Skip to content

[mlir][tosa] Fix tosa.Resize-to-linalg lowering #88514

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Apr 17, 2024
Merged

[mlir][tosa] Fix tosa.Resize-to-linalg lowering #88514

merged 1 commit into from
Apr 17, 2024

Conversation

fabrizio-indirli
Copy link
Contributor

Simplify the lowering of tosa.resize to linalg, to match more closely the implementation from the TOSA specification: some internal arithmetic operations are now performed on the integer types (rather than in floating point) to reduce the loss of precision and the num of generated operations.

@fabrizio-indirli
[mlir][tosa] Fix tosa.Resize-to-linalg lowering
e124355 
Simplify the lowering of tosa.resize to linalg, to match
more closely the implementation from the TOSA specification:
some internal arithmetic operations are now performed on
the integer types (rather than in floating point) to reduce
the loss of precision and the num of generated operations.

Signed-off-by: Fabrizio Indirli <[email protected]>
@llvmbot
Copy link
Member

llvmbot commented Apr 12, 2024

@llvm/pr-subscribers-mlir

Author: None (fabrizio-indirli)

Changes

Simplify the lowering of tosa.resize to linalg, to match more closely the implementation from the TOSA specification: some internal arithmetic operations are now performed on the integer types (rather than in floating point) to reduce the loss of precision and the num of generated operations.

Full diff: https://github.com/llvm/llvm-project/pull/88514.diff

2 Files Affected:

  • (modified) mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp (+10-11)
  • (modified) mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir (+50-62)
diff --git a/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp b/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
index 7c477f2e1412be..1a743282adc487 100644
--- a/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
+++ b/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
@@ -1578,17 +1578,16 @@ class GenericResizeConverter : public OpRewritePattern<tosa::ResizeOp> {
         }
         // x = x * scale_d + offset;
         // ix = floor(x / scale_n)
-        // dx = x / scale_n - ix
-        Value val = b.create<arith::UIToFPOp>(floatTy, in);
-        scaleN = b.create<arith::UIToFPOp>(floatTy, scaleN);
-        scaleD = b.create<arith::UIToFPOp>(floatTy, scaleD);
-        offset = b.create<arith::SIToFPOp>(floatTy, offset);
-        val = b.create<arith::MulFOp>(val, scaleD);
-        val = b.create<arith::AddFOp>(val, offset);
-        val = b.create<arith::DivFOp>(val, scaleN);
-        index = b.create<math::FloorOp>(val);
-        delta = b.create<arith::SubFOp>(val, index);
-        index = b.create<arith::FPToSIOp>(b.getI32Type(), index);
+        Value val = b.create<arith::MulIOp>(in, scaleD);
+        val = b.create<arith::AddIOp>(val, offset);
+        index = b.create<arith::FloorDivSIOp>(val, scaleN);
+
+        // rx = x % scale_n
+        // dx = rx / scale_n
+        Value r = b.create<arith::RemSIOp>(val, scaleN);
+        Value rFp = b.create<arith::SIToFPOp>(floatTy, r);
+        Value scaleNfp = b.create<arith::UIToFPOp>(floatTy, scaleN);
+        delta = b.create<arith::DivFOp>(rFp, scaleNfp);
       };
 
