swr/rast: Add autogen of helper llvm intrinsics.

Replace sqrt, maskload, fp min/max, cttz, ctlz with llvm equivalent.
Replace AVX maskedstore intrinsic with LLVM intrinsic. Add helper llvm
macros for stacksave, stackrestore, popcnt.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

13 files changed, 130 insertions, 126 deletions

diff --git a/src/gallium/drivers/swr/Makefile.am b/src/gallium/drivers/swr/Makefile.am
index 5ec9213c34..32dd9e59ff 100644
--- a/src/gallium/drivers/swr/Makefile.am
+++ b/src/gallium/drivers/swr/Makefile.am
@@ -81,6 +81,7 @@ BUILT_SOURCES = \
 	rasterizer/jitter/gen_state_llvm.h \
 	rasterizer/jitter/gen_builder.hpp \
 	rasterizer/jitter/gen_builder_x86.hpp \
+	rasterizer/jitter/gen_builder_intrin.hpp \
 	rasterizer/archrast/gen_ar_event.hpp \
 	rasterizer/archrast/gen_ar_event.cpp \
 	rasterizer/archrast/gen_ar_eventhandler.hpp \
@@ -140,6 +141,13 @@ rasterizer/jitter/gen_builder_x86.hpp: rasterizer/codegen/gen_llvm_ir_macros.py
 		--output rasterizer/jitter \
 		--gen_x86_h
 
+rasterizer/jitter/gen_builder_intrin.hpp: rasterizer/codegen/gen_llvm_ir_macros.py rasterizer/codegen/templates/gen_builder.hpp rasterizer/codegen/gen_common.py
+	$(MKDIR_GEN)
+	$(PYTHON_GEN) \
+		$(srcdir)/rasterizer/codegen/gen_llvm_ir_macros.py \
+		--output rasterizer/jitter \
+		--gen_intrin_h
+
 rasterizer/archrast/gen_ar_event.hpp: rasterizer/codegen/gen_archrast.py rasterizer/codegen/templates/gen_ar_event.hpp rasterizer/archrast/events.proto rasterizer/archrast/events_private.proto rasterizer/codegen/gen_common.py
 	$(MKDIR_GEN)
 	$(PYTHON_GEN) \
diff --git a/src/gallium/drivers/swr/SConscript b/src/gallium/drivers/swr/SConscript
index cc4025b0e5..5097be67bb 100644
--- a/src/gallium/drivers/swr/SConscript
+++ b/src/gallium/drivers/swr/SConscript
@@ -85,6 +85,15 @@ Depends('rasterizer/jitter/gen_builder.hpp',
         swrroot + 'rasterizer/codegen/templates/gen_builder.hpp')
 
 env.CodeGenerate(
+    target = 'rasterizer/jitter/gen_builder_intrin.hpp',
+    script = swrroot + 'rasterizer/codegen/gen_llvm_ir_macros.py',
+    source = '',
+    command = python_cmd + ' $SCRIPT --output ' + bldroot + '/rasterizer/jitter --gen_intrin_h'
+)
+Depends('rasterizer/jitter/gen_builder.hpp',
+        swrroot + 'rasterizer/codegen/templates/gen_builder.hpp')
+
+env.CodeGenerate(
     target = './gen_swr_context_llvm.h',
     script = swrroot + 'rasterizer/codegen/gen_llvm_types.py',
     source = 'swr_context.h',
diff --git a/src/gallium/drivers/swr/meson.build b/src/gallium/drivers/swr/meson.build
index e722602403..91e03abbeb 100644
--- a/src/gallium/drivers/swr/meson.build
+++ b/src/gallium/drivers/swr/meson.build
@@ -300,7 +300,7 @@ endif
 libmesaswr = static_library(
   'mesaswr',
   [files_swr_mesa, files_swr_common, gen_knobs_h, gen_knobs_cpp,
-   gen_builder_hpp, gen_builder_x86_hpp],
+   gen_builder_hpp, gen_builder_x86_hpp, gen_builder_intrin_hpp],
   cpp_args : [cpp_vis_args, swr_cpp_args, swr_avx_args, swr_arch_defines],
   include_directories : [inc_common, swr_incs],
   dependencies : dep_llvm,
diff --git a/src/gallium/drivers/swr/rasterizer/codegen/gen_llvm_ir_macros.py b/src/gallium/drivers/swr/rasterizer/codegen/gen_llvm_ir_macros.py
index 3e1fbfe88e..9dfc1e7559 100644
--- a/src/gallium/drivers/swr/rasterizer/codegen/gen_llvm_ir_macros.py
+++ b/src/gallium/drivers/swr/rasterizer/codegen/gen_llvm_ir_macros.py
@@ -42,32 +42,40 @@ inst_aliases = {
 }
 
