aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added avx2-i64x4 and avx1.1-i64x4 targets

Browse Source
This commit is contained in:
egaburov
2013-10-15 10:02:10 +02:00
parent ba4977cd84
commit 7e9b4c0924
5 changed files with 556 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
ispc.cpp
View File

@@ -507,6 +507,25 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
#if !defined(LLVM_3_1)
// LLVM 3.2+ only
this->m_hasRand = true;
#endif
}
else if (!strcasecmp(isa, "avx1.1-i64x4")) {
this->m_isa = Target::AVX11;
this->m_nativeVectorWidth = 8; /* native vector width in terms of floats */
this->m_vectorWidth = 4;
this->m_attributes = "+avx,+popcnt,+cmov,+f16c"
#if defined(LLVM_3_4)
",+rdrnd"
#else
",+rdrand"
#endif
;
this->m_maskingIsFree = false;
this->m_maskBitCount = 64;
this->m_hasHalf = true;
#if !defined(LLVM_3_1)
// LLVM 3.2+ only
this->m_hasRand = true;
#endif
}
else if (!strcasecmp(isa, "avx2") ||
@@ -555,6 +574,29 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
// LLVM 3.2+ only
this->m_hasRand = true;
this->m_hasGather = true;
#endif
}
else if (!strcasecmp(isa, "avx2-i64x4")) {
this->m_isa = Target::AVX2;
this->m_nativeVectorWidth = 8; /* native vector width in terms of floats */
this->m_vectorWidth = 4;
this->m_attributes = "+avx2,+popcnt,+cmov,+f16c"
#if defined(LLVM_3_4)
",+rdrnd"
#else
",+rdrand"
#endif
#ifndef LLVM_3_1
",+fma"
#endif // !LLVM_3_1
;
this->m_maskingIsFree = false;
this->m_maskBitCount = 64;
this->m_hasHalf = true;
#if !defined(LLVM_3_1)
// LLVM 3.2+ only
this->m_hasRand = true;
this->m_hasGather = true;
#endif
}
#ifdef ISPC_ARM_ENABLED
@@ -715,8 +757,8 @@ Target::SupportedTargets() {
"sse2-i32x4, sse2-i32x8, "
"sse4-i32x4, sse4-i32x8, sse4-i16x8, sse4-i8x16, "
"avx1-i32x8, avx1-i32x16, avx1-i64x4, "
"avx1.1-i32x8, avx1.1-i32x16, "
"avx2-i32x8, avx2-i32x16, "
"avx1.1-i32x8, avx1.1-i32x16, avx1.1-i64x4 "
"avx2-i32x8, avx2-i32x16, avx2-i64x4, "
"generic-x1, generic-x4, generic-x8, generic-x16, "
"generic-x32, generic-x64";
}