Skip to content

Commit

Permalink
Update aie2_colorThreshold.py RTPs
Browse files Browse the repository at this point in the history
  • Loading branch information
@jgmelber
jgmelber authored Dec 5, 2023
1 parent b6016fd commit dc48dbb
Showing 1 changed file with 46 additions and 38 deletions.
84 changes: 46 additions & 38 deletions reference_designs/ipu-xrt/vision_pipelines/color_threshold/aie2_colorThreshold.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -112,12 +112,6 @@ def device_body():
# Compute tile 2
@core(ComputeTile2, "threshold.cc.o")
def core_body():
# thresholdValue = arith.trunci(T.i16(), memref.load(rtpComputeTile2, [0]))
# maxValue = arith.trunci(T.i16(), memref.load(rtpComputeTile2, [1]))
# thresholdType = arith.trunci(T.i8(), memref.load(rtpComputeTile2, [2]))
maxValue = arith.constant(255, T.i16())
thresholdValue = arith.constant(50, T.i16())
thresholdType = arith.constant(0, T.i8())
# for _ in for_(4096):
for _ in for_(sys.maxsize):
elemIn = acquire(
Expand All @@ -133,6 +127,14 @@ def core_body():
T.memref(lineWidth, T.ui8()),
).acquired_elem()

# RTPs written from the instruction stream must be read right before the kernel
# after the ObjectFIFO acquires
thresholdValue = arith.trunci(T.i16(), memref.load(rtpComputeTile2, [0]))
maxValue = arith.trunci(T.i16(), memref.load(rtpComputeTile2, [1]))
thresholdType = arith.trunci(T.i8(), memref.load(rtpComputeTile2, [2]))
# maxValue = arith.constant(255, T.i16())
# thresholdValue = arith.constant(50, T.i16())
# thresholdType = arith.constant(0, T.i8())
Call(
thresholdLine,
[
Expand All @@ -152,12 +154,6 @@ def core_body():
# Compute tile 3
@core(ComputeTile3, "threshold.cc.o")
def core_body():
# thresholdValue = arith.trunci(T.i16(), memref.load(rtpComputeTile3, [0]))
# maxValue = arith.trunci(T.i16(), memref.load(rtpComputeTile3, [1]))
# thresholdType = arith.trunci(T.i8(), memref.load(rtpComputeTile3, [2]))
maxValue = arith.constant(255, T.i16())
thresholdValue = arith.constant(50, T.i16())
thresholdType = arith.constant(0, T.i8())
# for _ in for_(4096):
for _ in for_(sys.maxsize):
elemIn = acquire(
Expand All @@ -172,7 +168,15 @@ def core_body():
1,
T.memref(lineWidth, T.ui8()),
).acquired_elem()


# RTPs written from the instruction stream must be read right before the kernel
# after the ObjectFIFO acquires
thresholdValue = arith.trunci(T.i16(), memref.load(rtpComputeTile3, [0]))
maxValue = arith.trunci(T.i16(), memref.load(rtpComputeTile3, [1]))
thresholdType = arith.trunci(T.i8(), memref.load(rtpComputeTile3, [2]))
# maxValue = arith.constant(255, T.i16())
# thresholdValue = arith.constant(50, T.i16())
# thresholdType = arith.constant(0, T.i8())
Call(
thresholdLine,
[
Expand All @@ -192,12 +196,6 @@ def core_body():
# Compute tile 4
@core(ComputeTile4, "threshold.cc.o")
def core_body():
# thresholdValue = arith.trunci(T.i16(), memref.load(rtpComputeTile4, [0]))
# maxValue = arith.trunci(T.i16(), memref.load(rtpComputeTile4, [1]))
# thresholdType = arith.trunci(T.i8(), memref.load(rtpComputeTile4, [2]))
maxValue = arith.constant(255, T.i16())
thresholdValue = arith.constant(50, T.i16())
thresholdType = arith.constant(0, T.i8())
# for _ in for_(4096):
for _ in for_(sys.maxsize):
elemIn = acquire(
Expand All @@ -213,6 +211,14 @@ def core_body():
T.memref(lineWidth, T.ui8()),
).acquired_elem()

# RTPs written from the instruction stream must be read right before the kernel
# after the ObjectFIFO acquires
thresholdValue = arith.trunci(T.i16(), memref.load(rtpComputeTile4, [0]))
maxValue = arith.trunci(T.i16(), memref.load(rtpComputeTile4, [1]))
thresholdType = arith.trunci(T.i8(), memref.load(rtpComputeTile4, [2]))
# maxValue = arith.constant(255, T.i16())
# thresholdValue = arith.constant(50, T.i16())
# thresholdType = arith.constant(0, T.i8())
Call(
thresholdLine,
[
Expand All @@ -231,13 +237,7 @@ def core_body():

