Skip to content

Inline snrt

inline_snrt

SnrtConstants dataclass

Bases: ABC

Constants used when compiling the snitch runtime, depend on the exact snitch architecture target.

Source code in xdsl/transforms/inline_snrt.py 
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
@dataclass(frozen=True)
class SnrtConstants(ABC):
    """
    Constants used when compiling the snitch runtime, depend on the exact snitch
    architecture target.
    """

    cluster_num: int = 4
    """
    Number of snitch clusters on the chip
    """
    cluster_core_num: int = 9
    """
    Number of cores per cluster (1 dma + 8 compute)
    """
    base_hartid: int = 0
    """
    SNRT expects the snitch cores to be numbered contiguously starting at base_hartid
    """
    cluster_dm_core_num: int = 1
    """
    Number of dm cores per cluster. Compute core num = cluster_core_num - cluster_dm_core_num
    """

cluster_num: int = 4 class-attribute instance-attribute

Number of snitch clusters on the chip

cluster_core_num: int = 9 class-attribute instance-attribute

Number of cores per cluster (1 dma + 8 compute)

base_hartid: int = 0 class-attribute instance-attribute

SNRT expects the snitch cores to be numbered contiguously starting at base_hartid

cluster_dm_core_num: int = 1 class-attribute instance-attribute

Number of dm cores per cluster. Compute core num = cluster_core_num - cluster_dm_core_num

__init__(cluster_num: int = 4, cluster_core_num: int = 9, base_hartid: int = 0, cluster_dm_core_num: int = 1) -> None

LowerClusterHWBarrier

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
class LowerClusterHWBarrier(RewritePattern):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.ClusterHwBarrierOp, rewriter: PatternRewriter, /
    ):
        """
        Lowers to:

        /// Synchronize cores in a cluster with a hardware barrier
        inline void snrt_cluster_hw_barrier() {
            asm volatile("csrr x0, 0x7C2" ::: "memory");
        }
        """
        rewriter.replace_op(
            op,
            [
                zero := riscv.GetRegisterOp(riscv.Registers.ZERO),
                riscv.CsrrsOp(
                    rd=riscv.Registers.ZERO,
                    rs1=zero,
                    csr=IntegerAttr(0x7C2, 12),
                ),
            ],
            [],
        )

match_and_rewrite(op: snitch_runtime.ClusterHwBarrierOp, rewriter: PatternRewriter)

Lowers to:

/// Synchronize cores in a cluster with a hardware barrier inline void snrt_cluster_hw_barrier() { asm volatile("csrr x0, 0x7C2" ::: "memory"); }

Source code in xdsl/transforms/inline_snrt.py 
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.ClusterHwBarrierOp, rewriter: PatternRewriter, /
):
    """
    Lowers to:

    /// Synchronize cores in a cluster with a hardware barrier
    inline void snrt_cluster_hw_barrier() {
        asm volatile("csrr x0, 0x7C2" ::: "memory");
    }
    """
    rewriter.replace_op(
        op,
        [
            zero := riscv.GetRegisterOp(riscv.Registers.ZERO),
            riscv.CsrrsOp(
                rd=riscv.Registers.ZERO,
                rs1=zero,
                csr=IntegerAttr(0x7C2, 12),
            ),
        ],
        [],
    )

LowerSSRDisable

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
class LowerSSRDisable(RewritePattern):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.SsrDisableOp, rewriter: PatternRewriter, /
    ):
        """
        Lowers to:

            /// Disable SSR.
            inline void snrt_ssr_disable() {
            #ifdef __TOOLCHAIN_LLVM__
                __builtin_ssr_disable();
            #else
                asm volatile("csrci 0x7C0, 1\n");
            #endif
            }

        P.S. This specific rewrite ignores the LLVM case and goes
                straight to the generic one.
        """
        rewriter.replace_op(
            op,
            [
                riscv.CsrrciOp(csr=IntegerAttr(0x7C0, 12), immediate=IntegerAttr(1, 4)),
            ],
            [],
        )

match_and_rewrite(op: snitch_runtime.SsrDisableOp, rewriter: PatternRewriter) 

