r224879 - [x86] Add the (v)cmpps/pd/ss/sd builtins to match gcc. Use them in the sse intrinsic files.
Craig Topper
craig.topper at gmail.com
Fri Dec 26 22:59:57 PST 2014
Author: ctopper
Date: Sat Dec 27 00:59:57 2014
New Revision: 224879
URL: http://llvm.org/viewvc/llvm-project?rev=224879&view=rev
Log:
[x86] Add the (v)cmpps/pd/ss/sd builtins to match gcc. Use them in the sse intrinsic files.
This still lower to the same intrinsics as before.
This is preparation for bounds checking the immediate on the avx version of the builtin so we don't pass illegal immediates into the backend. Since SSE uses a smaller size immediate its not possible to bounds check when using a shared builtin. Rather than creating a clang specific builtin for the different immediate, I decided (after consulting with Chandler) that it was better to match gcc.
Modified:
cfe/trunk/include/clang/Basic/BuiltinsX86.def
cfe/trunk/lib/CodeGen/CGBuiltin.cpp
cfe/trunk/lib/Headers/emmintrin.h
cfe/trunk/lib/Headers/xmmintrin.h
cfe/trunk/test/CodeGen/sse-builtins.c
Modified: cfe/trunk/include/clang/Basic/BuiltinsX86.def
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/include/clang/Basic/BuiltinsX86.def?rev=224879&r1=224878&r2=224879&view=diff
==============================================================================
--- cfe/trunk/include/clang/Basic/BuiltinsX86.def (original)
+++ cfe/trunk/include/clang/Basic/BuiltinsX86.def Sat Dec 27 00:59:57 2014
@@ -187,13 +187,45 @@ BUILTIN(__builtin_ia32_ucomisdgt, "iV2dV
BUILTIN(__builtin_ia32_ucomisdge, "iV2dV2d", "")
BUILTIN(__builtin_ia32_ucomisdneq, "iV2dV2d", "")
BUILTIN(__builtin_ia32_cmpps, "V4fV4fV4fIc", "")
+BUILTIN(__builtin_ia32_cmpeqps, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpltps, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpleps, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpunordps, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpneqps, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpnltps, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpnleps, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpordps, "V4fV4fV4f", "")
BUILTIN(__builtin_ia32_cmpss, "V4fV4fV4fIc", "")
+BUILTIN(__builtin_ia32_cmpeqss, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpltss, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpless, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpunordss, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpneqss, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpnltss, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpnless, "V4fV4fV4f", "")
+BUILTIN(__builtin_ia32_cmpordss, "V4fV4fV4f", "")
BUILTIN(__builtin_ia32_minps, "V4fV4fV4f", "")
BUILTIN(__builtin_ia32_maxps, "V4fV4fV4f", "")
BUILTIN(__builtin_ia32_minss, "V4fV4fV4f", "")
BUILTIN(__builtin_ia32_maxss, "V4fV4fV4f", "")
BUILTIN(__builtin_ia32_cmppd, "V2dV2dV2dIc", "")
+BUILTIN(__builtin_ia32_cmpeqpd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpltpd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmplepd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpunordpd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpneqpd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpnltpd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpnlepd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpordpd, "V2dV2dV2d", "")
BUILTIN(__builtin_ia32_cmpsd, "V2dV2dV2dIc", "")
+BUILTIN(__builtin_ia32_cmpeqsd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpltsd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmplesd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpunordsd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpneqsd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpnltsd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpnlesd, "V2dV2dV2d", "")
+BUILTIN(__builtin_ia32_cmpordsd, "V2dV2dV2d", "")
BUILTIN(__builtin_ia32_minpd, "V2dV2dV2d", "")
BUILTIN(__builtin_ia32_maxpd, "V2dV2dV2d", "")
BUILTIN(__builtin_ia32_minsd, "V2dV2dV2d", "")
Modified: cfe/trunk/lib/CodeGen/CGBuiltin.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGBuiltin.cpp?rev=224879&r1=224878&r2=224879&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGBuiltin.cpp (original)
+++ cfe/trunk/lib/CodeGen/CGBuiltin.cpp Sat Dec 27 00:59:57 2014
@@ -5986,8 +5986,8 @@ Value *CodeGenFunction::EmitX86BuiltinEx
// 3DNow!
