Register allocation during phi elimination. (#2613)

* Register allocation during phi elimination.

* Enhance the test case.

* Cleanup line breaks in test case.

* remove unncessary line break changes.

* More cleanups.

---------

Co-authored-by: Yong He <yhe@nvidia.com>

7 files changed, 105 insertions, 61 deletions

diff --git a/tests/experimental/liveness/liveness-3.slang.expected b/tests/experimental/liveness/liveness-3.slang.expected
index 58f562d86..dac9ff1bd 100644
--- a/tests/experimental/liveness/liveness-3.slang.expected
+++ b/tests/experimental/liveness/liveness-3.slang.expected
@@ -52,13 +52,13 @@ int calcThing_0(int offset_0)
         livenessStart_2(idx_0, 0);
         const int  _S3[3] = { 0, 0, 0 };
         idx_0 = _S3;
-        int _S4 = _S2;
         int i_0;
-        int _S5;
+        int _S4;
+        int _S5 = _S2;
         livenessStart_1(i_0, 0);
         i_0 = 0;
-        livenessStart_1(_S5, 0);
-        _S5 = _S4;
+        livenessStart_1(_S4, 0);
+        _S4 = _S5;
         for(;;)
         {
             if(i_0 < 17)
@@ -74,32 +74,32 @@ int calcThing_0(int offset_0)
             int _S7;
             if(_S6 != 0)
             {
-                int _S8 = _S5;
-                livenessEnd_0(_S5, 0);
+                int _S8 = _S4;
+                livenessEnd_0(_S4, 0);
                 int _S9 = _S8 + 1;
                 livenessStart_1(_S7, 0);
                 _S7 = _S9;
             }
             else
             {
-                int _S10 = _S5;
-                livenessEnd_0(_S5, 0);
+                int _S10 = _S4;
+                livenessEnd_0(_S4, 0);
                 livenessStart_1(_S7, 0);
                 _S7 = _S10;
             }
             idx_0[modRange_0] = idx_0[modRange_0] + (_S7 + i_0);
             i_0 = i_0 + 1;
-            livenessStart_1(_S5, 0);
+            livenessStart_1(_S4, 0);
             int _S11 = _S7;
             livenessEnd_0(_S7, 0);
-            _S5 = _S11;
+            _S4 = _S11;
         }
         livenessEnd_0(i_0, 0);
         livenessEnd_0(_S2, 0);
         int _S12 = (k_0 + 7) % 5;
         if(_S12 == 4)
         {
-            livenessEnd_0(_S5, 0);
+            livenessEnd_0(_S4, 0);
             livenessEnd_1(idx_0, 0);
             livenessEnd_0(k_0, 0);
             livenessEnd_2(another_0, 0);
@@ -114,8 +114,8 @@ int calcThing_0(int offset_0)
         int total_1 = _S16 + _S15;
         k_0 = k_0 + 1;
         livenessStart_1(_S2, 0);
-        int _S17 = _S5;
-        livenessEnd_0(_S5, 0);
+        int _S17 = _S4;
+        livenessEnd_0(_S4, 0);
         _S2 = _S17;
         livenessStart_1(total_0, 0);
         total_0 = total_1;
diff --git a/tests/experimental/liveness/liveness-5.slang.expected b/tests/experimental/liveness/liveness-5.slang.expected
index ea6e37036..a8a9707d7 100644
--- a/tests/experimental/liveness/liveness-5.slang.expected
+++ b/tests/experimental/liveness/liveness-5.slang.expected
@@ -72,35 +72,26 @@ int calcThing_0(int offset_0)
     }
     livenessEnd_0(k_0, 0);
     livenessEnd_1(another_0, 0);
-    int total_2;
     if(total_0 > 4)
     {
-        int _S5 = total_0;
-        livenessEnd_0(total_0, 0);
-        int _S6 = - _S5;
-        livenessStart_1(total_2, 0);
-        total_2 = _S6;
+        total_0 = - total_0;
     }
     else
     {
-        int _S7 = total_0;
-        livenessEnd_0(total_0, 0);
-        livenessStart_1(total_2, 0);
-        total_2 = _S7;
     }
-    return total_2;
+    return total_0;
 }
 