       // Compute the ix and dx values for the X and Y dimensions - int case.
diff --git a/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir b/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir
index 468e92e2a2661f..d42d0a46692d47 100644
--- a/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir
+++ b/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir
@@ -304,42 +304,36 @@ func.func @resize_nearest_fp32(%input: tensor<1x50x48x1xf32>) -> () {
   // CHECK-DAG: %[[XMAX:.*]] = arith.constant 47
   // CHECK: %[[Y:.+]] = arith.index_cast %[[IDX1]]
   // CHECK: %[[X:.+]] = arith.index_cast %[[IDX2]]
-  // CHECK-DAG: %[[ISCALE_Y_N:.*]] = arith.constant 64
-  // CHECK-DAG: %[[ISCALE_Y_D:.*]] = arith.constant 2
-  // CHECK-DAG: %[[ISCALE_X_N:.*]] = arith.constant 64
-  // CHECK-DAG: %[[ISCALE_X_D:.*]] = arith.constant 2
-  // CHECK-DAG: %[[IOFFSET_Y:.*]] = arith.constant -31
-  // CHECK-DAG: %[[IOFFSET_X:.*]] = arith.constant -31
-  // CHECK-DAG: %[[IBORDER_Y:.*]] = arith.constant 31
-  // CHECK-DAG: %[[IBORDER_X:.*]] = arith.constant 31
-
-  // CHECK: %[[Y0:.+]] = arith.uitofp %[[Y]]
-  // CHECK: %[[SCALE_Y_N:.*]] = arith.uitofp %[[ISCALE_Y_N]]
-  // CHECK: %[[SCALE_Y_D:.*]] = arith.uitofp %[[ISCALE_Y_D]]
-  // CHECK: %[[OFFSET_Y:.*]] = arith.sitofp %[[IOFFSET_Y]]
-  // CHECK: %[[VAL_29:.*]] = arith.mulf %[[Y0]], %[[SCALE_Y_D]]
-  // CHECK: %[[VAL_31:.*]] = arith.addf %[[VAL_29]], %[[OFFSET_Y]]
-  // CHECK: %[[VAL_33:.*]] = arith.divf %[[VAL_31]], %[[SCALE_Y_N]]
-  // CHECK: %[[VAL_35:.*]] = math.floor %[[VAL_33]]
-  // CHECK: %[[D_Y:.*]] = arith.subf %[[VAL_33]], %[[VAL_35]]
-  // CHECK: %[[VAL_39:.*]] = arith.fptosi %[[VAL_35]]
-
-  // CHECK: %[[X0:.+]] = arith.uitofp %[[X]]
-  // CHECK: %[[SCALE_X_N:.*]] = arith.uitofp %[[ISCALE_X_N]]
-  // CHECK: %[[SCALE_X_D:.*]] = arith.uitofp %[[ISCALE_X_D]]
-  // CHECK: %[[OFFSET_X:.*]] = arith.sitofp %[[IOFFSET_X]]
-  // CHECK: %[[VAL_30:.*]] = arith.mulf %[[X0]], %[[SCALE_X_D]]
-  // CHECK: %[[VAL_32:.*]] = arith.addf %[[VAL_30]], %[[OFFSET_X]]
-  // CHECK: %[[VAL_34:.*]] = arith.divf %[[VAL_32]], %[[SCALE_X_N]]
-  // CHECK: %[[VAL_36:.*]] = math.floor %[[VAL_34]]
-  // CHECK: %[[D_X:.*]] = arith.subf %[[VAL_34]], %[[VAL_36]]
-  // CHECK: %[[VAL_40:.*]] = arith.fptosi %[[VAL_36]]
+  // CHECK-DAG: %[[SCALE_Y_N:.*]] = arith.constant 64
+  // CHECK-DAG: %[[SCALE_Y_D:.*]] = arith.constant 2
+  // CHECK-DAG: %[[SCALE_X_N:.*]] = arith.constant 64
+  // CHECK-DAG: %[[SCALE_X_D:.*]] = arith.constant 2
+  // CHECK-DAG: %[[OFFSET_Y:.*]] = arith.constant -31
+  // CHECK-DAG: %[[OFFSET_X:.*]] = arith.constant -31
+  // CHECK-DAG: %[[BORDER_Y:.*]] = arith.constant 31
+  // CHECK-DAG: %[[BORDER_X:.*]] = arith.constant 31
+
+  // CHECK: %[[VAL_29:.*]] = arith.muli %[[Y]], %[[SCALE_Y_D]]
+  // CHECK: %[[Y_TEMP:.*]] = arith.addi %[[VAL_29]], %[[OFFSET_Y]]
+  // CHECK: %[[IY_TEMP:.*]] = arith.floordivsi %[[Y_TEMP]], %[[SCALE_Y_N]]
+  // CHECK: %[[RY:.*]] = arith.remsi %[[Y_TEMP]], %[[SCALE_Y_N]]
+  // CHECK: %[[RY_FP:.*]] = arith.sitofp %[[RY]]
+  // CHECK: %[[SCALE_Y_N_FP:.*]] = arith.uitofp %[[SCALE_Y_N]]
+  // CHECK: %[[D_Y:.*]] = arith.divf %[[RY_FP]], %[[SCALE_Y_N_FP]]
+
+  // CHECK: %[[VAL_30:.*]] = arith.muli %[[X]], %[[SCALE_X_D]]
+  // CHECK: %[[X_TEMP:.*]] = arith.addi %[[VAL_30]], %[[OFFSET_X]]
+  // CHECK: %[[IX_TEMP:.*]] = arith.floordivsi %[[X_TEMP]], %[[SCALE_X_N]]
+  // CHECK: %[[RX:.*]] = arith.remsi %[[X_TEMP]], %[[SCALE_X_N]]
+  // CHECK: %[[RX_FP:.*]] = arith.sitofp %[[RX]]
+  // CHECK: %[[SCALE_X_N_FP:.*]] = arith.uitofp %[[SCALE_X_N]]
+  // CHECK: %[[D_X:.*]] = arith.divf %[[RX_FP]], %[[SCALE_X_N_FP]]
 