case X86::BI__builtin_ia32_pswapdsf:
case X86::BI__builtin_ia32_pswapdsi: {
- const char *name = nullptr;
- Intrinsic::ID ID = Intrinsic::not_intrinsic;
+ const char *name;
+ Intrinsic::ID ID;
switch(BuiltinID) {
default: llvm_unreachable("Unsupported intrinsic!");
case X86::BI__builtin_ia32_pswapdsf:
@@ -6041,6 +6041,154 @@ Value *CodeGenFunction::EmitX86BuiltinEx
Value *F = CGM.getIntrinsic(Intrinsic::x86_avx2_vbroadcasti128);
return Builder.CreateCall(F, Builder.CreateBitCast(VecTmp, Int8PtrTy));
}
+ // SSE comparison intrisics
+ case X86::BI__builtin_ia32_cmpeqps:
+ case X86::BI__builtin_ia32_cmpltps:
+ case X86::BI__builtin_ia32_cmpleps:
+ case X86::BI__builtin_ia32_cmpunordps:
+ case X86::BI__builtin_ia32_cmpneqps:
+ case X86::BI__builtin_ia32_cmpnltps:
+ case X86::BI__builtin_ia32_cmpnleps:
+ case X86::BI__builtin_ia32_cmpordps:
+ case X86::BI__builtin_ia32_cmpeqss:
+ case X86::BI__builtin_ia32_cmpltss:
+ case X86::BI__builtin_ia32_cmpless:
+ case X86::BI__builtin_ia32_cmpunordss:
+ case X86::BI__builtin_ia32_cmpneqss:
+ case X86::BI__builtin_ia32_cmpnltss:
+ case X86::BI__builtin_ia32_cmpnless:
+ case X86::BI__builtin_ia32_cmpordss:
+ case X86::BI__builtin_ia32_cmpeqpd:
+ case X86::BI__builtin_ia32_cmpltpd:
+ case X86::BI__builtin_ia32_cmplepd:
+ case X86::BI__builtin_ia32_cmpunordpd:
+ case X86::BI__builtin_ia32_cmpneqpd:
+ case X86::BI__builtin_ia32_cmpnltpd:
+ case X86::BI__builtin_ia32_cmpnlepd:
+ case X86::BI__builtin_ia32_cmpordpd:
+ case X86::BI__builtin_ia32_cmpeqsd:
+ case X86::BI__builtin_ia32_cmpltsd:
+ case X86::BI__builtin_ia32_cmplesd:
+ case X86::BI__builtin_ia32_cmpunordsd:
+ case X86::BI__builtin_ia32_cmpneqsd:
+ case X86::BI__builtin_ia32_cmpnltsd:
+ case X86::BI__builtin_ia32_cmpnlesd:
+ case X86::BI__builtin_ia32_cmpordsd:
+ // These exist so that the builtin that takes an immediate can be bounds
+ // checked by clang to avoid passing bad immediates to the backend. Since
+ // AVX has a larger immediate than SSE we would need separate builtins to
+ // do the different bounds checking. Rather than create a clang specific
+ // SSE only builtin, this implements eight separate builtins to match gcc
+ // implementation.
+
+ // Choose the immediate.