    Lowers to:

        /// Disable SSR.
        inline void snrt_ssr_disable() {
        #ifdef __TOOLCHAIN_LLVM__
            __builtin_ssr_disable();
        #else
            asm volatile("csrci 0x7C0, 1

"); #endif }

    P.S. This specific rewrite ignores the LLVM case and goes
            straight to the generic one.
Source code in xdsl/transforms/inline_snrt.py 
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.SsrDisableOp, rewriter: PatternRewriter, /
):
    """
    Lowers to:

        /// Disable SSR.
        inline void snrt_ssr_disable() {
        #ifdef __TOOLCHAIN_LLVM__
            __builtin_ssr_disable();
        #else
            asm volatile("csrci 0x7C0, 1\n");
        #endif
        }

    P.S. This specific rewrite ignores the LLVM case and goes
            straight to the generic one.
    """
    rewriter.replace_op(
        op,
        [
            riscv.CsrrciOp(csr=IntegerAttr(0x7C0, 12), immediate=IntegerAttr(1, 4)),
        ],
        [],
    )

LowerDMAStart1D

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
class LowerDMAStart1D(RewritePattern):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.DmaStart1DOp, rewriter: PatternRewriter, /
    ):
        """
        Lowers to:

        /// Initiate an asynchronous 1D DMA transfer.
        inline snrt_dma_txid_t snrt_dma_start_1d(void *dst, const void *src,
                                                size_t size) {
            return snrt_dma_start_1d_wideptr((size_t)dst, (size_t)src, size);
        }
        """
        reg_t = riscv.Registers.UNALLOCATED_INT
        rewriter.replace_op(
            op,
            [
                zero := riscv.GetRegisterOp(riscv.Registers.ZERO),
                # "Take a void* (assumed 32bit) and make it a 32 bit-wide RISC-V register"
                i32_dst := builtin.UnrealizedConversionCastOp.get(
                    [op.dst],
                    [reg_t],
                ),
                # "Take a void* (assumed 32bit) and make it a 32 bit-wide RISC-V register"
                i32_src := builtin.UnrealizedConversionCastOp.get(
                    [op.src],
                    [reg_t],
                ),
                # "Convert an IR-level i32 to a RISC-V register"
                i32_size := builtin.UnrealizedConversionCastOp.get(
                    [op.size],
                    [reg_t],
                ),
                riscv_snitch.DMSourceOp(i32_src, zero),
                riscv_snitch.DMDestinationOp(i32_dst, zero),
                copy_imm := riscv_snitch.DMCopyImmOp(i32_size, 0),
                tx_id := builtin.UnrealizedConversionCastOp.get(
                    [copy_imm], [builtin.i32]
                ),
            ],
            new_results=tx_id.results,
        )

match_and_rewrite(op: snitch_runtime.DmaStart1DOp, rewriter: PatternRewriter)

Lowers to:

/// Initiate an asynchronous 1D DMA transfer. inline snrt_dma_txid_t snrt_dma_start_1d(void dst, const void src, size_t size) { return snrt_dma_start_1d_wideptr((size_t)dst, (size_t)src, size); }

Source code in xdsl/transforms/inline_snrt.py 
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.DmaStart1DOp, rewriter: PatternRewriter, /
):
    """
    Lowers to:

    /// Initiate an asynchronous 1D DMA transfer.
    inline snrt_dma_txid_t snrt_dma_start_1d(void *dst, const void *src,
                                            size_t size) {
        return snrt_dma_start_1d_wideptr((size_t)dst, (size_t)src, size);
    }
    """
    reg_t = riscv.Registers.UNALLOCATED_INT
    rewriter.replace_op(
        op,
        [
            zero := riscv.GetRegisterOp(riscv.Registers.ZERO),
            # "Take a void* (assumed 32bit) and make it a 32 bit-wide RISC-V register"
            i32_dst := builtin.UnrealizedConversionCastOp.get(
                [op.dst],
                [reg_t],
            ),
            # "Take a void* (assumed 32bit) and make it a 32 bit-wide RISC-V register"
            i32_src := builtin.UnrealizedConversionCastOp.get(
                [op.src],
                [reg_t],
            ),
            # "Convert an IR-level i32 to a RISC-V register"
            i32_size := builtin.UnrealizedConversionCastOp.get(
                [op.size],
                [reg_t],
            ),
            riscv_snitch.DMSourceOp(i32_src, zero),
            riscv_snitch.DMDestinationOp(i32_dst, zero),
            copy_imm := riscv_snitch.DMCopyImmOp(i32_size, 0),
            tx_id := builtin.UnrealizedConversionCastOp.get(
                [copy_imm], [builtin.i32]
            ),
        ],
        new_results=tx_id.results,
    )

