[DAP] Fix 'dap.fir' operation. (#432)

* [DAP] Reformat CMakeLists.txt. * [DAP] Debug 'dap.fir' operation. Co-authorized-by: ShiHaoGao <[email protected]>
buddy-compiler · Nov 27, 2024 · 94f276d · 94f276d
1 parent b8f1ec9
commit 94f276d
Show file tree

Hide file tree

Showing 4 changed files with 145 additions and 77 deletions.
diff --git a/examples/DAPDialect/FIRLowpass.cpp b/examples/DAPDialect/FIRLowpass.cpp
@@ -62,7 +62,7 @@ int main() {
   printLogLabel();
   std::cout << "Running FIR operation..." << std::endl;
   const auto loadStart = std::chrono::high_resolution_clock::now();
-  dap::fir(&inputContainer, &kernel, &outputMemRef);
+  dap::FIR(&inputContainer, &kernel, &outputMemRef);
   const auto loadEnd = std::chrono::high_resolution_clock::now();
   const std::chrono::duration<double, std::milli> loadTime =
       loadEnd - loadStart;

diff --git a/frontend/Interfaces/buddy/DAP/DSP/FIR.h b/frontend/Interfaces/buddy/DAP/DSP/FIR.h
@@ -31,8 +31,8 @@ namespace detail {
 extern "C" {
 // TODO: support both float and double.
 void _mlir_ciface_buddy_fir(MemRef<float, 1> *inputBuddyConv1D,
-                               MemRef<float, 1> *kernelBuddyConv1D,
-                               MemRef<float, 1> *outputBuddyConv1D);
+                            MemRef<float, 1> *kernelBuddyConv1D,
+                            MemRef<float, 1> *outputBuddyConv1D);
 }
 } // namespace detail
 
@@ -67,7 +67,7 @@ void firLowpass(MemRef<T, N> &input, WINDOW_TYPE type, size_t len, T cutoff,
 }
 
 template <typename T, size_t N>
-void fir(MemRef<float, N> *input, MemRef<T, N> *filter,
+void FIR(MemRef<float, N> *input, MemRef<T, N> *filter,
          MemRef<float, N> *output) {
   if (N != 1)
     assert(0 && "Only mono audio is supported for now.");

diff --git a/frontend/Interfaces/lib/CMakeLists.txt b/frontend/Interfaces/lib/CMakeLists.txt
@@ -10,6 +10,10 @@ elseif (HAVE_NEON)
     set(SPLITING_SIZE 16)
 endif ()
 
+#-------------------------------------------------------------------------------
+# Generate Buddy DIP Library: BuddyLibDIP
+#-------------------------------------------------------------------------------
+
 add_custom_command(OUTPUT DIP.o
         COMMAND ${CMAKE_BINARY_DIR}/bin/buddy-opt ${CMAKE_CURRENT_SOURCE_DIR}/DIP.mlir
         -lower-dip="DIP-strip-mining=${SPLITING_SIZE}"
@@ -37,87 +41,97 @@ SET_TARGET_PROPERTIES(BuddyLibDIP PROPERTIES
   ARCHIVE_OUTPUT_DIRECTORY ${LIBRARY_OUTPUT_DIRECTORY}
   )
 
-add_custom_command(OUTPUT DAP.o
-  COMMAND ${CMAKE_BINARY_DIR}/bin/buddy-opt ${CMAKE_CURRENT_SOURCE_DIR}/DAP.mlir 
-              -lower-dap="DAP-vector-splitting=${SPLITING_SIZE}"
-              --convert-linalg-to-affine-loops
-              -arith-expand
-              -lower-affine
-              -convert-scf-to-cf
-              -convert-math-to-llvm
-              -convert-vector-to-llvm
-              -finalize-memref-to-llvm
-              -llvm-request-c-wrappers
-              -convert-func-to-llvm
-              -reconcile-unrealized-casts | 
-          ${LLVM_TOOLS_BINARY_DIR}/mlir-translate --mlir-to-llvmir |
-          ${LLVM_TOOLS_BINARY_DIR}/llc 
-              -mtriple=${BUDDY_TARGET_TRIPLE}
-              -mattr=${BUDDY_OPT_ATTR}
-              --filetype=obj
-              -o ${CMAKE_CURRENT_BINARY_DIR}/DAP.o
+#-------------------------------------------------------------------------------
+# Generate Buddy DAP Library: BuddyLibDAP
+#-------------------------------------------------------------------------------
+
+add_custom_command(
+  OUTPUT DAP.o
+  COMMAND 
+    ${CMAKE_BINARY_DIR}/bin/buddy-opt ${CMAKE_CURRENT_SOURCE_DIR}/DAP.mlir 
+      -lower-dap="DAP-vector-splitting=${SPLITING_SIZE}"
+      --convert-linalg-to-affine-loops
+      -arith-expand
+      -lower-affine
+      -convert-scf-to-cf
+      -convert-math-to-llvm
+      -convert-vector-to-llvm
+      -finalize-memref-to-llvm
+      -llvm-request-c-wrappers
+      -convert-func-to-llvm
+      -reconcile-unrealized-casts | 
+    ${LLVM_TOOLS_BINARY_DIR}/mlir-translate --mlir-to-llvmir |
+    ${LLVM_TOOLS_BINARY_DIR}/llc 
+      -mtriple=${BUDDY_TARGET_TRIPLE}
+      -mattr=${BUDDY_OPT_ATTR}
+      --filetype=obj
+      -o ${CMAKE_CURRENT_BINARY_DIR}/DAP.o
   DEPENDS mlir-translate llc buddy-opt
-  )
+)
 