+ unsigned Imm;
+ switch (BuiltinID) {
+ default: llvm_unreachable("Unsupported intrinsic!");
+ case X86::BI__builtin_ia32_cmpeqps:
+ case X86::BI__builtin_ia32_cmpeqss:
+ case X86::BI__builtin_ia32_cmpeqpd:
+ case X86::BI__builtin_ia32_cmpeqsd:
+ Imm = 0;
+ break;
+ case X86::BI__builtin_ia32_cmpltps:
+ case X86::BI__builtin_ia32_cmpltss:
+ case X86::BI__builtin_ia32_cmpltpd:
+ case X86::BI__builtin_ia32_cmpltsd:
+ Imm = 1;
+ break;
+ case X86::BI__builtin_ia32_cmpleps:
+ case X86::BI__builtin_ia32_cmpless:
+ case X86::BI__builtin_ia32_cmplepd:
+ case X86::BI__builtin_ia32_cmplesd:
+ Imm = 2;
+ break;
+ case X86::BI__builtin_ia32_cmpunordps:
+ case X86::BI__builtin_ia32_cmpunordss:
+ case X86::BI__builtin_ia32_cmpunordpd:
+ case X86::BI__builtin_ia32_cmpunordsd:
+ Imm = 3;
+ break;
+ case X86::BI__builtin_ia32_cmpneqps:
+ case X86::BI__builtin_ia32_cmpneqss:
+ case X86::BI__builtin_ia32_cmpneqpd:
+ case X86::BI__builtin_ia32_cmpneqsd:
+ Imm = 4;
+ break;
+ case X86::BI__builtin_ia32_cmpnltps:
+ case X86::BI__builtin_ia32_cmpnltss:
+ case X86::BI__builtin_ia32_cmpnltpd:
+ case X86::BI__builtin_ia32_cmpnltsd:
+ Imm = 5;
+ break;
+ case X86::BI__builtin_ia32_cmpnleps:
+ case X86::BI__builtin_ia32_cmpnless:
+ case X86::BI__builtin_ia32_cmpnlepd:
+ case X86::BI__builtin_ia32_cmpnlesd:
+ Imm = 6;
+ break;
+ case X86::BI__builtin_ia32_cmpordps:
+ case X86::BI__builtin_ia32_cmpordss:
+ case X86::BI__builtin_ia32_cmpordpd:
+ case X86::BI__builtin_ia32_cmpordsd:
+ Imm = 7;
+ break;
+ }
+
+ // Choose the intrinsic ID.
+ const char *name;
+ Intrinsic::ID ID;
+ switch (BuiltinID) {
+ default: llvm_unreachable("Unsupported intrinsic!");
+ case X86::BI__builtin_ia32_cmpeqps:
+ case X86::BI__builtin_ia32_cmpltps:
+ case X86::BI__builtin_ia32_cmpleps:
+ case X86::BI__builtin_ia32_cmpunordps:
+ case X86::BI__builtin_ia32_cmpneqps:
+ case X86::BI__builtin_ia32_cmpnltps:
+ case X86::BI__builtin_ia32_cmpnleps:
+ case X86::BI__builtin_ia32_cmpordps:
+ name = "cmpps";
+ ID = Intrinsic::x86_sse_cmp_ps;
+ break;
+ case X86::BI__builtin_ia32_cmpeqss:
+ case X86::BI__builtin_ia32_cmpltss:
+ case X86::BI__builtin_ia32_cmpless:
+ case X86::BI__builtin_ia32_cmpunordss:
+ case X86::BI__builtin_ia32_cmpneqss:
+ case X86::BI__builtin_ia32_cmpnltss:
+ case X86::BI__builtin_ia32_cmpnless:
+ case X86::BI__builtin_ia32_cmpordss:
+ name = "cmpss";
+ ID = Intrinsic::x86_sse_cmp_ss;
+ break;
+ case X86::BI__builtin_ia32_cmpeqpd:
+ case X86::BI__builtin_ia32_cmpltpd:
+ case X86::BI__builtin_ia32_cmplepd:
+ case X86::BI__builtin_ia32_cmpunordpd:
+ case X86::BI__builtin_ia32_cmpneqpd:
+ case X86::BI__builtin_ia32_cmpnltpd:
+ case X86::BI__builtin_ia32_cmpnlepd:
+ case X86::BI__builtin_ia32_cmpordpd:
+ name = "cmppd";
+ ID = Intrinsic::x86_sse2_cmp_pd;
+ break;
+ case X86::BI__builtin_ia32_cmpeqsd:
+ case X86::BI__builtin_ia32_cmpltsd:
+ case X86::BI__builtin_ia32_cmplesd:
+ case X86::BI__builtin_ia32_cmpunordsd:
+ case X86::BI__builtin_ia32_cmpneqsd:
+ case X86::BI__builtin_ia32_cmpnltsd:
+ case X86::BI__builtin_ia32_cmpnlesd:
+ case X86::BI__builtin_ia32_cmpordsd:
+ name = "cmpsd";
+ ID = Intrinsic::x86_sse2_cmp_sd;
+ break;
+ }
+
+ Ops.push_back(llvm::ConstantInt::get(Int8Ty, Imm));
+ llvm::Function *F = CGM.getIntrinsic(ID);
+ return Builder.CreateCall(F, Ops, name);
}
}
Modified: cfe/trunk/lib/Headers/emmintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/emmintrin.h?rev=224879&r1=224878&r2=224879&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/emmintrin.h (original)
+++ cfe/trunk/lib/Headers/emmintrin.h Sat Dec 27 00:59:57 2014
@@ -155,148 +155,148 @@ _mm_xor_pd(__m128d __a, __m128d __b)
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpeq_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__a, __b, 0);
+ return (__m128d)__builtin_ia32_cmpeqpd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmplt_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__a, __b, 1);
+ return (__m128d)__builtin_ia32_cmpltpd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmple_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__a, __b, 2);
+ return (__m128d)__builtin_ia32_cmplepd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpgt_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__b, __a, 1);
+ return (__m128d)__builtin_ia32_cmpltpd(__b, __a);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpge_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__b, __a, 2);
