add a snitch_stream kernel for relu

kernels/relu/16x16xf64/Makefile

@@ -8,5 +8,6 @@ TESTS += ssr.x
 TESTS += ssr_frep.x
 TESTS += ssr_frep_unroll.x
 TESTS += linalg.x
+TESTS += snitch_stream.x
 
 include ../../Makefile.kernels

kernels/relu/16x16xf64/snitch_stream.xdsl.mlir

@@ -0,0 +1,22 @@
+riscv.assembly_section ".text" {
+  riscv.directive ".globl" "relu"
+  riscv.directive ".p2align" "2"
+  riscv_func.func @relu(%X : !riscv.reg<a0>, %Y : !riscv.reg<a1>) {
+    %X_moved = riscv.mv %X : (!riscv.reg<a0>) -> !riscv.reg<>
+    %Y_moved = riscv.mv %Y : (!riscv.reg<a1>) -> !riscv.reg<>
+
+    %zero_int = riscv.get_register : () -> !riscv.reg<zero>
+    %zero_float = riscv.fcvt.d.w %zero_int : (!riscv.reg<zero>) -> !riscv.freg<ft3>
+
+    %stride_pattern = "snitch_stream.stride_pattern"() {"ub" = [#int<16>, #int<16>], "strides" = [#int<128>, #int<8>], "dm" = #int<31>} : () -> !snitch_stream.stride_pattern_type
+    %X_stream = "snitch_stream.strided_read"(%X_moved, %stride_pattern) {"dm" = #int<0>, "rank" = #int<2>} : (!riscv.reg<>, !snitch_stream.stride_pattern_type) -> !stream.readable<!riscv.freg<ft0>>
+    %Y_stream = "snitch_stream.strided_write"(%Y_moved, %stride_pattern) {"dm" = #int<1>, "rank" = #int<2>} : (!riscv.reg<>, !snitch_stream.stride_pattern_type) -> !stream.writable<!riscv.freg<ft1>>
+    %count = riscv.li 256 : () -> !riscv.reg<>
+    "snitch_stream.generic"(%count, %X_stream, %Y_stream) <{"operandSegmentSizes" = array<i32: 1, 1, 1>}> ({
+    ^0(%x : !riscv.freg<ft0>):
+        %res = riscv.fmax.d %x, %zero_float : (!riscv.freg<ft0>, !riscv.freg<ft3>) -> !riscv.freg<ft1>
+        snitch_stream.yield %res : !riscv.freg<ft1>
+    }) : (!riscv.reg<>, !stream.readable<!riscv.freg<ft0>>, !stream.writable<!riscv.freg<ft1>>) -> ()
+    riscv_func.return
+  }
+}

snitch/Makefile.rules

@@ -71,14 +71,21 @@ LDFLAGS += -lsnRuntime
 
 # XDSL
 
-XDSLOPT_PASSES = 
+XDSLOPT_PASSES =
+XDSLOPT_PASSES += lower-affine
 XDSLOPT_PASSES += convert-func-to-riscv-func
 XDSLOPT_PASSES += convert-memref-to-riscv
 XDSLOPT_PASSES += convert-arith-to-riscv
 XDSLOPT_PASSES += convert-scf-to-riscv-scf
 XDSLOPT_PASSES += dce
 XDSLOPT_PASSES += reconcile-unrealized-casts
+XDSLOPT_PASSES += snitch-allocate-registers
+XDSLOPT_PASSES += convert-snitch-stream-to-snitch
+XDSLOPT_PASSES += lower-snitch
 XDSLOPT_PASSES += canonicalize
+XDSLOPT_PASSES += riscv-scf-loop-range-folding
+XDSLOPT_PASSES += canonicalize
+XDSLOPT_PASSES += riscv-reduce-register-pressure
 XDSLOPT_PASSES += riscv-allocate-registers
 XDSLOPT_PASSES += canonicalize
 XDSLOPT_PASSES += lower-riscv-func