-add_custom_command(OUTPUT DAP-extend.o
-  COMMAND ${CMAKE_BINARY_DIR}/bin/buddy-opt ${CMAKE_CURRENT_SOURCE_DIR}/DAP-extend.mlir 
-              -extend-dap
-              -one-shot-bufferize
-              -convert-linalg-to-loops
-              -convert-scf-to-cf
-              -expand-strided-metadata
-              -lower-affine
-              -convert-vector-to-llvm 
-              -memref-expand 
-              -arith-expand
-              -convert-arith-to-llvm
-              -finalize-memref-to-llvm 
-              -convert-math-to-llvm
-              -llvm-request-c-wrappers
-              -convert-func-to-llvm
-              -reconcile-unrealized-casts | 
-          ${LLVM_TOOLS_BINARY_DIR}/mlir-translate -mlir-to-llvmir |
-          ${LLVM_TOOLS_BINARY_DIR}/llc 
-              -mtriple=${BUDDY_TARGET_TRIPLE}
-              -mattr=${BUDDY_OPT_ATTR}
-              -filetype=obj -relocation-model=pic
-              -o ${CMAKE_CURRENT_BINARY_DIR}/DAP-extend.o
+add_custom_command(
+  OUTPUT DAP-extend.o
+  COMMAND 
+    ${CMAKE_BINARY_DIR}/bin/buddy-opt ${CMAKE_CURRENT_SOURCE_DIR}/DAP-extend.mlir 
+      -extend-dap
+      -one-shot-bufferize
+      -convert-linalg-to-loops
+      -convert-scf-to-cf
+      -expand-strided-metadata
+      -lower-affine
+      -convert-vector-to-llvm 
+      -memref-expand 
+      -arith-expand
+      -convert-arith-to-llvm
+      -finalize-memref-to-llvm 
+      -convert-math-to-llvm
+      -llvm-request-c-wrappers
+      -convert-func-to-llvm
+      -reconcile-unrealized-casts | 
+    ${LLVM_TOOLS_BINARY_DIR}/mlir-translate -mlir-to-llvmir |
+    ${LLVM_TOOLS_BINARY_DIR}/llc 
+      -mtriple=${BUDDY_TARGET_TRIPLE}
+      -mattr=${BUDDY_OPT_ATTR}
+      -filetype=obj -relocation-model=pic
+      -o ${CMAKE_CURRENT_BINARY_DIR}/DAP-extend.o
   DEPENDS mlir-translate llc buddy-opt
-  )
+)
 
-add_custom_command(OUTPUT DAPVectorization.o
-  COMMAND cat ${CMAKE_CURRENT_SOURCE_DIR}/DAP.mlir |
-          sed 's/buddy_fir/buddy_fir_vectorization/' |
-          sed 's/buddy_iir/buddy_iir_vectorization/' |
-          sed 's/buddy_biquad/buddy_biquad_vectorization/' |
-          ${CMAKE_BINARY_DIR}/bin/buddy-opt
-              -vectorize-dap
-              -convert-linalg-to-affine-loops
-              -arith-expand
-              -lower-affine
-              -convert-scf-to-cf
-              -convert-math-to-llvm
-              -convert-vector-to-llvm
-              -finalize-memref-to-llvm
-              -llvm-request-c-wrappers
-              -convert-func-to-llvm
-              -reconcile-unrealized-casts | 
-          ${LLVM_TOOLS_BINARY_DIR}/mlir-translate -mlir-to-llvmir |
-          ${LLVM_TOOLS_BINARY_DIR}/llc 
-              -mtriple=${BUDDY_TARGET_TRIPLE}
-              -mattr=${BUDDY_OPT_ATTR}
-              -filetype=obj
-              -o ${CMAKE_CURRENT_BINARY_DIR}/DAPVectorization.o
+add_custom_command(
+  OUTPUT DAPVectorization.o
+  COMMAND 
+    cat ${CMAKE_CURRENT_SOURCE_DIR}/DAP.mlir |
+    sed -e 's/@buddy_fir/@buddy_fir_vectorization/'
+        -e 's/@buddy_iir/@buddy_iir_vectorization/'
+        -e 's/@buddy_biquad/@buddy_biquad_vectorization/' |
+    ${CMAKE_BINARY_DIR}/bin/buddy-opt
+      -vectorize-dap
+      -convert-linalg-to-affine-loops
+      -arith-expand
+      -lower-affine
+      -convert-scf-to-cf
+      -convert-math-to-llvm
+      -convert-vector-to-llvm
+      -finalize-memref-to-llvm
+      -llvm-request-c-wrappers
+      -convert-func-to-llvm
+      -reconcile-unrealized-casts | 
+    ${LLVM_TOOLS_BINARY_DIR}/mlir-translate -mlir-to-llvmir |
+    ${LLVM_TOOLS_BINARY_DIR}/llc 
+      -mtriple=${BUDDY_TARGET_TRIPLE}
+      -mattr=${BUDDY_OPT_ATTR}
+      -filetype=obj
+      -o ${CMAKE_CURRENT_BINARY_DIR}/DAPVectorization.o
   DEPENDS mlir-translate llc buddy-opt
-  )
+)
 
 add_library(BuddyLibDAP STATIC 
   DAP.o 
   DAP-extend.o 
   DAPVectorization.o
-  )
+)
 