+ return (__m128d)__builtin_ia32_cmplepd(__b, __a);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpord_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__a, __b, 7);
+ return (__m128d)__builtin_ia32_cmpordpd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpunord_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__a, __b, 3);
+ return (__m128d)__builtin_ia32_cmpunordpd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpneq_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__a, __b, 4);
+ return (__m128d)__builtin_ia32_cmpneqpd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpnlt_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__a, __b, 5);
+ return (__m128d)__builtin_ia32_cmpnltpd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpnle_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__a, __b, 6);
+ return (__m128d)__builtin_ia32_cmpnlepd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpngt_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__b, __a, 5);
+ return (__m128d)__builtin_ia32_cmpnltpd(__b, __a);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpnge_pd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmppd(__b, __a, 6);
+ return (__m128d)__builtin_ia32_cmpnlepd(__b, __a);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpeq_sd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmpsd(__a, __b, 0);
+ return (__m128d)__builtin_ia32_cmpeqsd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmplt_sd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmpsd(__a, __b, 1);
+ return (__m128d)__builtin_ia32_cmpltsd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmple_sd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmpsd(__a, __b, 2);
+ return (__m128d)__builtin_ia32_cmplesd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpgt_sd(__m128d __a, __m128d __b)
{
- __m128d __c = __builtin_ia32_cmpsd(__b, __a, 1);
+ __m128d __c = __builtin_ia32_cmpltsd(__b, __a);
return (__m128d) { __c[0], __a[1] };
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpge_sd(__m128d __a, __m128d __b)
{
- __m128d __c = __builtin_ia32_cmpsd(__b, __a, 2);
+ __m128d __c = __builtin_ia32_cmplesd(__b, __a);
return (__m128d) { __c[0], __a[1] };
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpord_sd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmpsd(__a, __b, 7);
+ return (__m128d)__builtin_ia32_cmpordsd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpunord_sd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmpsd(__a, __b, 3);
+ return (__m128d)__builtin_ia32_cmpunordsd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpneq_sd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmpsd(__a, __b, 4);
+ return (__m128d)__builtin_ia32_cmpneqsd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpnlt_sd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmpsd(__a, __b, 5);
+ return (__m128d)__builtin_ia32_cmpnltsd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpnle_sd(__m128d __a, __m128d __b)
{
- return (__m128d)__builtin_ia32_cmpsd(__a, __b, 6);
+ return (__m128d)__builtin_ia32_cmpnlesd(__a, __b);
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpngt_sd(__m128d __a, __m128d __b)
{
- __m128d __c = __builtin_ia32_cmpsd(__b, __a, 5);
+ __m128d __c = __builtin_ia32_cmpnltsd(__b, __a);
return (__m128d) { __c[0], __a[1] };
}
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_cmpnge_sd(__m128d __a, __m128d __b)
{
- __m128d __c = __builtin_ia32_cmpsd(__b, __a, 6);
+ __m128d __c = __builtin_ia32_cmpnlesd(__b, __a);
return (__m128d) { __c[0], __a[1] };
}
Modified: cfe/trunk/lib/Headers/xmmintrin.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Headers/xmmintrin.h?rev=224879&r1=224878&r2=224879&view=diff
==============================================================================
--- cfe/trunk/lib/Headers/xmmintrin.h (original)
+++ cfe/trunk/lib/Headers/xmmintrin.h Sat Dec 27 00:59:57 2014
@@ -182,153 +182,153 @@ _mm_xor_ps(__m128 __a, __m128 __b)
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpeq_ss(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpss(__a, __b, 0);
+ return (__m128)__builtin_ia32_cmpeqss(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpeq_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__a, __b, 0);
+ return (__m128)__builtin_ia32_cmpeqps(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmplt_ss(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpss(__a, __b, 1);