LowerDMAStart1DWidePtr

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
class LowerDMAStart1DWidePtr(RewritePattern):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.DmaStart1DWideptrOp, rewriter: PatternRewriter, /
    ):
        """
        Lowers to:

        /// Initiate an asynchronous 1D DMA transfer with wide 64-bit pointers.
        inline snrt_dma_txid_t snrt_dma_start_1d_wideptr(uint64_t dst, uint64_t src,
                                                        size_t size) {
            // Current DMA does not allow transfers with size == 0 (blocks)
            // TODO(colluca) remove this check once new DMA is integrated
            if (size > 0) {
                register uint32_t reg_dst_low asm("a0") = dst >> 0;    // 10
                register uint32_t reg_dst_high asm("a1") = dst >> 32;  // 11
                register uint32_t reg_src_low asm("a2") = src >> 0;    // 12
                register uint32_t reg_src_high asm("a3") = src >> 32;  // 13
                register uint32_t reg_size asm("a4") = size;           // 14

                // dmsrc a2, a3
                asm volatile(
                    ".word (0b0000000 << 25) | \
                        (     (13) << 20) | \
                        (     (12) << 15) | \
                        (    0b000 << 12) | \
                        (0b0101011 <<  0)   \n" ::"r"(reg_src_high),
                    "r"(reg_src_low));

                // dmdst a0, a1
                asm volatile(
                    ".word (0b0000001 << 25) | \
                        (     (11) << 20) | \
                        (     (10) << 15) | \
                        (    0b000 << 12) | \
                        (0b0101011 <<  0)   \n" ::"r"(reg_dst_high),
                    "r"(reg_dst_low));

                // dmcpyi a0, a4, 0b00
                register uint32_t reg_txid asm("a0");  // 10
                asm volatile(
                    ".word (0b0000010 << 25) | \
                        (  0b00000 << 20) | \
                        (     (14) << 15) | \
                        (    0b000 << 12) | \
                        (     (10) <<  7) | \
                        (0b0101011 <<  0)   \n"
                    : "=r"(reg_txid)
                    : "r"(reg_size));

                return reg_txid;
            } else {
                return -1;
            }
        }

        P.S. We only implement taking the top branch for now.
        """
        reg_t = riscv.Registers.UNALLOCATED_INT
        rewriter.replace_op(
            op,
            [
                # "Take an ui64 and split it in two 32 bit-wide RISC-V registers"
                split_i64_dst := builtin.UnrealizedConversionCastOp.get(
                    [op.dst],
                    [reg_t, reg_t],
                ),
                # "Take an ui64 and split it in two 32 bit-wide RISC-V registers"
                split_i64_src := builtin.UnrealizedConversionCastOp.get(
                    [op.src],
                    [reg_t, reg_t],
                ),
                # "Convert an IR-level i32 to a RISC-V register"
                i32_size := builtin.UnrealizedConversionCastOp.get(
                    [op.size],
                    [reg_t],
                ),
                riscv_snitch.DMSourceOp(
                    split_i64_src.results[0], split_i64_src.results[1]
                ),
                riscv_snitch.DMDestinationOp(
                    split_i64_dst.results[0], split_i64_dst.results[1]
                ),
                copy_imm := riscv_snitch.DMCopyImmOp(i32_size, 0),
                tx_id := builtin.UnrealizedConversionCastOp.get(
                    [copy_imm], [builtin.i32]
                ),
            ],
            new_results=tx_id.results,
        )

match_and_rewrite(op: snitch_runtime.DmaStart1DWideptrOp, rewriter: PatternRewriter) 