 intrinsics = [
-        ['VGATHERPD', 'x86_avx2_gather_d_pd_256', ['src', 'pBase', 'indices', 'mask', 'scale']],
-        ['VGATHERPS', 'x86_avx2_gather_d_ps_256', ['src', 'pBase', 'indices', 'mask', 'scale']],
-        ['VGATHERPS_16', 'x86_avx512_gather_dps_512', ['src', 'pBase', 'indices', 'mask', 'scale']],
-        ['VGATHERDD', 'x86_avx2_gather_d_d_256', ['src', 'pBase', 'indices', 'mask', 'scale']],
-        ['VGATHERDD_16', 'x86_avx512_gather_dpi_512', ['src', 'pBase', 'indices', 'mask', 'scale']],
-        ['VSQRTPS', 'x86_avx_sqrt_ps_256', ['a']],
-        ['VRSQRTPS', 'x86_avx_rsqrt_ps_256', ['a']],
-        ['VRCPPS', 'x86_avx_rcp_ps_256', ['a']],
-        ['VMINPS', 'x86_avx_min_ps_256', ['a', 'b']],
-        ['VMAXPS', 'x86_avx_max_ps_256', ['a', 'b']],
-        ['VROUND', 'x86_avx_round_ps_256', ['a', 'rounding']],
-        ['BEXTR_32', 'x86_bmi_bextr_32', ['src', 'control']],
-        ['VPSHUFB', 'x86_avx2_pshuf_b', ['a', 'b']],
-        ['VPERMD', 'x86_avx2_permd', ['a', 'idx']],
-        ['VPERMPS', 'x86_avx2_permps', ['idx', 'a']],
-        ['VCVTPD2PS', 'x86_avx_cvt_pd2_ps_256', ['a']],
-        ['VCVTPH2PS', 'x86_vcvtph2ps_256', ['a']],
-        ['VCVTPS2PH', 'x86_vcvtps2ph_256', ['a', 'round']],
-        ['VHSUBPS', 'x86_avx_hsub_ps_256', ['a', 'b']],
-        ['VPTESTC', 'x86_avx_ptestc_256', ['a', 'b']],
-        ['VPTESTZ', 'x86_avx_ptestz_256', ['a', 'b']],
-        ['VFMADDPS', 'x86_fma_vfmadd_ps_256', ['a', 'b', 'c']],
-        ['VMOVMSKPS', 'x86_avx_movmsk_ps_256', ['a']],
-        ['INTERRUPT', 'x86_int', ['a']],
-        ['VPHADDD', 'x86_avx2_phadd_d', ['a', 'b']],
-    ]
+    ['VGATHERPD', 'x86_avx2_gather_d_pd_256', ['src', 'pBase', 'indices', 'mask', 'scale']],
+    ['VGATHERPS', 'x86_avx2_gather_d_ps_256', ['src', 'pBase', 'indices', 'mask', 'scale']],
+    ['VGATHERPS_16', 'x86_avx512_gather_dps_512', ['src', 'pBase', 'indices', 'mask', 'scale']],
+    ['VGATHERDD', 'x86_avx2_gather_d_d_256', ['src', 'pBase', 'indices', 'mask', 'scale']],
+    ['VGATHERDD_16', 'x86_avx512_gather_dpi_512', ['src', 'pBase', 'indices', 'mask', 'scale']],
+    ['VRCPPS', 'x86_avx_rcp_ps_256', ['a']],
+    ['VROUND', 'x86_avx_round_ps_256', ['a', 'rounding']],
+    ['BEXTR_32', 'x86_bmi_bextr_32', ['src', 'control']],
+    ['VPSHUFB', 'x86_avx2_pshuf_b', ['a', 'b']],
+    ['VPERMD', 'x86_avx2_permd', ['a', 'idx']],
+    ['VPERMPS', 'x86_avx2_permps', ['idx', 'a']],
+    ['VCVTPD2PS', 'x86_avx_cvt_pd2_ps_256', ['a']],
+    ['VCVTPH2PS', 'x86_vcvtph2ps_256', ['a']],
+    ['VCVTPS2PH', 'x86_vcvtps2ph_256', ['a', 'round']],
+    ['VHSUBPS', 'x86_avx_hsub_ps_256', ['a', 'b']],
+    ['VPTESTC', 'x86_avx_ptestc_256', ['a', 'b']],
+    ['VPTESTZ', 'x86_avx_ptestz_256', ['a', 'b']],
+    ['VFMADDPS', 'x86_fma_vfmadd_ps_256', ['a', 'b', 'c']],
+    ['VMOVMSKPS', 'x86_avx_movmsk_ps_256', ['a']],
+    ['INTERRUPT', 'x86_int', ['a']],
+    ['VPHADDD', 'x86_avx2_phadd_d', ['a', 'b']],
+]
+
+llvm_intrinsics = [
+    ['CTTZ', 'cttz', ['a', 'flag'], ['a']],
+    ['CTLZ', 'ctlz', ['a', 'flag'], ['a']],
+    ['VSQRTPS', 'sqrt', ['a'], ['a']],
+    ['STACKSAVE', 'stacksave', [], []],
+    ['STACKRESTORE', 'stackrestore', ['a'], []],
+    ['VMINPS', 'minnum', ['a', 'b'], ['a']],
+    ['VMAXPS', 'maxnum', ['a', 'b'], ['a']],
+    ['DEBUGTRAP', 'debugtrap', [], []],
+    ['POPCNT', 'ctpop', ['a'], ['a']]
+]
 