+ return (__m128)__builtin_ia32_cmpltss(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmplt_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__a, __b, 1);
+ return (__m128)__builtin_ia32_cmpltps(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmple_ss(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpss(__a, __b, 2);
+ return (__m128)__builtin_ia32_cmpless(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmple_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__a, __b, 2);
+ return (__m128)__builtin_ia32_cmpleps(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpgt_ss(__m128 __a, __m128 __b)
{
return (__m128)__builtin_shufflevector(__a,
- __builtin_ia32_cmpss(__b, __a, 1),
+ __builtin_ia32_cmpltss(__b, __a),
4, 1, 2, 3);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpgt_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__b, __a, 1);
+ return (__m128)__builtin_ia32_cmpltps(__b, __a);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpge_ss(__m128 __a, __m128 __b)
{
return (__m128)__builtin_shufflevector(__a,
- __builtin_ia32_cmpss(__b, __a, 2),
+ __builtin_ia32_cmpless(__b, __a),
4, 1, 2, 3);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpge_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__b, __a, 2);
+ return (__m128)__builtin_ia32_cmpleps(__b, __a);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpneq_ss(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpss(__a, __b, 4);
+ return (__m128)__builtin_ia32_cmpneqss(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpneq_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__a, __b, 4);
+ return (__m128)__builtin_ia32_cmpneqps(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpnlt_ss(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpss(__a, __b, 5);
+ return (__m128)__builtin_ia32_cmpnltss(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpnlt_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__a, __b, 5);
+ return (__m128)__builtin_ia32_cmpnltps(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpnle_ss(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpss(__a, __b, 6);
+ return (__m128)__builtin_ia32_cmpnless(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpnle_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__a, __b, 6);
+ return (__m128)__builtin_ia32_cmpnleps(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpngt_ss(__m128 __a, __m128 __b)
{
return (__m128)__builtin_shufflevector(__a,
- __builtin_ia32_cmpss(__b, __a, 5),
+ __builtin_ia32_cmpnltss(__b, __a),
4, 1, 2, 3);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpngt_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__b, __a, 5);
+ return (__m128)__builtin_ia32_cmpnltps(__b, __a);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpnge_ss(__m128 __a, __m128 __b)
{
return (__m128)__builtin_shufflevector(__a,
- __builtin_ia32_cmpss(__b, __a, 6),
+ __builtin_ia32_cmpnless(__b, __a),
4, 1, 2, 3);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpnge_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__b, __a, 6);
+ return (__m128)__builtin_ia32_cmpnleps(__b, __a);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpord_ss(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpss(__a, __b, 7);
+ return (__m128)__builtin_ia32_cmpordss(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpord_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__a, __b, 7);
+ return (__m128)__builtin_ia32_cmpordps(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpunord_ss(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpss(__a, __b, 3);
+ return (__m128)__builtin_ia32_cmpunordss(__a, __b);
}
static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
_mm_cmpunord_ps(__m128 __a, __m128 __b)
{
- return (__m128)__builtin_ia32_cmpps(__a, __b, 3);
+ return (__m128)__builtin_ia32_cmpunordps(__a, __b);
}
static __inline__ int __attribute__((__always_inline__, __nodebug__))
Modified: cfe/trunk/test/CodeGen/sse-builtins.c
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/sse-builtins.c?rev=224879&r1=224878&r2=224879&view=diff
==============================================================================
--- cfe/trunk/test/CodeGen/sse-builtins.c (original)
+++ cfe/trunk/test/CodeGen/sse-builtins.c Sat Dec 27 00:59:57 2014
@@ -255,3 +255,291 @@ __m128i test_blend_epi16(__m128i V1, __m