    Lowers to:

    /// Initiate an asynchronous 1D DMA transfer with wide 64-bit pointers.
    inline snrt_dma_txid_t snrt_dma_start_1d_wideptr(uint64_t dst, uint64_t src,
                                                    size_t size) {
        // Current DMA does not allow transfers with size == 0 (blocks)
        // TODO(colluca) remove this check once new DMA is integrated
        if (size > 0) {
            register uint32_t reg_dst_low asm("a0") = dst >> 0;    // 10
            register uint32_t reg_dst_high asm("a1") = dst >> 32;  // 11
            register uint32_t reg_src_low asm("a2") = src >> 0;    // 12
            register uint32_t reg_src_high asm("a3") = src >> 32;  // 13
            register uint32_t reg_size asm("a4") = size;           // 14

            // dmsrc a2, a3
            asm volatile(
                ".word (0b0000000 << 25) |                         (     (13) << 20) |                         (     (12) << 15) |                         (    0b000 << 12) |                         (0b0101011 <<  0)

" ::"r"(reg_src_high), "r"(reg_src_low));

            // dmdst a0, a1
            asm volatile(
                ".word (0b0000001 << 25) |                         (     (11) << 20) |                         (     (10) << 15) |                         (    0b000 << 12) |                         (0b0101011 <<  0)

" ::"r"(reg_dst_high), "r"(reg_dst_low));

            // dmcpyi a0, a4, 0b00
            register uint32_t reg_txid asm("a0");  // 10
            asm volatile(
                ".word (0b0000010 << 25) |                         (  0b00000 << 20) |                         (     (14) << 15) |                         (    0b000 << 12) |                         (     (10) <<  7) |                         (0b0101011 <<  0)

" : "=r"(reg_txid) : "r"(reg_size));

            return reg_txid;
        } else {
            return -1;
        }
    }

    P.S. We only implement taking the top branch for now.
Source code in xdsl/transforms/inline_snrt.py 
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.DmaStart1DWideptrOp, rewriter: PatternRewriter, /
):
    """
    Lowers to:

    /// Initiate an asynchronous 1D DMA transfer with wide 64-bit pointers.
    inline snrt_dma_txid_t snrt_dma_start_1d_wideptr(uint64_t dst, uint64_t src,
                                                    size_t size) {
        // Current DMA does not allow transfers with size == 0 (blocks)
        // TODO(colluca) remove this check once new DMA is integrated
        if (size > 0) {
            register uint32_t reg_dst_low asm("a0") = dst >> 0;    // 10
            register uint32_t reg_dst_high asm("a1") = dst >> 32;  // 11
            register uint32_t reg_src_low asm("a2") = src >> 0;    // 12
            register uint32_t reg_src_high asm("a3") = src >> 32;  // 13
            register uint32_t reg_size asm("a4") = size;           // 14

            // dmsrc a2, a3
            asm volatile(
                ".word (0b0000000 << 25) | \
                    (     (13) << 20) | \
                    (     (12) << 15) | \
                    (    0b000 << 12) | \
                    (0b0101011 <<  0)   \n" ::"r"(reg_src_high),
                "r"(reg_src_low));

            // dmdst a0, a1
            asm volatile(
                ".word (0b0000001 << 25) | \
                    (     (11) << 20) | \
                    (     (10) << 15) | \
                    (    0b000 << 12) | \
                    (0b0101011 <<  0)   \n" ::"r"(reg_dst_high),
                "r"(reg_dst_low));

