Small IR cleanups. (#2441)

5 files changed, 99 insertions, 48 deletions

diff --git a/source/slang/slang-emit.cpp b/source/slang/slang-emit.cpp
index a916d0d63..6d85e1ce0 100644
--- a/source/slang/slang-emit.cpp
+++ b/source/slang/slang-emit.cpp
@@ -830,6 +830,9 @@ Result linkAndOptimizeIR(
         }
     }
 
+    // Run a final round of DCE to clean up unused things after phi-elimination.
+    eliminateDeadCode(irModule);
+
     // We include one final step to (optionally) dump the IR and validate
     // it after all of the optimization passes are complete. This should
     // reflect the IR that code is generated from as closely as possible.
diff --git a/source/slang/slang-ir-eliminate-phis.cpp b/source/slang/slang-ir-eliminate-phis.cpp
index c4f0b8b9d..07d0e7374 100644
--- a/source/slang/slang-ir-eliminate-phis.cpp
+++ b/source/slang/slang-ir-eliminate-phis.cpp
@@ -807,7 +807,10 @@ struct PhiEliminationContext
             // so that any logic that might have moved another parameter
             // into a temporary will influence our result.
             //
-            m_builder.emitStore(dstParam.temp, *srcArg.currentValPtr);
+            if ((*srcArg.currentValPtr)->getOp() != kIROp_undefined)
+            {
+                m_builder.emitStore(dstParam.temp, *srcArg.currentValPtr);
+            }
 
             //
             // Once the store is emitted, the assignment has been performed,
diff --git a/source/slang/slang-ir-simplify-cfg.cpp b/source/slang/slang-ir-simplify-cfg.cpp
index db0274d32..a1bc38b64 100644
--- a/source/slang/slang-ir-simplify-cfg.cpp
+++ b/source/slang/slang-ir-simplify-cfg.cpp
@@ -12,6 +12,9 @@ bool processFunc(IRGlobalValueWithCode* func)
     if (!firstBlock)
         return false;
 
+    SharedIRBuilder sharedBuilder(func->getModule());
+    IRBuilder builder(&sharedBuilder);
+
     bool changed = false;
 
     List<IRBlock*> workList;
@@ -25,12 +28,27 @@ bool processFunc(IRGlobalValueWithCode* func)
         {
             if (auto loop = as<IRLoop>(block->getTerminator()))
             {
+                // If continue block is unreachable, remove it.
                 auto continueBlock = loop->getContinueBlock();
                 if (continueBlock && !continueBlock->hasMoreThanOneUse())
                 {
                     loop->continueBlock.set(loop->getTargetBlock());
                     continueBlock->removeAndDeallocate();
                 }
+                // If there isn't any actual back jumps into loop target, remove the header
+                // and turn it into a normal branch.
+                auto targetBlock = loop->getTargetBlock();
+                if (targetBlock->getPredecessors().getCount() == 1 && *targetBlock->getPredecessors().begin() == block)
+                {
+                    builder.setInsertBefore(loop);
+                    List<IRInst*> args;
+                    for (UInt i = 0; i < loop->getArgCount(); i++)
+                    {
+                        args.add(loop->getArg(i));
+                    }
+                    builder.emitBranch(targetBlock, args.getCount(), args.getBuffer());
+                    loop->removeAndDeallocate();
+                }
             }
             // If `block` does not end with an unconditional branch, bail.
             if (block->getTerminator()->getOp() != kIROp_unconditionalBranch)
diff --git a/tests/experimental/liveness/liveness-7.slang b/tests/experimental/liveness/liveness-7.slang
index 5d8ed48cc..1bf53b09d 100644
--- a/tests/experimental/liveness/liveness-7.slang
+++ b/tests/experimental/liveness/liveness-7.slang
@@ -1,4 +1,6 @@
-//TEST:SIMPLE:-target glsl -entry computeMain -profile cs_6_3 -track-liveness -line-directive-mode none
+//DISABLED_TEST:SIMPLE:-target glsl -entry computeMain -profile cs_6_3 -track-liveness -line-directive-mode none
+
+// This test is disabled because the loop will always be optimized out so no liveness info will be inserted.
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
diff --git a/tests/pipeline/ray-tracing/trace-ray-inline.slang.glsl b/tests/pipeline/ray-tracing/trace-ray-inline.slang.glsl
index e591fc391..ae26ab498 100644
--- a/tests/pipeline/ray-tracing/trace-ray-inline.slang.glsl
+++ b/tests/pipeline/ray-tracing/trace-ray-inline.slang.glsl
@@ -1,6 +1,3 @@
-// trace-ray-inline.slang.glsl
-//TEST_IGNORE_FILE:
-
 #version 460
 #extension GL_EXT_ray_tracing : require
 #extension GL_EXT_ray_query : require
@@ -24,14 +21,6 @@ layout(std140) uniform _S1
     SLANG_ParameterGroup_C_0 _data;
 } C_0;
 
-struct RayDesc_0
-{
-    vec3 Origin_0;
-    float TMin_0;
-    vec3 Direction_0;
-    float TMax_0;
-};
-
 layout(binding = 0)
 uniform accelerationStructureEXT myAccelerationStructure_0;
 
@@ -78,13 +67,17 @@ void myMiss_0(inout MyRayPayload_0 payload_4)
     return;
 }
 
+struct RayDesc_0
+{
+    vec3 Origin_0;
+    float TMin_0;
+    vec3 Direction_0;
+    float TMax_0;
+};
+
 layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
 void main()
 {
-    MyProceduralHitAttrs_0 committedProceduralAttrs_0;
-    MyProceduralHitAttrs_0 committedProceduralAttrs_1;
-    MyProceduralHitAttrs_0 committedProceduralAttrs_2;
-    MyProceduralHitAttrs_0 committedProceduralAttrs_3;
 
     rayQueryEXT query_0;
 