 this_dir = os.path.dirname(os.path.abspath(__file__))
 template = os.path.join(this_dir, 'templates', 'gen_builder.hpp')
@@ -195,7 +203,7 @@ def generate_gen_h(functions, output_dir):
         templfuncs.append({
             'decl'      : decl,
             'intrin'    : func['name'],
-            'args'      : ', '.join(func['arg_names']),
+            'args'      : func['arg_names'],
         })
 
     MakoTemplateWriter.to_file(
@@ -205,7 +213,7 @@ def generate_gen_h(functions, output_dir):
         comment='Builder IR Wrappers',
         filename=filename,
         functions=templfuncs,
-        isX86=False)
+        isX86=False, isIntrin=False)
 
 '''
     Auto-generates macros for LLVM IR
@@ -221,8 +229,8 @@ def generate_x86_h(output_dir):
 
         functions.append({
             'decl'      : 'Value* %s(%s, const llvm::Twine& name = "")' % (inst[0], declargs),
-            'args'      : ', '.join(inst[2]),
             'intrin'    : inst[1],
+            'args'      : inst[2],
         })
 
     MakoTemplateWriter.to_file(
@@ -232,8 +240,36 @@ def generate_x86_h(output_dir):
         comment='x86 intrinsics',
         filename=filename,
         functions=functions,
-        isX86=True)
+        isX86=True, isIntrin=False)
 