 SET_TARGET_PROPERTIES(BuddyLibDAP PROPERTIES
   LINKER_LANGUAGE CXX
   ARCHIVE_OUTPUT_DIRECTORY ${LIBRARY_OUTPUT_DIRECTORY}
-  )
+)
diff --git a/midend/lib/Conversion/LowerDAP/LowerDAPPass.cpp b/midend/lib/Conversion/LowerDAP/LowerDAPPass.cpp
@@ -51,13 +51,67 @@ class DAPFirLowering : public OpRewritePattern<dap::FirOp> {
   LogicalResult matchAndRewrite(dap::FirOp op,
                                 PatternRewriter &rewriter) const override {
     auto loc = op->getLoc();
-    // auto ctx = op->getContext();
+    auto ctx = op->getContext();
     Value input = op->getOperand(0);
     Value kernel = op->getOperand(1);
     Value output = op->getOperand(2);
 
-    rewriter.create<linalg::Conv1DOp>(loc, ValueRange{input, kernel},
-                                      ValueRange{output});
+    Value c0 = rewriter.create<arith::ConstantIndexOp>(loc, 0);
+    Value c1 = rewriter.create<arith::ConstantIndexOp>(loc, 1);
+    FloatType f32 = FloatType::getF32(ctx);
+    Value f0 =
+        rewriter.create<arith::ConstantFloatOp>(loc, APFloat(float(0.0)), f32);
+
+    Value kernelSize = rewriter.create<memref::DimOp>(loc, kernel, c0);
+    Value dataLen = rewriter.create<memref::DimOp>(loc, output, c0);
+
+    // Populate the FIR pipeline by padding the `input` with [`kernelSize`-1]
+    // zeros at the beginning. Compute only the padding section of the input
+    // data.
+    Value fillInLen = rewriter.create<arith::SubIOp>(loc, kernelSize, c1);
+    rewriter.create<scf::ForOp>(
+        loc, c0, fillInLen, c1, ValueRange{std::nullopt},
+        [&](OpBuilder &b, Location loc, Value iv_n, ValueRange iargs) {
+          Value upperBound = b.create<arith::AddIOp>(loc, iv_n, c1);
+          Value outFinal =
+              b.create<scf::ForOp>(
+                   loc, c0, upperBound, c1, ValueRange{f0},
+                   [&](OpBuilder &b, Location loc, Value iv_k,
+                       ValueRange iargs) {
+                     Value i = b.create<arith::SubIOp>(loc, iv_n, iv_k);
+                     Value in = b.create<memref::LoadOp>(loc, input, i);
+                     Value k = b.create<memref::LoadOp>(loc, kernel, iv_k);
+                     Value mul = b.create<arith::MulFOp>(loc, in, k);
+                     Value outNext =
+                         b.create<arith::AddFOp>(loc, iargs[0], mul);
+                     b.create<scf::YieldOp>(loc, outNext);
+                   })
+                  .getResult(0);
+          b.create<memref::StoreOp>(loc, outFinal, output, ValueRange{iv_n});
+          b.create<scf::YieldOp>(loc, std::nullopt);
+        });
+
+    // Compute the input data following the padding section.
+    rewriter.create<scf::ForOp>(
+        loc, fillInLen, dataLen, c1, ValueRange{std::nullopt},
+        [&](OpBuilder &b, Location loc, Value iv_n, ValueRange iargs) {
+          Value outFinal =
+              b.create<scf::ForOp>(
+                   loc, c0, kernelSize, c1, ValueRange{f0},
+                   [&](OpBuilder &b, Location loc, Value iv_k,
+                       ValueRange iargs) {
+                     Value i = b.create<arith::SubIOp>(loc, iv_n, iv_k);
+                     Value in = b.create<memref::LoadOp>(loc, input, i);
+                     Value k = b.create<memref::LoadOp>(loc, kernel, iv_k);
+                     Value mul = b.create<arith::MulFOp>(loc, in, k);
+                     Value outNext =
+                         b.create<arith::AddFOp>(loc, iargs[0], mul);
+                     b.create<scf::YieldOp>(loc, outNext);
+                   })
+                  .getResult(0);
+          b.create<memref::StoreOp>(loc, outFinal, output, ValueRange{iv_n});
+          b.create<scf::YieldOp>(loc, std::nullopt);
+        });
 
     rewriter.eraseOp(op);
     return success();