# Compute tile 5
@core(ComputeTile5, "threshold.cc.o")
def core_body():
# thresholdValue = arith.trunci(T.i16(), memref.load(rtpComputeTile5, [0]))
# maxValue = arith.trunci(T.i16(), memref.load(rtpComputeTile5, [1]))
# thresholdType = arith.trunci(T.i8(), memref.load(rtpComputeTile5, [2]))
maxValue = arith.constant(255, T.i16())
thresholdValue = arith.constant(50, T.i16())
thresholdType = arith.constant(0, T.i8())
def core_body():)
# for _ in for_(4096):
for _ in for_(sys.maxsize):
elemIn = acquire(
Expand All @@ -253,6 +253,14 @@ def core_body():
T.memref(lineWidth, T.ui8()),
).acquired_elem()

# RTPs written from the instruction stream must be read right before the kernel
# after the ObjectFIFO acquires
thresholdValue = arith.trunci(T.i16(), memref.load(rtpComputeTile5, [0]))
maxValue = arith.trunci(T.i16(), memref.load(rtpComputeTile5, [1]))
thresholdType = arith.trunci(T.i8(), memref.load(rtpComputeTile5, [2]))
# maxValue = arith.constant(255, T.i16())
# thresholdValue = arith.constant(50, T.i16())
# thresholdType = arith.constant(0, T.i8()
Call(
thresholdLine,
[
Expand Down Expand Up @@ -281,21 +289,21 @@ def core_body():
)
def sequence(inTensor, notUsed, outTensor):
# thresholdValue, maxValue, thresholdType
# IpuWriteRTPOp("rtpComputeTile2", col = 0, row = 2, index = 0, value = 50)
# IpuWriteRTPOp("rtpComputeTile2", col = 0, row = 2, index = 1, value = 255)
# IpuWriteRTPOp("rtpComputeTile2", col = 0, row = 2, index = 2, value = 0)
IpuWriteRTPOp("rtpComputeTile2", col = 0, row = 2, index = 0, value = 50)
IpuWriteRTPOp("rtpComputeTile2", col = 0, row = 2, index = 1, value = 255)
IpuWriteRTPOp("rtpComputeTile2", col = 0, row = 2, index = 2, value = 0)

# IpuWriteRTPOp("rtpComputeTile3", col = 0, row = 3, index = 0, value = 50)
# IpuWriteRTPOp("rtpComputeTile3", col = 0, row = 3, index = 1, value = 255)
# IpuWriteRTPOp("rtpComputeTile3", col = 0, row = 3, index = 2, value = 0)
IpuWriteRTPOp("rtpComputeTile3", col = 0, row = 3, index = 0, value = 50)
IpuWriteRTPOp("rtpComputeTile3", col = 0, row = 3, index = 1, value = 255)
IpuWriteRTPOp("rtpComputeTile3", col = 0, row = 3, index = 2, value = 0)

# IpuWriteRTPOp("rtpComputeTile4", col = 0, row = 4, index = 0, value = 50)
# IpuWriteRTPOp("rtpComputeTile4", col = 0, row = 4, index = 1, value = 255)
# IpuWriteRTPOp("rtpComputeTile4", col = 0, row = 4, index = 2, value = 0)
IpuWriteRTPOp("rtpComputeTile4", col = 0, row = 4, index = 0, value = 50)
IpuWriteRTPOp("rtpComputeTile4", col = 0, row = 4, index = 1, value = 255)
IpuWriteRTPOp("rtpComputeTile4", col = 0, row = 4, index = 2, value = 0)

# IpuWriteRTPOp("rtpComputeTile5", col = 0, row = 5, index = 0, value = 50)
# IpuWriteRTPOp("rtpComputeTile5", col = 0, row = 5, index = 1, value = 255)
# IpuWriteRTPOp("rtpComputeTile5", col = 0, row = 5, index = 2, value = 0)
IpuWriteRTPOp("rtpComputeTile5", col = 0, row = 5, index = 0, value = 50)
IpuWriteRTPOp("rtpComputeTile5", col = 0, row = 5, index = 1, value = 255)
IpuWriteRTPOp("rtpComputeTile5", col = 0, row = 5, index = 2, value = 0)

ipu_dma_memcpy_nd(
metadata="inOOB_L3L2",
Expand Down

0 comments on commit dc48dbb

Please sign in to comment.