   // CHECK-DAG: %[[ONE:.*]] = arith.constant 1
   // CHECK-DAG: %[[HALF:.*]] = arith.constant 5.000000e-01
   // CHECK: %[[PRED_Y:.*]] = arith.cmpf oge, %[[D_Y]], %[[HALF]]
   // CHECK: %[[ROUND_Y:.*]] = arith.select %[[PRED_Y]], %[[ONE]], %[[ZERO]]
-  // CHECK: %[[VAL_48:.*]] = arith.addi %[[VAL_39]], %[[ROUND_Y]]
+  // CHECK: %[[VAL_48:.*]] = arith.addi %[[IY_TEMP]], %[[ROUND_Y]]
   // CHECK: %[[LOWER:.*]] = arith.maxsi %[[ZERO]], %[[VAL_48]]
   // CHECK: %[[CLAMPED:.*]] = arith.minsi %[[YMAX]], %[[LOWER]]
   // CHECK: %[[IDY:.*]] = arith.index_cast %[[CLAMPED]]
@@ -347,7 +341,7 @@ func.func @resize_nearest_fp32(%input: tensor<1x50x48x1xf32>) -> () {
   // CHECK-DAG: %[[HALF:.*]] = arith.constant 5.000000e-01
   // CHECK: %[[PRED_X:.*]] = arith.cmpf oge, %[[D_X]], %[[HALF]]
   // CHECK: %[[ROUND_X:.*]] = arith.select %[[PRED_X]], %[[ONE]], %[[ZERO]]
-  // CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_40]], %[[ROUND_X]]
+  // CHECK: %[[VAL_49:.*]] = arith.addi %[[IX_TEMP]], %[[ROUND_X]]
   // CHECK: %[[LOWER:.*]] = arith.maxsi %[[ZERO]], %[[VAL_49]]
   // CHECK: %[[CLAMPED:.*]] = arith.minsi %[[XMAX]], %[[LOWER]]
   // CHECK: %[[IDX:.*]] = arith.index_cast %[[CLAMPED]]
@@ -374,36 +368,30 @@ func.func @resize_bilinear_fp(%input: tensor<1x23x24x1xf32>) -> () {
   // CHECK-DAG: %[[X_MAX:.*]] = arith.constant 23
   // CHECK: %[[Y:.+]] = arith.index_cast %[[IDX_1]]
   // CHECK: %[[X:.+]] = arith.index_cast %[[IDX_2]]
-  // CHECK-DAG: %[[ISCALE_Y_N:.*]] = arith.constant 4
-  // CHECK-DAG: %[[ISCALE_Y_D:.*]] = arith.constant 1
-  // CHECK-DAG: %[[ISCALE_X_N:.*]] = arith.constant 4
-  // CHECK-DAG: %[[ISCALE_X_D:.*]] = arith.constant 1
-  // CHECK-DAG: %[[IOFFSET_Y:.*]] = arith.constant 0
-  // CHECK-DAG: %[[IOFFSET_X:.*]] = arith.constant 0
-  // CHECK-DAG: %[[IBORDER_Y:.*]] = arith.constant 0
-  // CHECK-DAG: %[[IBORDER_X:.*]] = arith.constant 0
-
-  // CHECK: %[[Y0:.+]] = arith.uitofp %[[Y]]
-  // CHECK: %[[SCALE_Y_N:.*]] = arith.uitofp %[[ISCALE_Y_N]]
-  // CHECK: %[[SCALE_Y_D:.*]] = arith.uitofp %[[ISCALE_Y_D]]
-  // CHECK: %[[OFFSET_Y:.*]] = arith.sitofp %[[IOFFSET_Y]]
-  // CHECK: %[[VAL_29:.*]] = arith.mulf %[[Y0]], %[[SCALE_Y_D]]
