From d45c5767d8ff81e01f56418095be3d115eb0507c Mon Sep 17 00:00:00 2001
From: "james.brodman" <james.brodman@intel.com>
Date: Tue, 1 Oct 2013 12:17:57 -0400
Subject: [PATCH 1/3] Due diligence tweaks.

---
 examples/intrinsics/knc-i1x16.h | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/examples/intrinsics/knc-i1x16.h b/examples/intrinsics/knc-i1x16.h
index ae9c4130..730141ec 100644
--- a/examples/intrinsics/knc-i1x16.h
+++ b/examples/intrinsics/knc-i1x16.h
@@ -1021,12 +1021,22 @@ static FORCEINLINE __vec16_i64 __mul(const __vec16_i32 &a, const __vec16_i64 &_b
         _mm512_mulhi_epi32(a.v, b.v_lo))).cvt2zmm();
 }
 
-#if __ICC_VERSION == 1400
+#if __ICC_VERSION >= 1400
 static FORCEINLINE __vec16_i64 __mul(__vec16_i64 a, __vec16_i64 b) {
   return __vec16_i64(_mm512_mullox_epi64(a.v1, b.v1), _mm512_mullox_epi64(a.v2,b.v2));
 }
 #else
-BINARY_OP(__vec16_i64, __mul, *)
+static FORCEINLINE __vec16_i64 __mul(const __vec16_i64 &a, const __vec16_i64 &b)
+{
+    __vec16_i32 lo = _mm512_mullo_epi32(a.v_lo,b.v_lo);
+    __vec16_i32 hi_m1 = _mm512_mulhi_epi32(a.v_lo, b.v_lo);
+    __vec16_i32 hi_m2 = _mm512_mullo_epi32(a.v_hi, b.v_lo);
+    __vec16_i32 hi_m3 = _mm512_mullo_epi32(a.v_lo, b.v_hi);
+    __mmask16 carry = 0;
+    __vec16_i32 hi_p23 = _mm512_addsetc_epi32(hi_m2, hi_m1, &carry);
+    __vec16_i32 hi = _mm512_adc_epi32(hi_m3, carry, hi_p23, &carry);
+    return __vec16_i64(lo, hi);
+}
 #endif
 #endif
 

From ac79f3f34555a97da837834076c8adc89e6c50a4 Mon Sep 17 00:00:00 2001
From: "james.brodman" <james.brodman@intel.com>
Date: Tue, 1 Oct 2013 12:31:33 -0400
Subject: [PATCH 2/3] format change

---
 examples/intrinsics/knc-i1x16.h | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/examples/intrinsics/knc-i1x16.h b/examples/intrinsics/knc-i1x16.h
index 730141ec..84a1f7aa 100644
--- a/examples/intrinsics/knc-i1x16.h
+++ b/examples/intrinsics/knc-i1x16.h
@@ -1026,8 +1026,10 @@ static FORCEINLINE __vec16_i64 __mul(__vec16_i64 a, __vec16_i64 b) {
   return __vec16_i64(_mm512_mullox_epi64(a.v1, b.v1), _mm512_mullox_epi64(a.v2,b.v2));
 }
 #else
-static FORCEINLINE __vec16_i64 __mul(const __vec16_i64 &a, const __vec16_i64 &b)
+static FORCEINLINE __vec16_i64 __mul(const __vec16_i64 &_a, const __vec16_i64 &_b)
 {
+    const __vec16_i64 a = _a.cvt2hilo();
+    const __vec16_i64 b = _b.cvt2hilo();
     __vec16_i32 lo = _mm512_mullo_epi32(a.v_lo,b.v_lo);
     __vec16_i32 hi_m1 = _mm512_mulhi_epi32(a.v_lo, b.v_lo);
     __vec16_i32 hi_m2 = _mm512_mullo_epi32(a.v_hi, b.v_lo);
@@ -1035,7 +1037,7 @@ static FORCEINLINE __vec16_i64 __mul(const __vec16_i64 &a, const __vec16_i64 &b)
     __mmask16 carry = 0;
     __vec16_i32 hi_p23 = _mm512_addsetc_epi32(hi_m2, hi_m1, &carry);
     __vec16_i32 hi = _mm512_adc_epi32(hi_m3, carry, hi_p23, &carry);
-    return __vec16_i64(lo, hi);
+    return __vec16_i64(hi,lo).cvt2zmm();
 }
 #endif
 #endif

From 44912e6b1e0478da79fe4c3a00fd5a64f2904ece Mon Sep 17 00:00:00 2001
From: "james.brodman" <james.brodman@intel.com>
Date: Tue, 8 Oct 2013 18:27:03 -0400
Subject: [PATCH 3/3] Fix segfault when using both -g and -MMM

---
 module.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/module.cpp b/module.cpp
index 755a5dc4..41861a2d 100644
--- a/module.cpp
+++ b/module.cpp
@@ -936,7 +936,7 @@ Module::AddExportedTypes(const std::vector<std::pair<const Type *,
 bool
 Module::writeOutput(OutputType outputType, const char *outFileName,
                     const char *includeFileName) {
-    if (diBuilder != NULL && outputType != Header) {
+    if (diBuilder != NULL && (outputType != Header && outputType != Deps)) {
         diBuilder->finalize();
 
         lStripUnusedDebugInfo(module);