:meta-name-twitter_title: Gradle User Manual: Version {gradleVersion}
:meta-name-twitter_description: {description}
:meta-name-twitter_url: {docsUrl}/{gradleVersion}/userguide/{docname}.html
:meta-name-twitter_image: {website}/images/gradle-256x256.png

toc::[leveloffset=+2]

[[part:about_gradle]]
== **OVERVIEW**

include::userguide.adoc[leveloffset=+2]
include::about_manual.adoc[leveloffset=+2]

'''

[[part:releases]]
== **RELEASES**

}

// CHECK-LABEL: func @testEmptyf16
func.func @testEmptyf16() -> (tensor<5xf16>) {
  %0 = "tf.Const"() { value = dense<5> : tensor<i32> } : () -> tensor<i32>

  // CHECK: [[VAL:%.+]] = "tf.Const"() <{value =  dense<0.000000e+00> : tensor<5xf16>}>
  // CHECK: return [[VAL]]
  %1 = "tf.Empty"(%0) : (tensor<i32>) -> (tensor<5xf16>)
  func.return %1 : tensor<5xf16>
}

// CHECK-LABEL: func @testEmptybf16

// Float16-DAG: %[[b:.*]] = arith.constant dense<0.000000e+00> : tensor<16xf16>
// Float16-DAG: %[[const:.*]] = "tfl.no_value"() <{value}> : () -> none
// Float16-DAG: %[[dq_w:.*]] = "tfl.dequantize"(%[[w]]) : (tensor<3x3x3x8x16xf16>) -> tensor<3x3x3x8x16xf32>
// Float16-DAG: %[[dq_b:.*]] = "tfl.dequantize"(%[[b]]) : (tensor<16xf16>) -> tensor<16xf32>

// CHECK-LABEL: func @floordiv_f16_broadcast
func.func @floordiv_f16_broadcast(%arg0: tensor<2x3xf16>, %arg1: tensor<3xf16>) -> tensor<2x3xf16> {
  // CHECK-NEXT:  chlo.broadcast_divide
  // CHECK-NEXT:  mhlo.floor
  // CHECK-NEXT:  return
  %0 = "tf.FloorDiv"(%arg0, %arg1) : (tensor<2x3xf16>, tensor<3xf16>) -> tensor<2x3xf16>
  func.return %0: tensor<2x3xf16>
}

// -----

// CHECK-LABEL: func @floordiv_dynamic

// -----

func.func @testInvalidSelect(%arg0: tensor<3xi1>, %arg1: tensor<2x3xf16>, %arg2: tensor<2x3xf16>) -> tensor<2x3xf16> {
  // expected-error @+1 {{requires that, when pred is a vector, the shape matches the first dimension of t and e}}
  %0 = "tf.Select"(%arg0, %arg1, %arg2) : (tensor<3xi1>, tensor<2x3xf16>, tensor<2x3xf16>) -> tensor<2x3xf16>
  func.return %0: tensor<2x3xf16>
}

// -----

  %cst_1 = arith.constant dense<1.5> : tensor<256xf32>
  %cst_2 = arith.constant dense<2.0> : tensor<1x1x1x256xf32>
  %0 = "tfl.conv_2d"(%arg0, %cst_0, %cst_1) {dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32} : (tensor<1x8x8x207xf32>, tensor<256x3x3x207xf32>, tensor<256xf32>) -> tensor<1x8x8x256xf32>

#define m(off)	(32*4 + off)(SP)
#define zsqr(off) (32*5 + off)(SP)
#define tmp(off)  (32*6 + off)(SP)
#define rptr	  (32*7)(SP)

//func p256PointDoubleAsm(res, in *P256Point)
TEXT ·p256PointDoubleAsm(SB),NOSPLIT,$256-16
	// Move input to stack in order to free registers
	MOVQ res+0(FP), AX
	MOVQ in+8(FP), BX

	MOVOU (16*0)(BX), X0
	MOVOU (16*1)(BX), X1
	MOVOU (16*2)(BX), X2
	MOVOU (16*3)(BX), X3

        input_shape=(1, 16, 16, 8),
        # Uses large filter to exceed the constant size threshold of 64KiB
        # (specified by `kDefaultConstantSizeThresholdInBytes`) for unfreezing.
        filter_shape=(256, 8, 8, 16),
        use_variable=True,
    )

    signature_keys = [signature_key]

    quantization_options = quant_opts_pb2.QuantizationOptions(
        quantization_method=quant_opts_pb2.QuantizationMethod(