-  // CHECK: %[[VAL_31:.*]] = arith.addf %[[VAL_29]], %[[OFFSET_Y]]
-  // CHECK: %[[VAL_33:.*]] = arith.divf %[[VAL_31]], %[[SCALE_Y_N]]
-  // CHECK: %[[VAL_35:.*]] = math.floor %[[VAL_33]]
-  // CHECK: %[[D_Y:.*]] = arith.subf %[[VAL_33]], %[[VAL_35]]
-  // CHECK: %[[I_Y:.*]] = arith.fptosi %[[VAL_35]]
-
-  // CHECK: %[[X0:.+]] = arith.uitofp %[[X]]
-  // CHECK: %[[SCALE_X_N:.*]] = arith.uitofp %[[ISCALE_X_N]]
-  // CHECK: %[[SCALE_X_D:.*]] = arith.uitofp %[[ISCALE_X_D]]
-  // CHECK: %[[OFFSET_X:.*]] = arith.sitofp %[[IOFFSET_X]]
-  // CHECK: %[[VAL_30:.*]] = arith.mulf %[[X0]], %[[SCALE_X_D]]
-  // CHECK: %[[VAL_32:.*]] = arith.addf %[[VAL_30]], %[[OFFSET_X]]
-  // CHECK: %[[VAL_34:.*]] = arith.divf %[[VAL_32]], %[[SCALE_X_N]]
-  // CHECK: %[[VAL_36:.*]] = math.floor %[[VAL_34]]
-  // CHECK: %[[D_X:.*]] = arith.subf %[[VAL_34]], %[[VAL_36]]
-  // CHECK: %[[I_X:.*]] = arith.fptosi %[[VAL_36]]
+  // CHECK-DAG: %[[SCALE_Y_N:.*]] = arith.constant 4
+  // CHECK-DAG: %[[SCALE_Y_D:.*]] = arith.constant 1
+  // CHECK-DAG: %[[SCALE_X_N:.*]] = arith.constant 4
+  // CHECK-DAG: %[[SCALE_X_D:.*]] = arith.constant 1
+  // CHECK-DAG: %[[OFFSET_Y:.*]] = arith.constant 0
+  // CHECK-DAG: %[[OFFSET_X:.*]] = arith.constant 0
+  // CHECK-DAG: %[[BORDER_Y:.*]] = arith.constant 0
+  // CHECK-DAG: %[[BORDER_X:.*]] = arith.constant 0
+
+  // CHECK: %[[VAL_29:.*]] = arith.muli %[[Y]], %[[SCALE_Y_D]]
+  // CHECK: %[[Y_TEMP:.*]] = arith.addi %[[VAL_29]], %[[OFFSET_Y]]
+  // CHECK: %[[I_Y:.*]] = arith.floordivsi %[[Y_TEMP]], %[[SCALE_Y_N]]
+  // CHECK: %[[RY:.*]] = arith.remsi %[[Y_TEMP]], %[[SCALE_Y_N]]
+  // CHECK: %[[RY_FP:.*]] = arith.sitofp %[[RY]]
+  // CHECK: %[[SCALE_Y_N_FP:.*]] = arith.uitofp %[[SCALE_Y_N]]
+  // CHECK: %[[D_Y:.*]] = arith.divf %[[RY_FP]], %[[SCALE_Y_N_FP]]
+
+  // CHECK: %[[VAL_30:.*]] = arith.muli %[[X]], %[[SCALE_X_D]]
+  // CHECK: %[[X_TEMP:.*]] = arith.addi %[[VAL_30]], %[[OFFSET_X]]
+  // CHECK: %[[I_X:.*]] = arith.floordivsi %[[X_TEMP]], %[[SCALE_X_N]]
+  // CHECK: %[[RX:.*]] = arith.remsi %[[X_TEMP]], %[[SCALE_X_N]]
+  // CHECK: %[[RX_FP:.*]] = arith.sitofp %[[RX]]
+  // CHECK: %[[SCALE_X_N_FP:.*]] = arith.uitofp %[[SCALE_X_N]]
+  // CHECK: %[[D_X:.*]] = arith.divf %[[RX_FP]], %[[SCALE_X_N_FP]]
 