            // dmcpyi a0, a4, 0b00
            register uint32_t reg_txid asm("a0");  // 10
            asm volatile(
                ".word (0b0000010 << 25) | \
                    (  0b00000 << 20) | \
                    (     (14) << 15) | \
                    (    0b000 << 12) | \
                    (     (10) <<  7) | \
                    (0b0101011 <<  0)   \n"
                : "=r"(reg_txid)
                : "r"(reg_size));

            return reg_txid;
        } else {
            return -1;
        }
    }

    P.S. We only implement taking the top branch for now.
    """
    reg_t = riscv.Registers.UNALLOCATED_INT
    rewriter.replace_op(
        op,
        [
            # "Take an ui64 and split it in two 32 bit-wide RISC-V registers"
            split_i64_dst := builtin.UnrealizedConversionCastOp.get(
                [op.dst],
                [reg_t, reg_t],
            ),
            # "Take an ui64 and split it in two 32 bit-wide RISC-V registers"
            split_i64_src := builtin.UnrealizedConversionCastOp.get(
                [op.src],
                [reg_t, reg_t],
            ),
            # "Convert an IR-level i32 to a RISC-V register"
            i32_size := builtin.UnrealizedConversionCastOp.get(
                [op.size],
                [reg_t],
            ),
            riscv_snitch.DMSourceOp(
                split_i64_src.results[0], split_i64_src.results[1]
            ),
            riscv_snitch.DMDestinationOp(
                split_i64_dst.results[0], split_i64_dst.results[1]
            ),
            copy_imm := riscv_snitch.DMCopyImmOp(i32_size, 0),
            tx_id := builtin.UnrealizedConversionCastOp.get(
                [copy_imm], [builtin.i32]
            ),
        ],
        new_results=tx_id.results,
    )

LowerDMAStart2DBase

Bases: RewritePattern, ABC

Source code in xdsl/transforms/inline_snrt.py 
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
class LowerDMAStart2DBase(RewritePattern, ABC):
    any_reg = riscv.Registers.UNALLOCATED_INT

    def generate_dma_instructions(
        self,
        dst_low: SSAValue | Operation,
        dst_high: SSAValue | Operation,
        src_low: SSAValue | Operation,
        src_high: SSAValue | Operation,
        size: SSAValue | Operation,
        dst_stride: SSAValue | Operation,
        src_stride: SSAValue | Operation,
        repeat: SSAValue | Operation,
    ) -> tuple[Sequence[Operation], Sequence[SSAValue]]:
        """
        Common function to generate the following sequence of operations:
            dmsrc %src_low, %src_high
            dmdst %dst_low, %dst_high
            dmstr %src_stride, %dst_stride
            dmrep %repeat
            %tx_id = dmcpyi %size, 0b10
            %tx_id_i32 unrealized_conversion_cast %tx_id to i32
        """

        return [
            riscv_snitch.DMSourceOp(src_low, src_high),
            riscv_snitch.DMDestinationOp(dst_low, dst_high),
            riscv_snitch.DMStrideOp(src_stride, dst_stride),
            riscv_snitch.DMRepOp(repeat),
            cpy_op := riscv_snitch.DMCopyImmOp(size, 0b10),
            tx_id := builtin.UnrealizedConversionCastOp.get([cpy_op], [builtin.i32]),
        ], tx_id.results

    def cast_i32(self, input_val: SSAValue):
        """
        Cast an i32 to riscv registers
        """
        return builtin.UnrealizedConversionCastOp.get([input_val], [self.any_reg])

    def cast_i64(self, input_val: SSAValue):
        """
        Cast an i64 to two riscv registers
        """
        return builtin.UnrealizedConversionCastOp.get(
            [input_val], [self.any_reg, self.any_reg]
        )

any_reg = riscv.Registers.UNALLOCATED_INT class-attribute instance-attribute

generate_dma_instructions(dst_low: SSAValue | Operation, dst_high: SSAValue | Operation, src_low: SSAValue | Operation, src_high: SSAValue | Operation, size: SSAValue | Operation, dst_stride: SSAValue | Operation, src_stride: SSAValue | Operation, repeat: SSAValue | Operation) -> tuple[Sequence[Operation], Sequence[SSAValue]]

Common function to generate the following sequence of operations

dmsrc %src_low, %src_high dmdst %dst_low, %dst_high dmstr %src_stride, %dst_stride dmrep %repeat %tx_id = dmcpyi %size, 0b10 %tx_id_i32 unrealized_conversion_cast %tx_id to i32