-layout(std430, binding = 0) buffer _S8 {
+layout(std430, binding = 0) buffer _S5 {
     int _data[];
 } outputBuffer_0;
 layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in;
 void main()
 {
     int index_0 = int(gl_GlobalInvocationID.x);
-    uint _S9 = uint(index_0);
-    int _S10 = calcThing_0(index_0);
-    ((outputBuffer_0)._data[(_S9)]) = _S10;
+    uint _S6 = uint(index_0);
+    int _S7 = calcThing_0(index_0);
+    ((outputBuffer_0)._data[(_S6)]) = _S7;
     return;
 }
 
diff --git a/tests/experimental/liveness/liveness-6.slang.expected b/tests/experimental/liveness/liveness-6.slang.expected
index 26a537330..402e19886 100644
--- a/tests/experimental/liveness/liveness-6.slang.expected
+++ b/tests/experimental/liveness/liveness-6.slang.expected
@@ -81,35 +81,26 @@ int calcThing_0(int offset_0)
     }
     livenessEnd_0(k_0, 0);
     livenessEnd_1(another_0, 0);
-    int total_3;
     if(total_0 > 4)
     {
-        int _S8 = total_0;
-        livenessEnd_0(total_0, 0);
-        int _S9 = - _S8;
-        livenessStart_1(total_3, 0);
-        total_3 = _S9;
+        total_0 = - total_0;
     }
     else
     {
-        int _S10 = total_0;
-        livenessEnd_0(total_0, 0);
-        livenessStart_1(total_3, 0);
-        total_3 = _S10;
     }
-    return total_3;
+    return total_0;
 }
 
-layout(std430, binding = 0) buffer _S11 {
+layout(std430, binding = 0) buffer _S8 {
     int _data[];
 } outputBuffer_0;
 layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in;
 void main()
 {
     int index_0 = int(gl_GlobalInvocationID.x);
-    uint _S12 = uint(index_0);
-    int _S13 = calcThing_0(index_0);
-    ((outputBuffer_0)._data[(_S12)]) = _S13;
+    uint _S9 = uint(index_0);
+    int _S10 = calcThing_0(index_0);
+    ((outputBuffer_0)._data[(_S9)]) = _S10;
     return;
 }
 
diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang.1.expected
index 15221b921..847eab926 100644
--- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang.1.expected
+++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang.1.expected
@@ -65,16 +65,14 @@ uint calcValue_0(hitObjectNV hit_0)
     else
     {
         bool _S7 = (hitObjectIsMissNV((hit_0)));
-        uint r_3;
         if(_S7)
         {
-            r_3 = 1U;
+            r_0 = 1U;
         }
         else
         {
-            r_3 = 0U;
+            r_0 = 0U;
         }
-        r_0 = r_3;
     }
     return r_0;
 }
@@ -96,13 +94,13 @@ void main()
     RayDesc_0 _S11 = ray_2;
     hitObjectNV hitObj_0;
     hitObjectRecordHitWithIndexNV(hitObj_0, scene_0, int(uint(idx_0)), int(uint(idx_0 * 2)), int(uint(idx_0 * 3)), 0U, 0U, _S11.Origin_0, _S11.TMin_0, _S11.Direction_0, _S11.TMax_0, (0));
-    uint r_4 = calcValue_0(hitObj_0);
+    uint r_3 = calcValue_0(hitObj_0);
     RayDesc_0 _S12 = ray_2;
     hitObjectNV hitObj_1;
     hitObjectRecordHitNV(hitObj_1, scene_0, int(uint(idx_0)), int(uint(idx_0 * 3)), int(uint(idx_0 * 2)), 0U, 0U, 4U, _S12.Origin_0, _S12.TMin_0, _S12.Direction_0, _S12.TMax_0, (0));
     uint _S13 = calcValue_0(hitObj_1);
-    uint r_5 = r_4 + _S13;
-    ((outputBuffer_0)._data[(uint(idx_0))]) = r_5;
+    uint r_4 = r_3 + _S13;
+    ((outputBuffer_0)._data[(uint(idx_0))]) = r_4;
     return;
 }
 