   // Compute the left, right, and top indices for the bilinear interpolation.

@llvmbot
Copy link
Member

llvmbot commented Apr 12, 2024

@llvm/pr-subscribers-mlir-linalg

Author: None (fabrizio-indirli)

Changes

Simplify the lowering of tosa.resize to linalg, to match more closely the implementation from the TOSA specification: some internal arithmetic operations are now performed on the integer types (rather than in floating point) to reduce the loss of precision and the num of generated operations.

Full diff: https://github.com/llvm/llvm-project/pull/88514.diff

2 Files Affected:

  • (modified) mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp (+10-11)
  • (modified) mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir (+50-62)
diff --git a/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp b/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
index 7c477f2e1412be..1a743282adc487 100644
--- a/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
+++ b/mlir/lib/Conversion/TosaToLinalg/TosaToLinalg.cpp
@@ -1578,17 +1578,16 @@ class GenericResizeConverter : public OpRewritePattern<tosa::ResizeOp> {
         }
         // x = x * scale_d + offset;
         // ix = floor(x / scale_n)
-        // dx = x / scale_n - ix
-        Value val = b.create<arith::UIToFPOp>(floatTy, in);
-        scaleN = b.create<arith::UIToFPOp>(floatTy, scaleN);
-        scaleD = b.create<arith::UIToFPOp>(floatTy, scaleD);
-        offset = b.create<arith::SIToFPOp>(floatTy, offset);
-        val = b.create<arith::MulFOp>(val, scaleD);
-        val = b.create<arith::AddFOp>(val, offset);
-        val = b.create<arith::DivFOp>(val, scaleN);
-        index = b.create<math::FloorOp>(val);
-        delta = b.create<arith::SubFOp>(val, index);
-        index = b.create<arith::FPToSIOp>(b.getI32Type(), index);
+        Value val = b.create<arith::MulIOp>(in, scaleD);
+        val = b.create<arith::AddIOp>(val, offset);
+        index = b.create<arith::FloorDivSIOp>(val, scaleN);
+
+        // rx = x % scale_n
+        // dx = rx / scale_n
+        Value r = b.create<arith::RemSIOp>(val, scaleN);
+        Value rFp = b.create<arith::SIToFPOp>(floatTy, r);
+        Value scaleNfp = b.create<arith::UIToFPOp>(floatTy, scaleN);
+        delta = b.create<arith::DivFOp>(rFp, scaleNfp);
       };
 
