From ce2e20c483ad6ed39c9a8e5a6804a0f9e5986ae5 Mon Sep 17 00:00:00 2001
From: Nico Weber <nicolasweber@gmx.de>
Date: Mon, 16 May 2016 18:14:07 +0000
Subject: [PATCH] [ms] Reintroduce feature guards in intrinsic headers in
 Microsoft mode

Visual Studio's C++ standard library headers include intrin.h, so the intrinsic
headers get included a lot more often in Microsoft mode than elsewhere. The
AVX512 intrinsics are a lot of code (0.7 MB, causing 30% compile time overhead
for small programs including e.g. <string> and 6% compile time overhead for
larger projects like e.g. v8). Since multiversioning can't be relied on in
Microsoft mode (cl.exe doesn't support it), having faster compiles seems like
the much better tradeoff until we have a better intrinsic story going forward
(which we'll need for e.g. PR19898).

Actually using intrinsics on Windows already requires the right /arch:
settings, so this patch should have no big behavior change.

See also thread "The intrinsics headers (especially avx512) are too big. What
to do about it?" on cfe-dev.

http://reviews.llvm.org/D20291


git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@269675 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Headers/immintrin.h    | 82 +++++++++++++++++++++++++++++++++++++-
 lib/Headers/x86intrin.h    | 24 +++++++++++
 test/CodeGen/ms-mm-align.c |  2 +-
 3 files changed, 106 insertions(+), 2 deletions(-)

diff --git a/lib/Headers/immintrin.h b/lib/Headers/immintrin.h
index a74dad84a4..d6c6ae5e4c 100644
--- a/lib/Headers/immintrin.h
+++ b/lib/Headers/immintrin.h
@@ -24,22 +24,41 @@
 #ifndef __IMMINTRIN_H
 #define __IMMINTRIN_H
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__MMX__)
 #include <mmintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__SSE__)
 #include <xmmintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__SSE2__)
 #include <emmintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__SSE3__)
 #include <pmmintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__SSSE3__)
 #include <tmmintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || \
+    (defined(__SSE4_2__) || defined(__SSE4_1__))
 #include <smmintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || \
+    (defined(__AES__) || defined(__PCLMUL__))
 #include <wmmintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX__)
 #include <avxintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX2__)
 #include <avx2intrin.h>
 
 /* The 256-bit versions of functions in f16cintrin.h.
@@ -54,45 +73,90 @@ _mm256_cvtph_ps(__m128i __a)
 {
   return (__m256)__builtin_ia32_vcvtph2ps256((__v8hi)__a);
 }
+#endif /* __AVX2__ */
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__BMI__)
 #include <bmiintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__BMI2__)
 #include <bmi2intrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__LZCNT__)
 #include <lzcntintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__FMA__)
 #include <fmaintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512F__)
 #include <avx512fintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512VL__)
 #include <avx512vlintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512BW__)
 #include <avx512bwintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512CD__)
 #include <avx512cdintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512DQ__)
 #include <avx512dqintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || \
+    (defined(__AVX512VL__) && defined(__AVX512BW__))
 #include <avx512vlbwintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || \
+    (defined(__AVX512VL__) && defined(__AVX512CD__))
 #include <avx512vlcdintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || \
+    (defined(__AVX512VL__) && defined(__AVX512DQ__))
 #include <avx512vldqintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512ER__)
 #include <avx512erintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512IFMA__)
 #include <avx512ifmaintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || \
+    (defined(__AVX512IFMA__) && defined(__AVX512VL__))
 #include <avx512ifmavlintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512VBMI__)
 #include <avx512vbmiintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || \
+    (defined(__AVX512VBMI__) && defined(__AVX512VL__))
 #include <avx512vbmivlintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__AVX512PF__)
 #include <avx512pfintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__PKU__)
 #include <pkuintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__RDRND__)
 static __inline__ int __attribute__((__always_inline__, __nodebug__, __target__("rdrnd")))
 _rdrand16_step(unsigned short *__p)
 {
@@ -112,7 +176,9 @@ _rdrand64_step(unsigned long long *__p)
   return __builtin_ia32_rdrand64_step(__p);
 }
 #endif
+#endif /* __RDRND__ */
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__FSGSBASE__)
 #ifdef __x86_64__
 static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__, __target__("fsgsbase")))
 _readfsbase_u32(void)
@@ -162,22 +228,36 @@ _writegsbase_u64(unsigned long long __V)
   return __builtin_ia32_wrgsbase64(__V);
 }
 #endif
+#endif /* __FSGSBASE__ */
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__RTM__)
 #include <rtmintrin.h>
-
 #include <xtestintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__SHA__)
 #include <shaintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__FXSR__)
 #include <fxsrintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__XSAVE__)
 #include <xsaveintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__XSAVEOPT__)
 #include <xsaveoptintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__XSAVEC__)
 #include <xsavecintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__XSAVES__)
 #include <xsavesintrin.h>
+#endif
 
 /* Some intrinsics inside adxintrin.h are available only on processors with ADX,
  * whereas others are also available at all times. */
diff --git a/lib/Headers/x86intrin.h b/lib/Headers/x86intrin.h
index 4d8077e382..6f3c8c4f3f 100644
--- a/lib/Headers/x86intrin.h
+++ b/lib/Headers/x86intrin.h
@@ -28,29 +28,53 @@
 
 #include <immintrin.h>
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__3dNOW__)
 #include <mm3dnow.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__BMI__)
 #include <bmiintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__BMI2__)
 #include <bmi2intrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__LZCNT__)
 #include <lzcntintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__POPCNT__)
 #include <popcntintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__RDSEED__)
 #include <rdseedintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__PRFCHW__)
 #include <prfchwintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__SSE4A__)
 #include <ammintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__FMA4__)
 #include <fma4intrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__XOP__)
 #include <xopintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__TBM__)
 #include <tbmintrin.h>
+#endif
 
+#if !defined(_MSC_VER) || __has_feature(modules) || defined(__F16C__)
 #include <f16cintrin.h>
+#endif
 
 /* FIXME: LWP */
 
diff --git a/test/CodeGen/ms-mm-align.c b/test/CodeGen/ms-mm-align.c
index ae8e98086a..49fedddc01 100644
--- a/test/CodeGen/ms-mm-align.c
+++ b/test/CodeGen/ms-mm-align.c
@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -ffreestanding -fms-extensions -fms-compatibility -fms-compatibility-version=17.00 \
+// RUN: %clang_cc1 -ffreestanding -fms-extensions -fms-compatibility -fms-compatibility-version=17.00 -target-feature +sse \
 // RUN:         -triple i686--windows -emit-llvm %s -o - \
 // RUN:         | FileCheck %s -check-prefix CHECK
 
-- 
2.40.0