diff --git a/tests/ir/ssa-reg-alloc.slang b/tests/ir/ssa-reg-alloc.slang
new file mode 100644
index 000000000..3bfd795a8
--- /dev/null
+++ b/tests/ir/ssa-reg-alloc.slang
@@ -0,0 +1,68 @@
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<uint> outputBuffer;
+
+int test1(uint p)
+{
+    int a, b;
+    if (p > 1)
+    {
+        a = 1;
+        b = 2;
+    }
+    else
+    {
+        a = 2;
+        b = 3;
+    }
+    // b is not used and should not interfere the result of a.
+    return a;
+}
+
+int test2(uint p)
+{
+    int a, b;
+    if (p > 1)
+    {
+        a = 1;
+        b = 2;
+    }
+    else
+    {
+        a = 2;
+        b = 3;
+    }
+    // a is not used and should not interfere the result of b.
+    return b;
+}
+
+int test3(uint p)
+{
+    int a = 1;
+    int b = 5;
+    
+    if (p > 0) a = 2;
+    if (p > 0) b = 3;
+
+    // a and b are now register allocated.
+    // The first block of the loop will have IRParams in the form of (a, b)
+    for (int i = 0; i <= p + 2; i++)
+    {
+        let tmp = a;
+        a = b;
+        b = tmp;
+        // The branch back to the loop header will have phi args: (b, a)
+        // Phi-elmination must handle this case of concurrent assignment correctly.
+    }
+    return a - b; // should be 4 when p == 0.
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    let rs1 = test1(dispatchThreadID.x) + test2(dispatchThreadID.x);
+    outputBuffer[0] = rs1;
+    outputBuffer[1] = test3(0);
+}
diff --git a/tests/ir/ssa-reg-alloc.slang.expected.txt b/tests/ir/ssa-reg-alloc.slang.expected.txt
new file mode 100644
index 000000000..e43a2b945
--- /dev/null
+++ b/tests/ir/ssa-reg-alloc.slang.expected.txt
@@ -0,0 +1,4 @@
+5
+4
+0
+0
diff --git a/tests/pipeline/ray-tracing/trace-ray-inline.slang.glsl b/tests/pipeline/ray-tracing/trace-ray-inline.slang.glsl
index f9e94bb30..389dae05a 100644
--- a/tests/pipeline/ray-tracing/trace-ray-inline.slang.glsl
+++ b/tests/pipeline/ray-tracing/trace-ray-inline.slang.glsl
@@ -106,14 +106,10 @@ void main()
                 tHit_1 = 0.00000000000000000000;
                 bool _S6 = myProceduralIntersection_0(tHit_1, candidateProceduralAttrs_0);
 
-                MyProceduralHitAttrs_0 committedProceduralAttrs_2;
-
                 if(_S6)
                 {
                     bool _S7 = myProceduralAnyHit_0(payload_5);
 
-                    MyProceduralHitAttrs_0 committedProceduralAttrs_3;
-
                     if(_S7)
                     {
                         rayQueryGenerateIntersectionEXT(query_0, tHit_1);
@@ -126,28 +122,24 @@ void main()
                         {
                         }
 
-                        committedProceduralAttrs_3 = _S8;
+                        committedProceduralAttrs_1 = _S8;
 
                     }
                     else
                     {
 
-                        committedProceduralAttrs_3 = committedProceduralAttrs_0;
+                        committedProceduralAttrs_1 = committedProceduralAttrs_0;
 
                     }
 
-                    committedProceduralAttrs_2 = committedProceduralAttrs_3;
-
                 }
                 else
                 {
 
-                    committedProceduralAttrs_2 = committedProceduralAttrs_0;
+                    committedProceduralAttrs_1 = committedProceduralAttrs_0;
 
                 }
 
-                committedProceduralAttrs_1 = committedProceduralAttrs_2;
-
                 break;
             }
         case 0U: