Full address insts elimination for backward autodiff. (#2604)

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

4 files changed, 47 insertions, 43 deletions

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 8fc391feb..15221b921 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
@@ -57,15 +57,16 @@ uint calcValue_0(hitObjectNV hit_0)
         uint hitKind_0 = (hitObjectGetHitKindNV((hit_0)));
         uint r_1 = 0U + hitKind_0 + instanceIndex_0 + instanceID_0 + geometryIndex_0 + primitiveIndex_0;
         RayDesc_0 ray_1 = HitObject_GetRayDesc_0(hit_0);
-        uint r_2 = r_1 + uint(ray_1.TMin_0 > 0.00000000000000000000) + uint(ray_1.TMax_0 < ray_1.TMin_0);
+        float _S6 = ray_1.TMin_0;
+        uint r_2 = r_1 + uint(_S6 > 0.00000000000000000000) + uint(ray_1.TMax_0 < _S6);
         SomeValues_0 objSomeValues_0 = HitObject_GetAttributes_0(hit_0);
         r_0 = r_2 + uint(objSomeValues_0.a_0);
     }
     else
     {
-        bool _S6 = (hitObjectIsMissNV((hit_0)));
+        bool _S7 = (hitObjectIsMissNV((hit_0)));
         uint r_3;
-        if(_S6)
+        if(_S7)
         {
             r_3 = 1U;
         }
@@ -78,29 +79,29 @@ uint calcValue_0(hitObjectNV hit_0)
     return r_0;
 }
 
-layout(std430, binding = 1) buffer _S7 {
+layout(std430, binding = 1) buffer _S8 {
     uint _data[];
 } outputBuffer_0;
 void main()
 {
-    uvec3 _S8 = ((gl_LaunchIDEXT));
-    ivec2 launchID_0 = ivec2(_S8.xy);
-    uvec3 _S9 = ((gl_LaunchSizeEXT));
+    uvec3 _S9 = ((gl_LaunchIDEXT));
+    ivec2 launchID_0 = ivec2(_S9.xy);
+    uvec3 _S10 = ((gl_LaunchSizeEXT));
     int idx_0 = launchID_0.x;
     RayDesc_0 ray_2;
     ray_2.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000);
     ray_2.TMin_0 = 0.00999999977648258209;
     ray_2.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000);
     ray_2.TMax_0 = 10000.00000000000000000000;
-    RayDesc_0 _S10 = ray_2;
+    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, _S10.Origin_0, _S10.TMin_0, _S10.Direction_0, _S10.TMax_0, (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);
-    RayDesc_0 _S11 = ray_2;
+    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, _S11.Origin_0, _S11.TMin_0, _S11.Direction_0, _S11.TMax_0, (0));
-    uint _S12 = calcValue_0(hitObj_1);
-    uint r_5 = r_4 + _S12;
+    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;
     return;
 }
diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang.1.expected
index 90223115b..f250c1c92 100644
--- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang.1.expected
+++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang.1.expected
@@ -79,36 +79,37 @@ void main()
     ivec2 launchID_0 = ivec2(_S3.xy);
     uvec3 _S4 = ((gl_LaunchSizeEXT));
     int idx_0 = launchID_0.x;
-    SomeValues_0 someValues_0 = { idx_0, float(idx_0) * 2.00000000000000000000 };
+    float _S5 = float(idx_0);
+    SomeValues_0 someValues_0 = { idx_0, _S5 * 2.00000000000000000000 };
     RayDesc_0 ray_0;
-    ray_0.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000);
+    ray_0.Origin_0 = vec3(_S5, 0.00000000000000000000, 0.00000000000000000000);
     ray_0.TMin_0 = 0.00999999977648258209;
     ray_0.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000);
     ray_0.TMax_0 = 10000.00000000000000000000;
-    RayDesc_0 _S5 = ray_0;
+    RayDesc_0 _S6 = ray_0;
     p_0 = someValues_0;
     hitObjectNV hitObj_0;
-    hitObjectTraceRayNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S5.Origin_0, _S5.TMin_0, _S5.Direction_0, _S5.TMax_0, (0));
+    hitObjectTraceRayNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S6.Origin_0, _S6.TMin_0, _S6.Direction_0, _S6.TMax_0, (0));
     uint r_1 = calcValue_0(hitObj_0);
     reorderThreadNV(hitObj_0);
     SomeValues_0 otherValues_0;
-    SomeValues_0 _S6 = { idx_0 * -1, float(idx_0) * 4.00000000000000000000 };
-    otherValues_0 = _S6;
+    SomeValues_0 _S7 = { idx_0 * -1, _S5 * 4.00000000000000000000 };
+    otherValues_0 = _S7;
     HitObject_Invoke_0(scene_0, hitObj_0, otherValues_0);
-    uint _S7 = calcValue_0(hitObj_0);
-    uint r_2 = r_1 + _S7;
+    uint _S8 = calcValue_0(hitObj_0);
+    uint r_2 = r_1 + _S8;
     reorderThreadNV(hitObj_0, uint(idx_0 & 3), 2U);
-    SomeValues_0 _S8 = { idx_0 * -2, float(idx_0) * 8.00000000000000000000 };
-    otherValues_0 = _S8;
+    SomeValues_0 _S9 = { idx_0 * -2, _S5 * 8.00000000000000000000 };
+    otherValues_0 = _S9;
     HitObject_Invoke_0(scene_0, hitObj_0, otherValues_0);
-    uint _S9 = calcValue_0(hitObj_0);
-    uint r_3 = r_2 + _S9;
+    uint _S10 = calcValue_0(hitObj_0);
+    uint r_3 = r_2 + _S10;
     reorderThreadNV(uint(idx_0 & 1), 1U);
-    SomeValues_0 _S10 = { idx_0 * -4, float(idx_0) * 16.00000000000000000000 };
-    otherValues_0 = _S10;
+    SomeValues_0 _S11 = { idx_0 * -4, _S5 * 16.00000000000000000000 };
+    otherValues_0 = _S11;
     HitObject_Invoke_0(scene_0, hitObj_0, otherValues_0);
-    uint _S11 = calcValue_0(hitObj_0);
-    uint r_4 = r_3 + _S11;
+    uint _S12 = calcValue_0(hitObj_0);
+    uint r_4 = r_3 + _S12;
     ((outputBuffer_0)._data[(uint(idx_0))]) = r_4;
     return;
 }
diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang.1.expected
index a86dc6aa7..f6f6f132d 100644
--- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang.1.expected
+++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang.1.expected
@@ -70,19 +70,20 @@ void main()
     int idx_0 = launchID_0.x;
     int _S5 = idx_0 / 4;
     float currentTime_0 = float(_S5);
-    SomeValues_0 someValues_0 = { idx_0, float(idx_0) * 2.00000000000000000000 };
+    float _S6 = float(idx_0);
+    SomeValues_0 someValues_0 = { idx_0, _S6 * 2.00000000000000000000 };
     RayDesc_0 ray_0;
-    ray_0.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000);
+    ray_0.Origin_0 = vec3(_S6, 0.00000000000000000000, 0.00000000000000000000);
     ray_0.TMin_0 = 0.00999999977648258209;
     ray_0.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000);
     ray_0.TMax_0 = 10000.00000000000000000000;
-    RayDesc_0 _S6 = ray_0;
+    RayDesc_0 _S7 = ray_0;
     p_0 = someValues_0;
     hitObjectNV hitObj_0;
-    hitObjectTraceRayMotionNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S6.Origin_0, _S6.TMin_0, _S6.Direction_0, _S6.TMax_0, currentTime_0, (0));
-    uint _S7 = uint(idx_0);
-    uint _S8 = calcValue_0(hitObj_0);
-    ((outputBuffer_0)._data[(_S7)]) = _S8;
+    hitObjectTraceRayMotionNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S7.Origin_0, _S7.TMin_0, _S7.Direction_0, _S7.TMax_0, currentTime_0, (0));
+    uint _S8 = uint(idx_0);
+    uint _S9 = calcValue_0(hitObj_0);
+    ((outputBuffer_0)._data[(_S8)]) = _S9;
     return;
 }
 
diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang.1.expected
index 38ddbf233..16099b5e2 100644
--- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang.1.expected
+++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang.1.expected
@@ -67,19 +67,20 @@ void main()
     ivec2 launchID_0 = ivec2(_S3.xy);
     uvec3 _S4 = ((gl_LaunchSizeEXT));
     int idx_0 = launchID_0.x;
-    SomeValues_0 someValues_0 = { idx_0, float(idx_0) * 2.00000000000000000000 };
+    float _S5 = float(idx_0);
+    SomeValues_0 someValues_0 = { idx_0, _S5 * 2.00000000000000000000 };
     RayDesc_0 ray_0;
-    ray_0.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000);
+    ray_0.Origin_0 = vec3(_S5, 0.00000000000000000000, 0.00000000000000000000);
     ray_0.TMin_0 = 0.00999999977648258209;
     ray_0.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000);
     ray_0.TMax_0 = 10000.00000000000000000000;
-    RayDesc_0 _S5 = ray_0;
+    RayDesc_0 _S6 = ray_0;
     p_0 = someValues_0;
     hitObjectNV hitObj_0;
-    hitObjectTraceRayNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S5.Origin_0, _S5.TMin_0, _S5.Direction_0, _S5.TMax_0, (0));
-    uint _S6 = uint(idx_0);
-    uint _S7 = calcValue_0(hitObj_0);
-    ((outputBuffer_0)._data[(_S6)]) = _S7;
+    hitObjectTraceRayNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S6.Origin_0, _S6.TMin_0, _S6.Direction_0, _S6.TMax_0, (0));
+    uint _S7 = uint(idx_0);
+    uint _S8 = calcValue_0(hitObj_0);
+    ((outputBuffer_0)._data[(_S7)]) = _S8;
     return;
 }