re PR c++/35553 (-fkeep-inline-functions and -O errors out in SSE headers)

	PR target/35553
	* config/i386/i386.h (TARGET_CPU_CPP_BUILTINS): Define
	__SSE_USE_INLINED_FUNC__ when using -flag-keep-inline-functions
	together with optimization.
	* config/i386/xmmintrin.h: Use __SSE_USE_INLINED_FUNC__ instead of
	__OPTIMIZE__ to choose between inlined intrinsic SSE function having
	immediate arguments and its equivalent macro definition.
	* config/i386/bmintrin.h: Ditto.
	* config/i386/smmintrin.h: Ditto.
	* config/i386/tmmintrin.h: Ditto.
	* config/i386/mmintrin-common.h: Ditto.
	* config/i386/ammintrin.h: Ditto.
	* config/i386/emmintrin.h: Ditto.

testsuite/ChangeLog:

	PR target/35553
	* g++.dg/other/i386-3.C: New test.

From-SVN: r133164

11 files changed, 60 insertions, 26 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 2fa006f..d85d572 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,19 @@
+2008-03-13  Uros Bizjak  <ubizjak@gmail.com>
+
+	PR target/35553
+	* config/i386/i386.h (TARGET_CPU_CPP_BUILTINS): Define
+	__SSE_USE_INLINED_FUNC__ when using -flag-keep-inline-functions
+	together with optimization.
+	* config/i386/xmmintrin.h: Use __SSE_USE_INLINED_FUNC__ instead of
+	__OPTIMIZE__ to choose between inlined intrinsic SSE function having
+	immediate arguments and its equivalent macro definition.
+	* config/i386/bmintrin.h: Ditto.
+	* config/i386/smmintrin.h: Ditto.
+	* config/i386/tmmintrin.h: Ditto.
+	* config/i386/mmintrin-common.h: Ditto.
+	* config/i386/ammintrin.h: Ditto.
+	* config/i386/emmintrin.h: Ditto.
+
 2008-03-13  Jakub Jelinek  <jakub@redhat.com>
 
 	PR middle-end/35185
diff --git a/gcc/config/i386/ammintrin.h b/gcc/config/i386/ammintrin.h
index 1351ebd..69ee246 100644
--- a/gcc/config/i386/ammintrin.h
+++ b/gcc/config/i386/ammintrin.h
@@ -55,7 +55,7 @@ _mm_extract_si64 (__m128i __X, __m128i __Y)
   return (__m128i) __builtin_ia32_extrq ((__v2di) __X, (__v16qi) __Y);
 }
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_extracti_si64 (__m128i __X, unsigned const int __I, unsigned const int __L)
 {
@@ -73,7 +73,7 @@ _mm_insert_si64 (__m128i __X,__m128i __Y)
   return (__m128i) __builtin_ia32_insertq ((__v2di)__X, (__v2di)__Y);
 }
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_inserti_si64(__m128i __X, __m128i __Y, unsigned const int __I, unsigned const int __L)
 {
diff --git a/gcc/config/i386/bmmintrin.h b/gcc/config/i386/bmmintrin.h
index 4254d0f..72f0efd 100644
--- a/gcc/config/i386/bmmintrin.h
+++ b/gcc/config/i386/bmmintrin.h
@@ -350,7 +350,7 @@ _mm_rot_epi64(__m128i __A,  __m128i __B)
 
 
 /* Rotates - Immediate form */
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_roti_epi8(__m128i __A, const int __B)
 {
diff --git a/gcc/config/i386/emmintrin.h b/gcc/config/i386/emmintrin.h
index c5bbe8b..1244958 100644
--- a/gcc/config/i386/emmintrin.h
+++ b/gcc/config/i386/emmintrin.h
@@ -880,7 +880,7 @@ _mm_cvtss_sd (__m128d __A, __m128 __B)
   return (__m128d)__builtin_ia32_cvtss2sd ((__v2df) __A, (__v4sf)__B);
 }
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128d __attribute__((__always_inline__, __artificial__))
 _mm_shuffle_pd(__m128d __A, __m128d __B, const int __mask)
 {
@@ -1144,7 +1144,7 @@ _mm_srai_epi32 (__m128i __A, int __B)
   return (__m128i)__builtin_ia32_psradi128 ((__v4si)__A, __B);
 }
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_srli_si128 (__m128i __A, const int __N)
 {
@@ -1307,7 +1307,7 @@ _mm_cmpgt_epi32 (__m128i __A, __m128i __B)
   return (__m128i)__builtin_ia32_pcmpgtd128 ((__v4si)__A, (__v4si)__B);
 }
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline int __attribute__((__always_inline__, __artificial__))
 _mm_extract_epi16 (__m128i const __A, int const __N)
 {
@@ -1363,7 +1363,7 @@ _mm_mulhi_epu16 (__m128i __A, __m128i __B)
   return (__m128i)__builtin_ia32_pmulhuw128 ((__v8hi)__A, (__v8hi)__B);
 }
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_shufflehi_epi16 (__m128i __A, const int __mask)
 {
diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h
index 98cb72a..ea0a820 100644
--- a/gcc/config/i386/i386.h
+++ b/gcc/config/i386/i386.h
@@ -691,6 +691,11 @@ extern const char *host_detect_local_cpu (int argc, const char **argv);
 	builtin_define ("__SSE_MATH__");			\
       if (TARGET_SSE_MATH && TARGET_SSE2)			\
 	builtin_define ("__SSE2_MATH__");			\
+								\
+      /* Use inlined intrinsic SSE function having immediate	\
+	 arguments instead of a macro definition.  */		\
+      if (optimize && !flag_keep_inline_functions)		\
+	builtin_define ("__SSE_USE_INLINED_FUNC__");		\
     }								\
   while (0)
 
diff --git a/gcc/config/i386/mmintrin-common.h b/gcc/config/i386/mmintrin-common.h
index 3be8333..308c2c2 100644
--- a/gcc/config/i386/mmintrin-common.h
+++ b/gcc/config/i386/mmintrin-common.h
@@ -92,7 +92,7 @@ _mm_testnzc_si128 (__m128i __M, __m128i __V)
 
 /* Packed/scalar double precision floating point rounding.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128d __attribute__((__always_inline__, __artificial__))
 _mm_round_pd (__m128d __V, const int __M)
 {
@@ -117,7 +117,7 @@ _mm_round_sd(__m128d __D, __m128d __V, const int __M)
 
 /* Packed/scalar single precision floating point rounding.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128 __attribute__((__always_inline__, __artificial__))
 _mm_round_ps (__m128 __V, const int __M)
 {
diff --git a/gcc/config/i386/smmintrin.h b/gcc/config/i386/smmintrin.h
index 08bfd10..bb70cb4 100644
--- a/gcc/config/i386/smmintrin.h
+++ b/gcc/config/i386/smmintrin.h
@@ -44,7 +44,7 @@
 /* Integer blend instructions - select data from 2 sources using
    constant/variable mask.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_blend_epi16 (__m128i __X, __m128i __Y, const int __M)
 {
@@ -69,7 +69,7 @@ _mm_blendv_epi8 (__m128i __X, __m128i __Y, __m128i __M)
 /* Single precision floating point blend instructions - select data
    from 2 sources using constant/variable mask.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128 __attribute__((__always_inline__, __artificial__))
 _mm_blend_ps (__m128 __X, __m128 __Y, const int __M)
 {
@@ -94,7 +94,7 @@ _mm_blendv_ps (__m128 __X, __m128 __Y, __m128 __M)
 /* Double precision floating point blend instructions - select data
    from 2 sources using constant/variable mask.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128d __attribute__((__always_inline__, __artificial__))
 _mm_blend_pd (__m128d __X, __m128d __Y, const int __M)
 {
@@ -119,7 +119,7 @@ _mm_blendv_pd (__m128d __X, __m128d __Y, __m128d __M)
 /* Dot product instructions with mask-defined summing and zeroing parts
    of result.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128 __attribute__((__always_inline__, __artificial__))
 _mm_dp_ps (__m128 __X, __m128 __Y, const int __M)
 {
@@ -224,7 +224,7 @@ _mm_mul_epi32 (__m128i __X, __m128i __Y)
    index, the bits [5-4] define D index, and bits [3-0] define
    zeroing mask for D.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128 __attribute__((__always_inline__, __artificial__))
 _mm_insert_ps (__m128 __D, __m128 __S, const int __N)
 {
@@ -244,7 +244,7 @@ _mm_insert_ps (__m128 __D, __m128 __S, const int __N)
 /* Extract binary representation of single precision float from packed
    single precision array element of X selected by index N.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline int __attribute__((__always_inline__, __artificial__))
 _mm_extract_ps (__m128 __X, const int __N)
 {
@@ -277,7 +277,7 @@ _mm_extract_ps (__m128 __X, const int __N)
 /* Insert integer, S, into packed integer array element of D
    selected by index N.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_insert_epi8 (__m128i __D, int __S, const int __N)
 {
@@ -319,7 +319,7 @@ _mm_insert_epi64 (__m128i __D, long long __S, const int __N)
 /* Extract integer from packed integer array element of X selected by
    index N.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline int __attribute__((__always_inline__, __artificial__))
 _mm_extract_epi8 (__m128i __X, const int __N)
 {
@@ -447,7 +447,7 @@ _mm_packus_epi32 (__m128i __X, __m128i __Y)
    byte integers in the first 2 operands.  Starting offsets within
    operands are determined by the 3rd mask operand.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_mpsadbw_epu8 (__m128i __X, __m128i __Y, const int __M)
 {
@@ -497,7 +497,7 @@ _mm_stream_load_si128 (__m128i *__X)
 