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> %{{.*}}, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 7>
return _mm_blend_epi16(V1, V2, 42);
}
+
+__m128 test_mm_cmpeq_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpeq_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 0)
+ return _mm_cmpeq_ss(__a, __b);
+}
+
+__m128 test_mm_cmplt_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmplt_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 1)
+ return _mm_cmplt_ss(__a, __b);
+}
+
+__m128 test_mm_cmple_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmple_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 2)
+ return _mm_cmple_ss(__a, __b);
+}
+
+__m128 test_mm_cmpunord_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpunord_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 3)
+ return _mm_cmpunord_ss(__a, __b);
+}
+
+__m128 test_mm_cmpneq_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpneq_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 4)
+ return _mm_cmpneq_ss(__a, __b);
+}
+
+__m128 test_mm_cmpnlt_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpnlt_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 5)
+ return _mm_cmpnlt_ss(__a, __b);
+}
+
+__m128 test_mm_cmpnle_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpnle_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 6)
+ return _mm_cmpnle_ss(__a, __b);
+}
+
+__m128 test_mm_cmpord_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpord_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 7)
+ return _mm_cmpord_ss(__a, __b);
+}
+
+__m128 test_mm_cmpgt_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpgt_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 1)
+ return _mm_cmpgt_ss(__a, __b);
+}
+
+__m128 test_mm_cmpge_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpge_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 2)
+ return _mm_cmpge_ss(__a, __b);
+}
+
+__m128 test_mm_cmpngt_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpngt_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 5)
+ return _mm_cmpngt_ss(__a, __b);
+}
+
+__m128 test_mm_cmpnge_ss(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpnge_ss
+ // CHECK: @llvm.x86.sse.cmp.ss(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 6)
+ return _mm_cmpnge_ss(__a, __b);
+}
+
+__m128 test_mm_cmpeq_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpeq_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 0)
+ return _mm_cmpeq_ps(__a, __b);
+}
+
+__m128 test_mm_cmplt_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmplt_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 1)
+ return _mm_cmplt_ps(__a, __b);
+}
+
+__m128 test_mm_cmple_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmple_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 2)
+ return _mm_cmple_ps(__a, __b);
+}
+
+__m128 test_mm_cmpunord_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpunord_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 3)
+ return _mm_cmpunord_ps(__a, __b);
+}
+
+__m128 test_mm_cmpneq_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpneq_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 4)
+ return _mm_cmpneq_ps(__a, __b);
+}
+
+__m128 test_mm_cmpnlt_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpnlt_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 5)
+ return _mm_cmpnlt_ps(__a, __b);
+}
+
+__m128 test_mm_cmpnle_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpnle_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 6)
+ return _mm_cmpnle_ps(__a, __b);
+}
+
+__m128 test_mm_cmpord_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpord_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 7)
+ return _mm_cmpord_ps(__a, __b);
+}
+
+__m128 test_mm_cmpgt_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpgt_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 1)
+ return _mm_cmpgt_ps(__a, __b);
+}
+
+__m128 test_mm_cmpge_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpge_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 2)
+ return _mm_cmpge_ps(__a, __b);
+}
+
+__m128 test_mm_cmpngt_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpngt_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 5)
+ return _mm_cmpngt_ps(__a, __b);
+}
+
+__m128 test_mm_cmpnge_ps(__m128 __a, __m128 __b) {
+ // CHECK-LABEL: @test_mm_cmpnge_ps
+ // CHECK: @llvm.x86.sse.cmp.ps(<4 x float> %{{.*}}, <4 x float> %{{.*}}, i8 6)