+def generate_intrin_h(output_dir):
+    filename = 'gen_builder_intrin.hpp'
+    output_filename = os.path.join(output_dir, filename)
+
+    functions = []
+    for inst in llvm_intrinsics:
+        #print('Inst: %s, x86: %s numArgs: %d' % (inst[0], inst[1], len(inst[2])))
+        if len(inst[2]) != 0:
+            declargs = 'Value* ' + ', Value* '.join(inst[2])
+            decl = 'Value* %s(%s, const llvm::Twine& name = "")' % (inst[0], declargs)
+        else:
+            decl = 'Value* %s(const llvm::Twine& name = "")' % (inst[0])
+
+        functions.append({
+            'decl'      : decl,
+            'intrin'    : inst[1],
+            'args'      : inst[2],
+            'types'     : inst[3],
+        })
+
+    MakoTemplateWriter.to_file(
+        template,
+        output_filename,
+        cmdline=sys.argv,
+        comment='llvm intrinsics',
+        filename=filename,
+        functions=functions,
+        isX86=False, isIntrin=True)
 '''
     Function which is invoked when this script is started from a command line.
     Will present and consume a set of arguments which will tell this script how
@@ -247,6 +283,7 @@ def main():
     parser.add_argument('--output-dir', '-o', action='store', dest='output', help='Path to output directory', required=True)
     parser.add_argument('--gen_h', help='Generate builder_gen.h', action='store_true', default=False)
     parser.add_argument('--gen_x86_h', help='Generate x86 intrinsics. No input is needed.', action='store_true', default=False)
+    parser.add_argument('--gen_intrin_h', help='Generate llvm intrinsics. No input is needed.', action='store_true', default=False)
     args = parser.parse_args()
 
     if not os.path.exists(args.output):
@@ -264,6 +301,9 @@ def main():
     if args.gen_x86_h:
         generate_x86_h(args.output)
 
+    if args.gen_intrin_h:
+        generate_intrin_h(args.output)
+
 if __name__ == '__main__':
     main()
 # END OF FILE
diff --git a/src/gallium/drivers/swr/rasterizer/codegen/templates/gen_builder.hpp b/src/gallium/drivers/swr/rasterizer/codegen/templates/gen_builder.hpp
index b6cf03e92f..5a47c9aa10 100644
--- a/src/gallium/drivers/swr/rasterizer/codegen/templates/gen_builder.hpp
+++ b/src/gallium/drivers/swr/rasterizer/codegen/templates/gen_builder.hpp
@@ -35,15 +35,27 @@
 //============================================================================
 // Auto-generated ${comment}
 //============================================================================
-
 %for func in functions:
+<%argList = ', '.join(func['args'])%>\
 ${func['decl']}
 {
 %if isX86:
-    Function *pFunc = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::${func['intrin']});
-    return CALL(pFunc, std::initializer_list<Value*>{${func['args']}}, name);
+    Function * pFunc = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::${func['intrin']});
+    return CALL(pFunc, std::initializer_list<Value*>{${argList}}, name);
+%elif isIntrin:
+    %if len(func['types']) != 0:
+    SmallVector<Type*, ${len(func['types'])}> args;
+    %for arg in func['types']:
+    args.push_back(${arg}->getType());
+    %endfor
+    Function * pFunc = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::${func['intrin']}, args);
+    return CALL(pFunc, std::initializer_list<Value*>{${argList}}, name);
+    %else:
+    Function * pFunc = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::${func['intrin']});
+    return CALL(pFunc, std::initializer_list<Value*>{${argList}}, name);
+    %endif
 %else:
-    return IRB()->${func['intrin']}(${func['args']});
+    return IRB()->${func['intrin']}(${argList});
 %endif
 }
 
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder.h b/src/gallium/drivers/swr/rasterizer/jitter/builder.h
index 763d29fdf6..516e872eb0 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder.h
@@ -92,6 +92,7 @@ namespace SwrJit
 
 #include "gen_builder.hpp"
 #include "gen_builder_x86.hpp"
+#include "gen_builder_intrin.hpp"
 #include "builder_misc.h"
 #include "builder_math.h"
 #include "builder_mem.h"
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp
index cc0f897e48..a825434de7 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp
@@ -129,30 +129,6 @@ namespace SwrJit
         return STORE(val, GEPA(basePtr, valIndices));
     }
 