Source code in xdsl/transforms/inline_snrt.py 
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
def generate_dma_instructions(
    self,
    dst_low: SSAValue | Operation,
    dst_high: SSAValue | Operation,
    src_low: SSAValue | Operation,
    src_high: SSAValue | Operation,
    size: SSAValue | Operation,
    dst_stride: SSAValue | Operation,
    src_stride: SSAValue | Operation,
    repeat: SSAValue | Operation,
) -> tuple[Sequence[Operation], Sequence[SSAValue]]:
    """
    Common function to generate the following sequence of operations:
        dmsrc %src_low, %src_high
        dmdst %dst_low, %dst_high
        dmstr %src_stride, %dst_stride
        dmrep %repeat
        %tx_id = dmcpyi %size, 0b10
        %tx_id_i32 unrealized_conversion_cast %tx_id to i32
    """

    return [
        riscv_snitch.DMSourceOp(src_low, src_high),
        riscv_snitch.DMDestinationOp(dst_low, dst_high),
        riscv_snitch.DMStrideOp(src_stride, dst_stride),
        riscv_snitch.DMRepOp(repeat),
        cpy_op := riscv_snitch.DMCopyImmOp(size, 0b10),
        tx_id := builtin.UnrealizedConversionCastOp.get([cpy_op], [builtin.i32]),
    ], tx_id.results

cast_i32(input_val: SSAValue)

Cast an i32 to riscv registers

Source code in xdsl/transforms/inline_snrt.py 
270
271
272
273
274
def cast_i32(self, input_val: SSAValue):
    """
    Cast an i32 to riscv registers
    """
    return builtin.UnrealizedConversionCastOp.get([input_val], [self.any_reg])

cast_i64(input_val: SSAValue)

Cast an i64 to two riscv registers

Source code in xdsl/transforms/inline_snrt.py 
276
277
278
279
280
281
282
def cast_i64(self, input_val: SSAValue):
    """
    Cast an i64 to two riscv registers
    """
    return builtin.UnrealizedConversionCastOp.get(
        [input_val], [self.any_reg, self.any_reg]
    )

LowerDMAStart2DWideptr

Bases: LowerDMAStart2DBase

Source code in xdsl/transforms/inline_snrt.py 
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
class LowerDMAStart2DWideptr(LowerDMAStart2DBase):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.DmaStart2DWideptrOp, rewriter: PatternRewriter, /
    ):
        """
        Lowers to the equivalent of the snitch_runtime implementation:

        inline snrt_dma_txid_t snrt_dma_start_2d_wideptr(uint64_t dst, uint64_t src,
                                                         size_t size, size_t dst_stride,
                                                         size_t src_stride,
                                                         size_t repeat) {
            // Current DMA does not allow transfers with size == 0 (blocks)
            // TODO(colluca) remove this check once new DMA is integrated
            if (size > 0) {
                register uint32_t reg_dst_low asm("a0") = dst >> 0;       // 10
                register uint32_t reg_dst_high asm("a1") = dst >> 32;     // 11
                register uint32_t reg_src_low asm("a2") = src >> 0;       // 12
                register uint32_t reg_src_high asm("a3") = src >> 32;     // 13
                register uint32_t reg_size asm("a4") = size;              // 14
                register uint32_t reg_dst_stride asm("a5") = dst_stride;  // 15
                register uint32_t reg_src_stride asm("a6") = src_stride;  // 16
                register uint32_t reg_repeat asm("a7") = repeat;          // 17

                // dmsrc a0, a1
                asm volatile(
                    ".word (0b0000000 << 25) | \
                        (     (13) << 20) | \
                        (     (12) << 15) | \
                        (    0b000 << 12) | \
                        (0b0101011 <<  0)   \n" ::"r"(reg_src_high),
                    "r"(reg_src_low));

                // dmdst a0, a1
                asm volatile(
                    ".word (0b0000001 << 25) | \
                        (     (11) << 20) | \
                        (     (10) << 15) | \
                        (    0b000 << 12) | \
                        (0b0101011 <<  0)   \n" ::"r"(reg_dst_high),
                    "r"(reg_dst_low));