+ return _mm_cmpnge_ps(__a, __b);
+}
+
+__m128d test_mm_cmpeq_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpeq_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 0)
+ return _mm_cmpeq_sd(__a, __b);
+}
+
+__m128d test_mm_cmplt_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmplt_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 1)
+ return _mm_cmplt_sd(__a, __b);
+}
+
+__m128d test_mm_cmple_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmple_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 2)
+ return _mm_cmple_sd(__a, __b);
+}
+
+__m128d test_mm_cmpunord_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpunord_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 3)
+ return _mm_cmpunord_sd(__a, __b);
+}
+
+__m128d test_mm_cmpneq_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpneq_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 4)
+ return _mm_cmpneq_sd(__a, __b);
+}
+
+__m128d test_mm_cmpnlt_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpnlt_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 5)
+ return _mm_cmpnlt_sd(__a, __b);
+}
+
+__m128d test_mm_cmpnle_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpnle_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 6)
+ return _mm_cmpnle_sd(__a, __b);
+}
+
+__m128d test_mm_cmpord_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpord_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 7)
+ return _mm_cmpord_sd(__a, __b);
+}
+
+__m128d test_mm_cmpgt_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpgt_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 1)
+ return _mm_cmpgt_sd(__a, __b);
+}
+
+__m128d test_mm_cmpge_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpge_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 2)
+ return _mm_cmpge_sd(__a, __b);
+}
+
+__m128d test_mm_cmpngt_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpngt_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 5)
+ return _mm_cmpngt_sd(__a, __b);
+}
+
+__m128d test_mm_cmpnge_sd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpnge_sd
+ // CHECK: @llvm.x86.sse2.cmp.sd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 6)
+ return _mm_cmpnge_sd(__a, __b);
+}
+
+__m128d test_mm_cmpeq_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpeq_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 0)
+ return _mm_cmpeq_pd(__a, __b);
+}
+
+__m128d test_mm_cmplt_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmplt_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 1)
+ return _mm_cmplt_pd(__a, __b);
+}
+
+__m128d test_mm_cmple_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmple_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 2)
+ return _mm_cmple_pd(__a, __b);
+}
+
+__m128d test_mm_cmpunord_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpunord_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 3)
+ return _mm_cmpunord_pd(__a, __b);
+}
+
+__m128d test_mm_cmpneq_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpneq_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 4)
+ return _mm_cmpneq_pd(__a, __b);
+}
+
+__m128d test_mm_cmpnlt_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpnlt_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 5)
+ return _mm_cmpnlt_pd(__a, __b);
+}
+
+__m128d test_mm_cmpnle_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpnle_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 6)
+ return _mm_cmpnle_pd(__a, __b);
+}
+
+__m128d test_mm_cmpord_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpord_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 7)
+ return _mm_cmpord_pd(__a, __b);
+}
+
+__m128d test_mm_cmpgt_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpgt_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 1)
+ return _mm_cmpgt_pd(__a, __b);
+}
+
+__m128d test_mm_cmpge_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpge_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 2)
+ return _mm_cmpge_pd(__a, __b);
+}
+
+__m128d test_mm_cmpngt_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpngt_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 5)
+ return _mm_cmpngt_pd(__a, __b);
+}
+
+__m128d test_mm_cmpnge_pd(__m128d __a, __m128d __b) {
+ // CHECK-LABEL: @test_mm_cmpnge_pd
+ // CHECK: @llvm.x86.sse2.cmp.pd(<2 x double> %{{.*}}, <2 x double> %{{.*}}, i8 6)
+ return _mm_cmpnge_pd(__a, __b);
+}
More information about the cfe-commits
mailing list