AD: Fix out-of-scope indexing rules for insts in loop header blocks during the primal-inst availability pass (#2918)

* add test case

* Fix out-of-scope indexing rules for loop header blocks

---------

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

4 files changed, 96 insertions, 7 deletions

diff --git a/source/slang/slang-ir-autodiff-primal-hoist.cpp b/source/slang/slang-ir-autodiff-primal-hoist.cpp
index 0016f25e3..6947fd7c5 100644
--- a/source/slang/slang-ir-autodiff-primal-hoist.cpp
+++ b/source/slang/slang-ir-autodiff-primal-hoist.cpp
@@ -961,6 +961,7 @@ IRInst* emitIndexedStoreAddressForVar(
 IRInst* emitIndexedLoadAddressForVar(
     IRBuilder* builder,
     IRVar* localVar,
+    IRBlock* defBlock,
     const List<IndexTrackingInfo>& defBlockIndices,
     const List<IndexTrackingInfo>& useBlockIndices)
 {
@@ -986,12 +987,15 @@ IRInst* emitIndexedLoadAddressForVar(
         {
             // If the use-block is outside this region, use the
             // last available value (by indexing with primal counter minus 1)
+            // An exception is if the stored inst is in a loop header block where 
+            // we use counter directly (since that block runs N+1 times)
             // 
             auto primalCounterCurrValue = index.primalCountParam;
-            auto primalCounterLastValue = builder->emitSub(
-                primalCounterCurrValue->getDataType(),
-                primalCounterCurrValue,
-                builder->getIntValue(builder->getIntType(), 1));
+            auto primalCounterLastValue = (index.loopHeaderBlock == defBlock) ? primalCounterCurrValue : 
+                builder->emitSub(
+                    primalCounterCurrValue->getDataType(),
+                    primalCounterCurrValue,
+                    builder->getIntValue(builder->getIntType(), 1));
 
             loadAddr = builder->emitElementAddress(
                 builder->getPtrType(currType),
@@ -1021,10 +1025,11 @@ IRVar* storeIndexedValue(
 IRInst* loadIndexedValue(
     IRBuilder* builder,
     IRVar* localVar,
+    IRBlock* defBlock,
     const List<IndexTrackingInfo>& defBlockIndices,
     const List<IndexTrackingInfo>& useBlockIndices)
 {
-    IRInst* addr = emitIndexedLoadAddressForVar(builder, localVar, defBlockIndices, useBlockIndices);
+    IRInst* addr = emitIndexedLoadAddressForVar(builder, localVar, defBlock, defBlockIndices, useBlockIndices);
 
     return builder->emitLoad(addr);
 }
@@ -1292,7 +1297,12 @@ RefPtr<HoistedPrimalsInfo> ensurePrimalAvailability(
 
                     List<IndexTrackingInfo>& useBlockIndices = indexedBlockInfo[getBlock(use->getUser())];
 
-                    IRInst* loadAddr = emitIndexedLoadAddressForVar(&builder, localVar, defBlockIndices, useBlockIndices);
+                    IRInst* loadAddr = emitIndexedLoadAddressForVar(
+                        &builder,
+                        localVar,
+                        defBlock,
+                        defBlockIndices,
+                        useBlockIndices);
                     builder.replaceOperand(use, loadAddr);
                 }
 
@@ -1323,7 +1333,9 @@ RefPtr<HoistedPrimalsInfo> ensurePrimalAvailability(
                 {
                     List<IndexTrackingInfo> useBlockIndices = indexedBlockInfo[getBlock(use->getUser())];
                     setInsertBeforeOrdinaryInst(&builder, getInstInBlock(use->getUser()));
-                    builder.replaceOperand(use, loadIndexedValue(&builder, localVar, defBlockIndices, useBlockIndices));
+                    builder.replaceOperand(
+                        use,
+                        loadIndexedValue(&builder, localVar, defBlock, defBlockIndices, useBlockIndices));
                 }
 
                 if (!isRecomputeInst)
@@ -1512,6 +1524,8 @@ void buildIndexedBlocks(
         IndexTrackingInfo indexInfo = {};
         lowerIndexedRegion(primalLoop, loop, indexInfo.primalCountParam, indexInfo.diffCountParam);
 
+        indexInfo.loopHeaderBlock = getLoopConditionBlock(primalLoop);
+
         SLANG_RELEASE_ASSERT(indexInfo.primalCountParam);
         SLANG_RELEASE_ASSERT(indexInfo.diffCountParam);
 
diff --git a/source/slang/slang-ir-autodiff-primal-hoist.h b/source/slang/slang-ir-autodiff-primal-hoist.h
index c9377d56b..59c70e862 100644
--- a/source/slang/slang-ir-autodiff-primal-hoist.h
+++ b/source/slang/slang-ir-autodiff-primal-hoist.h
@@ -211,6 +211,12 @@ namespace Slang
         IRInst* primalCountParam = nullptr;
         IRInst* diffCountParam = nullptr;
 
+        // Reference to the header block. Note that the header block 
+        // typically contains the loop condition and is executed N+1
+        // times if the loop body is executed N times.
+        // 
+        IRBlock* loopHeaderBlock = nullptr;
+
         enum CountStatus
         {
             Unresolved,
diff --git a/tests/autodiff/high-order-backward-diff-4.slang b/tests/autodiff/high-order-backward-diff-4.slang
new file mode 100644
index 000000000..9ee9aa4c4
--- /dev/null
+++ b/tests/autodiff/high-order-backward-diff-4.slang
@@ -0,0 +1,64 @@
+//TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute -output-using-type -shaderobj
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj -output-using-type
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -output-using-type
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+struct A : IDifferentiable
+{
+    float x;
+    int nx;
+}
+
+[BackwardDifferentiable]
+float f(A x)
+{
+    A rs;
+    rs.x = 1.0;
+    for (int i = 0; i < 2; i++)
+        rs.x = rs.x * x.x;
+    return rs.x;
+}
+
+[BackwardDifferentiable]
+float outerF(A x)
+{
+    A nx;
+    nx.x = 1.0;
+    int i = 0;
+    [MaxIters(3)]
+    do
+    {
+        nx.x = nx.x * x.x;
+        i++;
+    }
+    while (i < 2);
+    nx.nx = 2;//x.nx;
+    return f(nx);
+}
+
+[BackwardDifferentiable]
+float df(A x)
+{
+    A.Differential ad;
+    ad.x = 0.0;
+    var p = diffPair(x, ad);
+    __bwd_diff(outerF)(p, 1.0);
+    return p.d.x;
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    // Given f(x) = x^4,
+    // f''(x) = 12 * x^2
+    // Expect f''(4) = 192
+    A a;
+    a.x = 4.0;
+    a.nx = 54;
+    A.Differential ad;
+    ad.x = 1.0;
+    var p = diffPair(a, ad);
+    outputBuffer[0] = __fwd_diff(df)(p).d;
+}
diff --git a/tests/autodiff/high-order-backward-diff-4.slang.expected.txt b/tests/autodiff/high-order-backward-diff-4.slang.expected.txt
new file mode 100644
index 000000000..0f08247f0
--- /dev/null
+++ b/tests/autodiff/high-order-backward-diff-4.slang.expected.txt
@@ -0,0 +1,5 @@
+type: float
+192.000000
+0.000000
+0.000000
+0.000000