       // Compute the ix and dx values for the X and Y dimensions - int case.
diff --git a/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir b/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir
index 468e92e2a2661f..d42d0a46692d47 100644
--- a/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir
+++ b/mlir/test/Conversion/TosaToLinalg/tosa-to-linalg-resize.mlir
@@ -304,42 +304,36 @@ func.func @resize_nearest_fp32(%input: tensor<1x50x48x1xf32>) -> () {
   // CHECK-DAG: %[[XMAX:.*]] = arith.constant 47
   // CHECK: %[[Y:.+]] = arith.index_cast %[[IDX1]]
   // CHECK: %[[X:.+]] = arith.index_cast %[[IDX2]]
-  // CHECK-DAG: %[[ISCALE_Y_N:.*]] = arith.constant 64
-  // CHECK-DAG: %[[ISCALE_Y_D:.*]] = arith.constant 2
-  // CHECK-DAG: %[[ISCALE_X_N:.*]] = arith.constant 64
-  // CHECK-DAG: %[[ISCALE_X_D:.*]] = arith.constant 2
-  // CHECK-DAG: %[[IOFFSET_Y:.*]] = arith.constant -31
-  // CHECK-DAG: %[[IOFFSET_X:.*]] = arith.constant -31
-  // CHECK-DAG: %[[IBORDER_Y:.*]] = arith.constant 31
-  // CHECK-DAG: %[[IBORDER_X:.*]] = arith.constant 31
-
-  // CHECK: %[[Y0:.+]] = arith.uitofp %[[Y]]
-  // CHECK: %[[SCALE_Y_N:.*]] = arith.uitofp %[[ISCALE_Y_N]]
-  // CHECK: %[[SCALE_Y_D:.*]] = arith.uitofp %[[ISCALE_Y_D]]
-  // CHECK: %[[OFFSET_Y:.*]] = arith.sitofp %[[IOFFSET_Y]]
-  // CHECK: %[[VAL_29:.*]] = arith.mulf %[[Y0]], %[[SCALE_Y_D]]
-  // CHECK: %[[VAL_31:.*]] = arith.addf %[[VAL_29]], %[[OFFSET_Y]]
-  // CHECK: %[[VAL_33:.*]] = arith.divf %[[VAL_31]], %[[SCALE_Y_N]]
-  // CHECK: %[[VAL_35:.*]] = math.floor %[[VAL_33]]
-  // CHECK: %[[D_Y:.*]] = arith.subf %[[VAL_33]], %[[VAL_35]]
-  // CHECK: %[[VAL_39:.*]] = arith.fptosi %[[VAL_35]]
-
-  // CHECK: %[[X0:.+]] = arith.uitofp %[[X]]
-  // CHECK: %[[SCALE_X_N:.*]] = arith.uitofp %[[ISCALE_X_N]]
-  // CHECK: %[[SCALE_X_D:.*]] = arith.uitofp %[[ISCALE_X_D]]
-  // CHECK: %[[OFFSET_X:.*]] = arith.sitofp %[[IOFFSET_X]]
-  // CHECK: %[[VAL_30:.*]] = arith.mulf %[[X0]], %[[SCALE_X_D]]
-  // CHECK: %[[VAL_32:.*]] = arith.addf %[[VAL_30]], %[[OFFSET_X]]
-  // CHECK: %[[VAL_34:.*]] = arith.divf %[[VAL_32]], %[[SCALE_X_N]]
-  // CHECK: %[[VAL_36:.*]] = math.floor %[[VAL_34]]
-  // CHECK: %[[D_X:.*]] = arith.subf %[[VAL_34]], %[[VAL_36]]
-  // CHECK: %[[VAL_40:.*]] = arith.fptosi %[[VAL_36]]
+  // CHECK-DAG: %[[SCALE_Y_N:.*]] = arith.constant 64
+  // CHECK-DAG: %[[SCALE_Y_D:.*]] = arith.constant 2
+  // CHECK-DAG: %[[SCALE_X_N:.*]] = arith.constant 64
+  // CHECK-DAG: %[[SCALE_X_D:.*]] = arith.constant 2
+  // CHECK-DAG: %[[OFFSET_Y:.*]] = arith.constant -31
+  // CHECK-DAG: %[[OFFSET_X:.*]] = arith.constant -31
+  // CHECK-DAG: %[[BORDER_Y:.*]] = arith.constant 31
+  // CHECK-DAG: %[[BORDER_X:.*]] = arith.constant 31
+
+  // CHECK: %[[VAL_29:.*]] = arith.muli %[[Y]], %[[SCALE_Y_D]]
+  // CHECK: %[[Y_TEMP:.*]] = arith.addi %[[VAL_29]], %[[OFFSET_Y]]
+  // CHECK: %[[IY_TEMP:.*]] = arith.floordivsi %[[Y_TEMP]], %[[SCALE_Y_N]]
+  // CHECK: %[[RY:.*]] = arith.remsi %[[Y_TEMP]], %[[SCALE_Y_N]]
+  // CHECK: %[[RY_FP:.*]] = arith.sitofp %[[RY]]
+  // CHECK: %[[SCALE_Y_N_FP:.*]] = arith.uitofp %[[SCALE_Y_N]]
+  // CHECK: %[[D_Y:.*]] = arith.divf %[[RY_FP]], %[[SCALE_Y_N_FP]]
+
+  // CHECK: %[[VAL_30:.*]] = arith.muli %[[X]], %[[SCALE_X_D]]
+  // CHECK: %[[X_TEMP:.*]] = arith.addi %[[VAL_30]], %[[OFFSET_X]]
+  // CHECK: %[[IX_TEMP:.*]] = arith.floordivsi %[[X_TEMP]], %[[SCALE_X_N]]
+  // CHECK: %[[RX:.*]] = arith.remsi %[[X_TEMP]], %[[SCALE_X_N]]
+  // CHECK: %[[RX_FP:.*]] = arith.sitofp %[[RX]]
+  // CHECK: %[[SCALE_X_N_FP:.*]] = arith.uitofp %[[SCALE_X_N]]
+  // CHECK: %[[D_X:.*]] = arith.divf %[[RX_FP]], %[[SCALE_X_N_FP]]
 