 /* Intrinsics for text/string processing.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_cmpistrm (__m128i __X, __m128i __Y, const int __M)
 {
@@ -550,7 +550,7 @@ _mm_cmpestri (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
 /* Intrinsics for text/string processing and reading values of
    EFlags.  */
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline int __attribute__((__always_inline__, __artificial__))
 _mm_cmpistra (__m128i __X, __m128i __Y, const int __M)
 {
diff --git a/gcc/config/i386/tmmintrin.h b/gcc/config/i386/tmmintrin.h
index 0e6a0d4..6fd20b1 100644
--- a/gcc/config/i386/tmmintrin.h
+++ b/gcc/config/i386/tmmintrin.h
@@ -181,7 +181,7 @@ _mm_sign_pi32 (__m64 __X, __m64 __Y)
   return (__m64) __builtin_ia32_psignd ((__v2si)__X, (__v2si)__Y);
 }
 
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128i __attribute__((__always_inline__, __artificial__))
 _mm_alignr_epi8(__m128i __X, __m128i __Y, const int __N)
 {
diff --git a/gcc/config/i386/xmmintrin.h b/gcc/config/i386/xmmintrin.h
index cfe55ef..e0cf958 100644
--- a/gcc/config/i386/xmmintrin.h
+++ b/gcc/config/i386/xmmintrin.h
@@ -716,7 +716,7 @@ _mm_cvtps_pi8(__m128 __A)
 }
 
 /* Selects four specific SPFP values from A and B based on MASK.  */
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m128 __attribute__((__always_inline__, __artificial__))
 _mm_shuffle_ps (__m128 __A, __m128 __B, int const __mask)
 {
@@ -992,7 +992,7 @@ _mm_move_ss (__m128 __A, __m128 __B)
 }
 
 /* Extracts one of the four words of A.  The selector N must be immediate.  */
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline int __attribute__((__always_inline__, __artificial__))
 _mm_extract_pi16 (__m64 const __A, int const __N)
 {
@@ -1013,7 +1013,7 @@ _m_pextrw (__m64 const __A, int const __N)
 
 /* Inserts word D into one of four words of A.  The selector N must be
    immediate.  */
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m64 __attribute__((__always_inline__, __artificial__))
 _mm_insert_pi16 (__m64 const __A, int const __D, int const __N)
 {
@@ -1114,7 +1114,7 @@ _m_pmulhuw (__m64 __A, __m64 __B)
 
 /* Return a combination of the four 16-bit values in A.  The selector
    must be an immediate.  */
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline __m64 __attribute__((__always_inline__, __artificial__))
 _mm_shuffle_pi16 (__m64 __A, int const __N)
 {
@@ -1191,7 +1191,7 @@ _m_psadbw (__m64 __A, __m64 __B)
 
 /* Loads one cache line from address P to a location "closer" to the
    processor.  The selector I specifies the type of prefetch operation.  */
-#ifdef __OPTIMIZE__
+#ifdef __SSE_USE_INLINED_FUNC__
 static __inline void __attribute__((__always_inline__, __artificial__))
 _mm_prefetch (const void *__P, enum _mm_hint __I)
 {
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 4467bb9..1b1b7ea 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2008-03-13  Uros Bizjak  <ubizjak@gmail.com>
+
+	PR target/35553
+	* g++.dg/other/i386-3.C: New test.
+
 2008-03-13  Paolo Bonzini  <bonzini@gnu.org>
 
 	PR tree-opt/35422
diff --git a/gcc/testsuite/g++.dg/other/i386-3.C b/gcc/testsuite/g++.dg/other/i386-3.C
new file mode 100644
index 0000000..8ebc74e6
--- /dev/null
+++ b/gcc/testsuite/g++.dg/other/i386-3.C
@@ -0,0 +1,8 @@
+/* Test that {,x,e,p,t,s,a,b}mmintrin.h, mm3dnow.h and mm_malloc.h are
+   usable with -O -fkeep-inline-functions.  */
+/* { dg-do compile { target i?86-*-* x86_64-*-* } } */
+/* { dg-options "-O -fkeep-inline-functions -march=k8 -m3dnow -msse4 -msse5" } */
+
+#include <bmmintrin.h>
+#include <smmintrin.h>
+#include <mm3dnow.h>