@@ -95,63 +88,87 @@ void main()
     RayDesc_0 ray_0 = { C_0._data.origin_0, C_0._data.tMin_0, C_0._data.direction_0, C_0._data.tMax_0 };
     rayQueryInitializeEXT((query_0), (myAccelerationStructure_0), (C_0._data.rayFlags_0 | 512), (C_0._data.instanceMask_0), (ray_0.Origin_0), (ray_0.TMin_0), (ray_0.Direction_0), (ray_0.TMax_0));
 
-    MyProceduralHitAttrs_0 _S3;
-    committedProceduralAttrs_0 = _S3;
+    MyProceduralHitAttrs_0 committedProceduralAttrs_0;
 
     for(;;)
     {
-        bool _S4 = rayQueryProceedEXT(query_0);
-        if(!_S4)
+
+        bool _S3 = rayQueryProceedEXT(query_0);
+
+        if(!_S3)
         {
             break;
         }
-        uint _S5 = (rayQueryGetIntersectionTypeEXT((query_0), false));
-        switch(_S5)
+        uint _S4 = (rayQueryGetIntersectionTypeEXT((query_0), false));
+
+        MyProceduralHitAttrs_0 committedProceduralAttrs_1;
+
+        switch(_S4)
         {
-        case uint(1):
+        case 1U:
             {
                 MyProceduralHitAttrs_0 candidateProceduralAttrs_0;
-                MyProceduralHitAttrs_0 _S6 = { 0 };
-                candidateProceduralAttrs_0 = _S6;
+
+                MyProceduralHitAttrs_0 _S5 = { 0 };
+
+                candidateProceduralAttrs_0 = _S5;
                 float tHit_1;
                 tHit_1 = 0.00000000000000000000;
-                bool _S7 = myProceduralIntersection_0(tHit_1, candidateProceduralAttrs_0);
-                if(_S7)
+                bool _S6 = myProceduralIntersection_0(tHit_1, candidateProceduralAttrs_0);
+
+                MyProceduralHitAttrs_0 committedProceduralAttrs_2;
+
+                if(_S6)
                 {
-                    bool _S8 = myProceduralAnyHit_0(payload_5);
-                    if(_S8)
+                    bool _S7 = myProceduralAnyHit_0(payload_5);
+
+                    MyProceduralHitAttrs_0 committedProceduralAttrs_3;
+
+                    if(_S7)
                     {
                         rayQueryGenerateIntersectionEXT(query_0, tHit_1);
-                        MyProceduralHitAttrs_0 _S9 = candidateProceduralAttrs_0;
-                        if(bool(C_0._data.shouldStopAtFirstHit_0))
+                        MyProceduralHitAttrs_0 _S8 = candidateProceduralAttrs_0;
+                        if(C_0._data.shouldStopAtFirstHit_0 != 0U)
                         {
                             rayQueryTerminateEXT(query_0);
                         }
                         else
                         {
                         }
-                        committedProceduralAttrs_1 = _S9;
+
+                        committedProceduralAttrs_3 = _S8;
+
                     }
                     else
                     {
-                        committedProceduralAttrs_1 = committedProceduralAttrs_0;
+
+                        committedProceduralAttrs_3 = committedProceduralAttrs_0;
+
                     }
-                    committedProceduralAttrs_2 = committedProceduralAttrs_1;
+
+                    committedProceduralAttrs_2 = committedProceduralAttrs_3;
+
                 }
                 else
                 {
+
                     committedProceduralAttrs_2 = committedProceduralAttrs_0;
+
                 }
-                committedProceduralAttrs_3 = committedProceduralAttrs_2;
+
+                committedProceduralAttrs_1 = committedProceduralAttrs_2;
+
                 break;
             }
-        case uint(0):
+        case 0U:
             {
-                bool _S10 = myTriangleAnyHit_0(payload_5);
-                if(_S10)
+
+                bool _S9 = myTriangleAnyHit_0(payload_5);
+
+                if(_S9)
                 {
                     rayQueryConfirmIntersectionEXT(query_0);
-                    if(bool(C_0._data.shouldStopAtFirstHit_0))
+                    if(C_0._data.shouldStopAtFirstHit_0 != 0U)
                     {
                         rayQueryTerminateEXT(query_0);
                     }
@@ -162,31 +179,39 @@ void main()
                 else
                 {
                 }
-                committedProceduralAttrs_3 = committedProceduralAttrs_0;
+
+                committedProceduralAttrs_1 = committedProceduralAttrs_0;
+
                 break;
             }
         default:
             {
-                committedProceduralAttrs_3 = committedProceduralAttrs_0;
+
+                committedProceduralAttrs_1 = committedProceduralAttrs_0;
+
                 break;
             }
         }
-        committedProceduralAttrs_0 = committedProceduralAttrs_3;
+
+        committedProceduralAttrs_0 = committedProceduralAttrs_1;
+
     }
-    uint _S11 = (rayQueryGetIntersectionTypeEXT((query_0), true));
-    switch(_S11)
+
+    uint _S10 = (rayQueryGetIntersectionTypeEXT((query_0), true));
+
+    switch(_S10)
     {
-    case uint(1):
+    case 1U:
         {
             myTriangleClosestHit_0(payload_5);
             break;
         }
-    case uint(2):
+    case 2U:
         {
             myProceduralClosestHit_0(payload_5, committedProceduralAttrs_0);
             break;
         }
-    case uint(0):
+    case 0U:
         {
             myMiss_0(payload_5);
             break;