-    //////////////////////////////////////////////////////////////////////////
-    /// @brief Generate an i32 masked load operation in LLVM IR.  If not  
-    /// supported on the underlying platform, emulate it with float masked load
-    /// @param src - base address pointer for the load
-    /// @param vMask - SIMD wide mask that controls whether to access memory load 0
-    Value *Builder::MASKLOADD(Value* src, Value* mask)
-    {
-        Value* vResult;
-        // use avx2 gather instruction is available
-        if (JM()->mArch.AVX2())
-        {
-            Function *func = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::x86_avx2_maskload_d_256);
-            vResult = CALL(func, { src,mask });
-        }
-        else
-        {
-            // maskload intrinsic expects integer mask operand in llvm >= 3.8
-            mask = BITCAST(mask, VectorType::get(mInt32Ty, mVWidth));
-            Function *func = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::x86_avx_maskload_ps_256);
-            vResult = BITCAST(CALL(func, { src,mask }), VectorType::get(mInt32Ty, mVWidth));
-        }
-        return vResult;
-    }
-
     Value* Builder::OFFSET_TO_NEXT_COMPONENT(Value* base, Constant *offset)
     {
         return GEP(base, offset);
@@ -390,9 +366,7 @@ namespace SwrJit
     /// @param pVecPassthru - SIMD wide vector of values to load when lane is inactive
     Value* Builder::GATHER_PTR(Value* pVecSrcPtr, Value* pVecMask, Value* pVecPassthru)
     {
-        Function* pMaskedGather = llvm::Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::masked_gather, { pVecPassthru->getType(), pVecSrcPtr->getType() });
-
-        return CALL(pMaskedGather, { pVecSrcPtr, C(0), pVecMask, pVecPassthru });
+        return MASKED_GATHER(pVecSrcPtr, 4, pVecMask, pVecPassthru);
     }
 
     void Builder::Gather4(const SWR_FORMAT format, Value* pSrcBase, Value* byteOffsets,
@@ -791,14 +765,11 @@ namespace SwrJit
 
         Value* pMask = VMOVMSKPS(BITCAST(vMask, mSimdFP32Ty));
 
-        // Get cttz function
-        Function* pfnCttz = Intrinsic::getDeclaration(mpJitMgr->mpCurrentModule, Intrinsic::cttz, { mInt32Ty });
-
         // Setup loop basic block
         BasicBlock* pLoop = BasicBlock::Create(mpJitMgr->mContext, "Scatter_Loop", pFunc);
 
         // compute first set bit
-        Value* pIndex = CALL(pfnCttz, { pMask, C(false) });
+        Value* pIndex = CTTZ(pMask, C(false));
 
         Value* pIsUndef = ICMP_EQ(pIndex, C(32));
 
@@ -835,7 +806,7 @@ namespace SwrJit
         Value* pNewMask = AND(pMaskPhi, NOT(SHL(C(1), pIndexPhi)));
 
         // Terminator
-        Value* pNewIndex = CALL(pfnCttz, { pNewMask, C(false) });
+        Value* pNewIndex = CTTZ(pNewMask, C(false));
 
         pIsUndef = ICMP_EQ(pNewIndex, C(32));
         COND_BR(pIsUndef, pPostLoop, pLoop);
@@ -848,19 +819,4 @@ namespace SwrJit
         IRB()->SetInsertPoint(pPostLoop, pPostLoop->begin());
     }
 
-    //////////////////////////////////////////////////////////////////////////
-    /// @brief save/restore stack, providing ability to push/pop the stack and 
-    ///        reduce overall stack requirements for temporary stack use
-    Value* Builder::STACKSAVE()
-    {
-        Function* pfnStackSave = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::stacksave);
-        return CALLA(pfnStackSave);
-    }
-
-    void Builder::STACKRESTORE(Value* pSaved)
-    {
-        Function* pfnStackRestore = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::stackrestore);
-        CALL(pfnStackRestore, std::initializer_list<Value*>{pSaved});
-    }
-
 }
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.h b/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.h
index 5a755e63b4..b538342146 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.h
@@ -60,8 +60,6 @@ LoadInst *LOADV(Value *BasePtr, const std::initializer_list<Value*> &offset, con
 StoreInst *STORE(Value *Val, Value *BasePtr, const std::initializer_list<uint32_t> &offset);
 StoreInst *STOREV(Value *Val, Value *BasePtr, const std::initializer_list<Value*> &offset);
 
