ /dev/null
@@ -1,55 +0,0 @@
-declare i8 @__clc_clz_impl_s8(i8 %x)
-
-define i8 @__clc_clz_s8(i8 %x) nounwind readnone alwaysinline {
-  %call = call i8 @__clc_clz_impl_s8(i8 %x)
-  ret i8 %call
-}
-
-declare i8 @__clc_clz_impl_u8(i8 %x)
-
-define i8 @__clc_clz_u8(i8 %x) nounwind readnone alwaysinline {
-  %call = call i8 @__clc_clz_impl_u8(i8 %x)
-  ret i8 %call
-}
-
-declare i16 @__clc_clz_impl_s16(i16 %x)
-
-define i16 @__clc_clz_s16(i16 %x) nounwind readnone alwaysinline {
-  %call = call i16 @__clc_clz_impl_s16(i16 %x)
-  ret i16 %call
-}
-
-declare i16 @__clc_clz_impl_u16(i16 %x)
-
-define i16 @__clc_clz_u16(i16 %x) nounwind readnone alwaysinline {
-  %call = call i16 @__clc_clz_impl_u16(i16 %x)
-  ret i16 %call
-}
-
-declare i32 @__clc_clz_impl_s32(i32 %x)
-
-define i32 @__clc_clz_s32(i32 %x) nounwind readnone alwaysinline {
-  %call = call i32 @__clc_clz_impl_s32(i32 %x)
-  ret i32 %call
-}
-
-declare i32 @__clc_clz_impl_u32(i32 %x)
-
-define i32 @__clc_clz_u32(i32 %x) nounwind readnone alwaysinline {
-  %call = call i32 @__clc_clz_impl_u32(i32 %x)
-  ret i32 %call
-}
-
-declare i64 @__clc_clz_impl_s64(i64 %x)
-
-define i64 @__clc_clz_s64(i64 %x) nounwind readnone alwaysinline {
-  %call = call i64 @__clc_clz_impl_s64(i64 %x)
-  ret i64 %call
-}
-
-declare i64 @__clc_clz_impl_u64(i64 %x)
-
-define i64 @__clc_clz_u64(i64 %x) nounwind readnone alwaysinline {
-  %call = call i64 @__clc_clz_impl_u64(i64 %x)
-  ret i64 %call
-}

diff --git a/libclc/generic/lib/integer/clz_impl.ll b/libclc/generic/lib/integer/clz_impl.ll
deleted file mode 100644 (file)
index b5c3d98..0000000
--- a/libclc/generic/lib/integer/clz_impl.ll
+++ /dev/null
@@ -1,44 +0,0 @@
-declare i8 @llvm.ctlz.i8(i8, i1)
-declare i16 @llvm.ctlz.i16(i16, i1)
-declare i32 @llvm.ctlz.i32(i32, i1)
-declare i64 @llvm.ctlz.i64(i64, i1)
-
-define i8 @__clc_clz_impl_s8(i8 %x) nounwind readnone alwaysinline {
-  %call = call i8 @llvm.ctlz.i8(i8 %x, i1 0)
-  ret i8 %call
-}
-
-define i8 @__clc_clz_impl_u8(i8 %x) nounwind readnone alwaysinline {
-  %call = call i8 @llvm.ctlz.i8(i8 %x, i1 0)
-  ret i8 %call
-}
-
-define i16 @__clc_clz_impl_s16(i16 %x) nounwind readnone alwaysinline {
-  %call = call i16 @llvm.ctlz.i16(i16 %x, i1 0)
-  ret i16 %call
-}
-
-define i16 @__clc_clz_impl_u16(i16 %x) nounwind readnone alwaysinline {
-  %call = call i16 @llvm.ctlz.i16(i16 %x, i1 0)
-  ret i16 %call
-}
-
-define i32 @__clc_clz_impl_s32(i32 %x) nounwind readnone alwaysinline {
-  %call = call i32 @llvm.ctlz.i32(i32 %x, i1 0)
-  ret i32 %call
-}
-
-define i32 @__clc_clz_impl_u32(i32 %x) nounwind readnone alwaysinline {
-  %call = call i32 @llvm.ctlz.i32(i32 %x, i1 0)
-  ret i32 %call
-}
-
-define i64 @__clc_clz_impl_s64(i64 %x) nounwind readnone alwaysinline {
-  %call = call i64 @llvm.ctlz.i64(i64 %x, i1 0)
-  ret i64 %call
-}
-
-define i64 @__clc_clz_impl_u64(i64 %x) nounwind readnone alwaysinline {
-  %call = call i64 @llvm.ctlz.i64(i64 %x, i1 0)
-  ret i64 %call
-}