Release 1.4.2, 11 June 2013

docs/ReleaseNotes.txt

@@ -1,3 +1,19 @@
+=== v1.4.2 === (11 June 2013)
+
+A minor version update with a few important changes:
+
+* Stability fix for AVX2 target (Haswell) - problem with gather instructions was
+  released in LLVM 3.4, if you build with LLVM 3.2 or 3.3, it's available in our
+  repository (llvm_patches/r183327-AVX2-GATHER.patch) and needs to be applied
+  manually.
+
+* Stability fix for widespread issue on Win32 platform (#503).
+
+* Performance improvements for Xeon Phi related to mask representation.
+
+Also LLVM 3.3 has been released and now it's the recommended version for building ISPC.
+Precompiled binaries are also built with LLVM 3.3. 
+
 === v1.4.1 === (28 May 2013)
 
 A major new version of ispc has been released with stability and performance
@@ -20,7 +36,7 @@ Important bug fixes/changes:
 
 * FMA instructions are enabled for AVX2 instruction set.
 
-* Support of RDRAND instruction when availible via library function rdrand (Ivy Bridge).
+* Support of RDRAND instruction when available via library function rdrand (Ivy Bridge).
 
 Release also contains numerous bug fixes and minor improvements.
 

docs/news.rst

@@ -2,6 +2,17 @@
 ispc News
 =========
 
+ispc 1.4.2 is Released
+----------------------
+
+A minor update of ``ispc`` has been released with stability fix for AVX2
+(Haswell), fix for Win32 platform and performance improvements on Xeon Phi.
+As usual, it's available on all supported platforms (Windows, Linux and MacOS).
+This version supports LLVM 3.1, 3.2, 3.3 and 3.4, but now we are recommending
+to avoid 3.1, as it's known to contain a number of stability problems and we are
+planning to deprecate its support soon.
+The released binaries are built with 3.3.
+
 ispc 1.4.1 is Released
 ----------------------
 

doxygen.cfg

@@ -31,7 +31,7 @@ PROJECT_NAME           = "Intel SPMD Program Compiler"
 # This could be handy for archiving the generated documentation or
 # if some version control system is used.
 
-PROJECT_NUMBER         = 1.4.1
+PROJECT_NUMBER         = 1.4.2
 
 # The OUTPUT_DIRECTORY tag is used to specify the (relative or absolute)
 # base path where the generated documentation will be put.

ispc.h

@@ -38,7 +38,7 @@
 #ifndef ISPC_H
 #define ISPC_H
 
-#define ISPC_VERSION "1.4.1dev"
+#define ISPC_VERSION "1.4.2"
 
 #if !defined(LLVM_3_1) && !defined(LLVM_3_2) && !defined(LLVM_3_3) && !defined(LLVM_3_4)
 #error "Only LLVM 3.1, 3.2, 3.3 and the 3.4 development branch are supported"

llvm_patches/r183327-AVX2-GATHER.patch

@@ -0,0 +1,54 @@
+This patch needs to be applied to LLVM 3.2/3.3 to fix bunch of fails on AVX2 target.
+LLVM 3.4 contains this fix (r183327).
+
+Index: lib/Target/X86/X86ISelDAGToDAG.cpp
+===================================================================
+--- lib/Target/X86/X86ISelDAGToDAG.cpp	(revision 183626)
++++ lib/Target/X86/X86ISelDAGToDAG.cpp	(working copy)
+@@ -2013,6 +2013,8 @@
+     case Intrinsic::x86_avx2_gather_d_d_256:
+     case Intrinsic::x86_avx2_gather_q_d:
+     case Intrinsic::x86_avx2_gather_q_d_256: {
++      if (!Subtarget->hasAVX2())
++        break;
+       unsigned Opc;
+       switch (IntNo) {
+       default: llvm_unreachable("Impossible intrinsic");
+Index: lib/Target/X86/X86InstrSSE.td
+===================================================================
+--- lib/Target/X86/X86InstrSSE.td	(revision 183626)
++++ lib/Target/X86/X86InstrSSE.td	(working copy)
+@@ -8367,7 +8367,9 @@
+             []>, VEX_4VOp3, VEX_L;
+ }
+ 
+-let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
++let mayLoad = 1, Constraints
++  = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
++  in {
+   defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
+   defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
+   defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
+Index: test/CodeGen/X86/avx2-gather.ll
+===================================================================
+--- test/CodeGen/X86/avx2-gather.ll	(revision 0)
++++ test/CodeGen/X86/avx2-gather.ll	(working copy)
+@@ -0,0 +1,18 @@
++; RUN: not llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx
++; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=core-avx2 | FileCheck %s
++
++declare <4 x float> @llvm.x86.avx2.gather.d.ps(<4 x float>, i8*,
++                      <4 x i32>, <4 x float>, i8) nounwind readonly
++
++define <4 x float> @test_x86_avx2_gather_d_ps(i8* %a1,
++                     <4 x i32> %idx, <4 x float> %mask) {
++  %res = call <4 x float> @llvm.x86.avx2.gather.d.ps(<4 x float> undef,
++                            i8* %a1, <4 x i32> %idx, <4 x float> %mask, i8 2) ;
++  ret <4 x float> %res
++}
++
++; CHECK: test_x86_avx2_gather_d_ps
++; CHECK: vgatherdps
++; CHECK-NOT: [[DST]]
++; CHECK: [[DST:%xmm[0-9]+]]{{$}}
++; CHECK: ret