-Value *MASKLOADD(Value* src, Value* mask);
-
 void Gather4(const SWR_FORMAT format, Value* pSrcBase, Value* byteOffsets,
     Value* mask, Value* vGatherComponents[], bool bPackedOutput, JIT_MEM_CLIENT usage = MEM_CLIENT_INTERNAL);
 
@@ -87,9 +85,6 @@ void SCATTERPS(Value* pDst, Value* vSrc, Value* vOffsets, Value* vMask);
 void Shuffle8bpcGather4(const SWR_FORMAT_INFO &info, Value* vGatherInput, Value* vGatherOutput[], bool bPackedOutput);
 void Shuffle16bpcGather4(const SWR_FORMAT_INFO &info, Value* vGatherInput[], Value* vGatherOutput[], bool bPackedOutput);
 
-Value* STACKSAVE();
-void STACKRESTORE(Value* pSaved);
-
 // Static stack allocations for scatter operations
 Value* pScatterStackSrc{ nullptr };
 Value* pScatterStackOffsets{ nullptr };
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
index 704b0f259c..c2660181fd 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
@@ -335,13 +335,6 @@ namespace SwrJit
         return CALLA(Callee, args);
     }
 
-    //////////////////////////////////////////////////////////////////////////
-    Value *Builder::DEBUGTRAP()
-    {
-        Function *func = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::debugtrap);
-        return CALL(func);
-    }
-
     Value *Builder::VRCP(Value *va, const llvm::Twine& name)
     {
         return FDIV(VIMMED1(1.0f), va, name);  // 1 / a
@@ -841,12 +834,6 @@ namespace SwrJit
         return vOut;
     }
 
-    Value* Builder::POPCNT(Value* a)
-    {
-        Function* pCtPop = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::ctpop, { a->getType() });
-        return CALL(pCtPop, std::initializer_list<Value*>{a});
-    }
-
     //////////////////////////////////////////////////////////////////////////
     /// @brief pop count on vector mask (e.g. <8 x i1>)
     Value* Builder::VPOPCNT(Value* a)
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
index 549f328ff4..343a9b0367 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
@@ -124,15 +124,6 @@ Value *PMINUD(Value* a, Value* b);
 Value *VABSPS(Value* a);
 Value *FMADDPS(Value* a, Value* b, Value* c);
 
-// LLVM removed VPCMPGTD x86 intrinsic.  This emulates that behavior
-Value *VPCMPGTD(Value* a, Value* b)
-{
-    Value* vIndexMask = ICMP_UGT(a,b);
-
-    // need to set the high bit for x86 intrinsic masks
-    return S_EXT(vIndexMask,VectorType::get(mInt32Ty,JM()->mVWidth));
-}
-
 Value *ICLAMP(Value* src, Value* low, Value* high, const llvm::Twine& name = "");
 Value *FCLAMP(Value* src, Value* low, Value* high);
 Value *FCLAMP(Value* src, float low, float high);
@@ -140,10 +131,8 @@ Value *FCLAMP(Value* src, float low, float high);
 CallInst *PRINT(const std::string &printStr);
 CallInst *PRINT(const std::string &printStr,const std::initializer_list<Value*> &printArgs);
 
-Value* POPCNT(Value* a);
 Value* VPOPCNT(Value* a);
 
-Value* DEBUGTRAP();
 Value* INT3() { return DEBUGTRAP(); }
 
 
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
index 1ee66918d2..5c8d81332d 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
@@ -1884,13 +1884,11 @@ Value* FetchJit::GetSimdValid32bitIndices(Value* pIndices, Value* pLastIndex)
     //     vIndexMask    -1-1-1-1 0 0 0 0 : offsets < max pass
     //     vLoadedIndices 0 1 2 3 0 0 0 0 : offsets >= max masked to 0
     Value* vMaxIndex = VBROADCAST(numIndicesLeft);
-    Value* vIndexMask = VPCMPGTD(vMaxIndex, vIndexOffsets);
-
-    // VMASKLOAD takes an *i8 src pointer
-    pIndices = BITCAST(pIndices,PointerType::get(mInt8Ty,0));
+    Value* vIndexMask = ICMP_SGT(vMaxIndex, vIndexOffsets);
 