   // CHECK-DAG: %[[ONE:.*]] = arith.constant 1
   // CHECK-DAG: %[[HALF:.*]] = arith.constant 5.000000e-01
   // CHECK: %[[PRED_Y:.*]] = arith.cmpf oge, %[[D_Y]], %[[HALF]]
   // CHECK: %[[ROUND_Y:.*]] = arith.select %[[PRED_Y]], %[[ONE]], %[[ZERO]]
-  // CHECK: %[[VAL_48:.*]] = arith.addi %[[VAL_39]], %[[ROUND_Y]]
+  // CHECK: %[[VAL_48:.*]] = arith.addi %[[IY_TEMP]], %[[ROUND_Y]]
   // CHECK: %[[LOWER:.*]] = arith.maxsi %[[ZERO]], %[[VAL_48]]
   // CHECK: %[[CLAMPED:.*]] = arith.minsi %[[YMAX]], %[[LOWER]]
   // CHECK: %[[IDY:.*]] = arith.index_cast %[[CLAMPED]]
@@ -347,7 +341,7 @@ func.func @resize_nearest_fp32(%input: tensor<1x50x48x1xf32>) -> () {
   // CHECK-DAG: %[[HALF:.*]] = arith.constant 5.000000e-01
   // CHECK: %[[PRED_X:.*]] = arith.cmpf oge, %[[D_X]], %[[HALF]]
   // CHECK: %[[ROUND_X:.*]] = arith.select %[[PRED_X]], %[[ONE]], %[[ZERO]]
-  // CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_40]], %[[ROUND_X]]
+  // CHECK: %[[VAL_49:.*]] = arith.addi %[[IX_TEMP]], %[[ROUND_X]]
   // CHECK: %[[LOWER:.*]] = arith.maxsi %[[ZERO]], %[[VAL_49]]
   // CHECK: %[[CLAMPED:.*]] = arith.minsi %[[XMAX]], %[[LOWER]]
   // CHECK: %[[IDX:.*]] = arith.index_cast %[[CLAMPED]]
@@ -374,36 +368,30 @@ func.func @resize_bilinear_fp(%input: tensor<1x23x24x1xf32>) -> () {
   // CHECK-DAG: %[[X_MAX:.*]] = arith.constant 23
   // CHECK: %[[Y:.+]] = arith.index_cast %[[IDX_1]]
   // CHECK: %[[X:.+]] = arith.index_cast %[[IDX_2]]
-  // CHECK-DAG: %[[ISCALE_Y_N:.*]] = arith.constant 4
-  // CHECK-DAG: %[[ISCALE_Y_D:.*]] = arith.constant 1
-  // CHECK-DAG: %[[ISCALE_X_N:.*]] = arith.constant 4
-  // CHECK-DAG: %[[ISCALE_X_D:.*]] = arith.constant 1
-  // CHECK-DAG: %[[IOFFSET_Y:.*]] = arith.constant 0
-  // CHECK-DAG: %[[IOFFSET_X:.*]] = arith.constant 0
-  // CHECK-DAG: %[[IBORDER_Y:.*]] = arith.constant 0
-  // CHECK-DAG: %[[IBORDER_X:.*]] = arith.constant 0
-
-  // CHECK: %[[Y0:.+]] = arith.uitofp %[[Y]]
-  // CHECK: %[[SCALE_Y_N:.*]] = arith.uitofp %[[ISCALE_Y_N]]
-  // CHECK: %[[SCALE_Y_D:.*]] = arith.uitofp %[[ISCALE_Y_D]]
-  // CHECK: %[[OFFSET_Y:.*]] = arith.sitofp %[[IOFFSET_Y]]
-  // CHECK: %[[VAL_29:.*]] = arith.mulf %[[Y0]], %[[SCALE_Y_D]]
-  // CHECK: %[[VAL_31:.*]] = arith.addf %[[VAL_29]], %[[OFFSET_Y]]
-  // CHECK: %[[VAL_33:.*]] = arith.divf %[[VAL_31]], %[[SCALE_Y_N]]
-  // CHECK: %[[VAL_35:.*]] = math.floor %[[VAL_33]]
-  // CHECK: %[[D_Y:.*]] = arith.subf %[[VAL_33]], %[[VAL_35]]
-  // CHECK: %[[I_Y:.*]] = arith.fptosi %[[VAL_35]]
-
-  // CHECK: %[[X0:.+]] = arith.uitofp %[[X]]
-  // CHECK: %[[SCALE_X_N:.*]] = arith.uitofp %[[ISCALE_X_N]]
-  // CHECK: %[[SCALE_X_D:.*]] = arith.uitofp %[[ISCALE_X_D]]