     // Load the indices; OOB loads 0
-    return MASKLOADD(pIndices,vIndexMask);
+    pIndices = BITCAST(pIndices, PointerType::get(mSimdInt32Ty, 0));
+    return MASKED_LOAD(pIndices, 4, vIndexMask, VIMMED1(0));
 }
 
 //////////////////////////////////////////////////////////////////////////
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/meson.build b/src/gallium/drivers/swr/rasterizer/jitter/meson.build
index dd1ddcf80c..4a2f46ae1e 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/meson.build
+++ b/src/gallium/drivers/swr/rasterizer/jitter/meson.build
@@ -48,3 +48,14 @@ gen_builder_x86_hpp = custom_target(
   depend_files : swr_gen_builder_depends,
 )
 
+gen_builder_intrin_hpp = custom_target(
+  'gen_builder_intrin.hpp',
+  input : '../codegen/gen_llvm_ir_macros.py',
+  output : 'gen_builder_intrin.hpp',
+  command : [
+    prog_python2, '@INPUT0@', '--gen_intrin_h', '--output', '@OUTPUT@',
+    '--output-dir', '@OUTDIR@'
+  ],
+  depend_files : swr_gen_builder_depends,
+)
+
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp
index ac4436e5eb..f9d858090f 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp
@@ -112,11 +112,11 @@ struct StreamOutJit : public Builder
         {
             if (bitmask & (1 << i))
             {
-                indices.push_back(C(-1.0f));
+                indices.push_back(C(true));
             }
             else
             {
-                indices.push_back(C(0.0f));
+                indices.push_back(C(false));
             }
         }
         return ConstantVector::get(indices);
@@ -131,9 +131,6 @@ struct StreamOutJit : public Builder
     // @param decl - input decl
     void buildDecl(Value* pStream, Value* pOutBuffers[4], const STREAMOUT_DECL& decl)
     {
-        // @todo add this to x86 macros
-        Function* maskStore = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::x86_avx_maskstore_ps);
-
         uint32_t numComponents = _mm_popcnt_u32(decl.componentMask);
         uint32_t packedMask = (1 << numComponents) - 1;
         if (!decl.hole)
@@ -152,15 +149,14 @@ struct StreamOutJit : public Builder
 
             // store to output buffer
             // cast SO buffer to i8*, needed by maskstore
-            Value* pOut = BITCAST(pOutBuffers[decl.bufferIndex], PointerType::get(mInt8Ty, 0));
+            Value* pOut = BITCAST(pOutBuffers[decl.bufferIndex], PointerType::get(simd4Ty, 0));
 
             // cast input to <4xfloat>
             Value* src = BITCAST(vpackedAttrib, simd4Ty);
 
-            // cast mask to <4xint>
+            // cast mask to <4xi1>
             Value* mask = ToMask(packedMask);
-            mask = BITCAST(mask, VectorType::get(IRB()->getInt32Ty(), 4));
-            CALL(maskStore, {pOut, mask, src});
+            MASKED_STORE(src, pOut, 4, mask);
         }
 
         // increment SO buffer
@@ -325,13 +321,15 @@ struct StreamOutJit : public Builder
 /// @return PFN_SO_FUNC - pointer to SOS function
 PFN_SO_FUNC JitStreamoutFunc(HANDLE hJitMgr, const HANDLE hFunc)
 {
-    const llvm::Function *func = (const llvm::Function*)hFunc;
+    llvm::Function *func = (llvm::Function*)hFunc;
     JitManager* pJitMgr = reinterpret_cast<JitManager*>(hJitMgr);
     PFN_SO_FUNC pfnStreamOut;
     pfnStreamOut = (PFN_SO_FUNC)(pJitMgr->mpExec->getFunctionAddress(func->getName().str()));
     // MCJIT finalizes modules the first time you JIT code from them. After finalized, you cannot add new IR to the module
     pJitMgr->mIsModuleFinalized = true;
 
+    pJitMgr->DumpAsm(func, "SoFunc_optimized");
+
     return pfnStreamOut;
 }