-  // CHECK: %[[OFFSET_X:.*]] = arith.sitofp %[[IOFFSET_X]]
-  // CHECK: %[[VAL_30:.*]] = arith.mulf %[[X0]], %[[SCALE_X_D]]
-  // CHECK: %[[VAL_32:.*]] = arith.addf %[[VAL_30]], %[[OFFSET_X]]
-  // CHECK: %[[VAL_34:.*]] = arith.divf %[[VAL_32]], %[[SCALE_X_N]]
-  // CHECK: %[[VAL_36:.*]] = math.floor %[[VAL_34]]
-  // CHECK: %[[D_X:.*]] = arith.subf %[[VAL_34]], %[[VAL_36]]
-  // CHECK: %[[I_X:.*]] = arith.fptosi %[[VAL_36]]
+  // CHECK-DAG: %[[SCALE_Y_N:.*]] = arith.constant 4
+  // CHECK-DAG: %[[SCALE_Y_D:.*]] = arith.constant 1
+  // CHECK-DAG: %[[SCALE_X_N:.*]] = arith.constant 4
+  // CHECK-DAG: %[[SCALE_X_D:.*]] = arith.constant 1
+  // CHECK-DAG: %[[OFFSET_Y:.*]] = arith.constant 0
+  // CHECK-DAG: %[[OFFSET_X:.*]] = arith.constant 0
+  // CHECK-DAG: %[[BORDER_Y:.*]] = arith.constant 0
+  // CHECK-DAG: %[[BORDER_X:.*]] = arith.constant 0
+
+  // CHECK: %[[VAL_29:.*]] = arith.muli %[[Y]], %[[SCALE_Y_D]]
+  // CHECK: %[[Y_TEMP:.*]] = arith.addi %[[VAL_29]], %[[OFFSET_Y]]
+  // CHECK: %[[I_Y:.*]] = arith.floordivsi %[[Y_TEMP]], %[[SCALE_Y_N]]
+  // CHECK: %[[RY:.*]] = arith.remsi %[[Y_TEMP]], %[[SCALE_Y_N]]
+  // CHECK: %[[RY_FP:.*]] = arith.sitofp %[[RY]]
+  // CHECK: %[[SCALE_Y_N_FP:.*]] = arith.uitofp %[[SCALE_Y_N]]
+  // CHECK: %[[D_Y:.*]] = arith.divf %[[RY_FP]], %[[SCALE_Y_N_FP]]
+
+  // CHECK: %[[VAL_30:.*]] = arith.muli %[[X]], %[[SCALE_X_D]]
+  // CHECK: %[[X_TEMP:.*]] = arith.addi %[[VAL_30]], %[[OFFSET_X]]
+  // CHECK: %[[I_X:.*]] = arith.floordivsi %[[X_TEMP]], %[[SCALE_X_N]]
+  // CHECK: %[[RX:.*]] = arith.remsi %[[X_TEMP]], %[[SCALE_X_N]]
+  // CHECK: %[[RX_FP:.*]] = arith.sitofp %[[RX]]
+  // CHECK: %[[SCALE_X_N_FP:.*]] = arith.uitofp %[[SCALE_X_N]]
+  // CHECK: %[[D_X:.*]] = arith.divf %[[RX_FP]], %[[SCALE_X_N_FP]]
 
   // Compute the left, right, and top indices for the bilinear interpolation.

@fabrizio-indirli
Copy link
Contributor Author

@rsuderman @GeorgeARM @Hsiangkai
Could you please have a look when you have time? Thank you in advance :)

GeorgeARM
GeorgeARM approved these changes Apr 15, 2024
View reviewed changes
Copy link
Contributor

@GeorgeARM GeorgeARM left a comment •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. Aligned with the spec. Leaving to @kuhar @rsuderman

@GeorgeARM GeorgeARM requested a review from sjarus April 15, 2024 17:48
@Hsiangkai Hsiangkai merged commit 4edeaff into llvm:main Apr 17, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@sjarus sjarus sjarus approved these changes

@GeorgeARM GeorgeARM GeorgeARM approved these changes

Assignees
No one assigned
Labels
mlir:linalg mlir
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
5 participants
@fabrizio-indirli @llvmbot @sjarus @GeorgeARM @Hsiangkai