                // dmstr a5, a6
                asm volatile(
                    ".word (0b0000110 << 25) | \
                        (     (15) << 20) | \
                        (     (16) << 15) | \
                        (    0b000 << 12) | \
                        (0b0101011 <<  0)   \n"
                    :
                    : "r"(reg_dst_stride), "r"(reg_src_stride));

                // dmrep a7
                asm volatile(
                    ".word (0b0000111 << 25) | \
                        (     (17) << 15) | \
                        (    0b000 << 12) | \
                        (0b0101011 <<  0)   \n"
                    :
                    : "r"(reg_repeat));

                // dmcpyi a0, a4, 0b10
                register uint32_t reg_txid asm("a0");  // 10
                asm volatile(
                    ".word (0b0000010 << 25) | \
                        (  0b00010 << 20) | \
                        (     (14) << 15) | \
                        (    0b000 << 12) | \
                        (     (10) <<  7) | \
                        (0b0101011 <<  0)   \n"
                    : "=r"(reg_txid)
                    : "r"(reg_size));

                return reg_txid;
            } else {
                return -1;
            }
        }
        """
        rewriter.insert_op(
            [
                dst := self.cast_i64(op.dst),
                src := self.cast_i64(op.src),
                src_stride := self.cast_i32(op.src_stride),
                dst_stride := self.cast_i32(op.dst_stride),
                size := self.cast_i32(op.size),
                repeat := self.cast_i32(op.size),
            ]
        )
        rewriter.replace_op(
            op,
            *self.generate_dma_instructions(
                dst.results[0],
                dst.results[1],
                src.results[0],
                src.results[1],
                size,
                dst_stride,
                src_stride,
                repeat,
            ),
        )

match_and_rewrite(op: snitch_runtime.DmaStart2DWideptrOp, rewriter: PatternRewriter) 

    Lowers to the equivalent of the snitch_runtime implementation:

    inline snrt_dma_txid_t snrt_dma_start_2d_wideptr(uint64_t dst, uint64_t src,
                                                     size_t size, size_t dst_stride,
                                                     size_t src_stride,
                                                     size_t repeat) {
        // Current DMA does not allow transfers with size == 0 (blocks)
        // TODO(colluca) remove this check once new DMA is integrated
        if (size > 0) {
            register uint32_t reg_dst_low asm("a0") = dst >> 0;       // 10
            register uint32_t reg_dst_high asm("a1") = dst >> 32;     // 11
            register uint32_t reg_src_low asm("a2") = src >> 0;       // 12
            register uint32_t reg_src_high asm("a3") = src >> 32;     // 13
            register uint32_t reg_size asm("a4") = size;              // 14
            register uint32_t reg_dst_stride asm("a5") = dst_stride;  // 15
            register uint32_t reg_src_stride asm("a6") = src_stride;  // 16
            register uint32_t reg_repeat asm("a7") = repeat;          // 17

            // dmsrc a0, a1
            asm volatile(
                ".word (0b0000000 << 25) |                         (     (13) << 20) |                         (     (12) << 15) |                         (    0b000 << 12) |                         (0b0101011 <<  0)

" ::"r"(reg_src_high), "r"(reg_src_low));

            // dmdst a0, a1
            asm volatile(
                ".word (0b0000001 << 25) |                         (     (11) << 20) |                         (     (10) << 15) |                         (    0b000 << 12) |                         (0b0101011 <<  0)

" ::"r"(reg_dst_high), "r"(reg_dst_low));

            // dmstr a5, a6
            asm volatile(
                ".word (0b0000110 << 25) |                         (     (15) << 20) |                         (     (16) << 15) |                         (    0b000 << 12) |                         (0b0101011 <<  0)

" : : "r"(reg_dst_stride), "r"(reg_src_stride));

            // dmrep a7
            asm volatile(
                ".word (0b0000111 << 25) |                         (     (17) << 15) |                         (    0b000 << 12) |                         (0b0101011 <<  0)

" : : "r"(reg_repeat));

            // dmcpyi a0, a4, 0b10
            register uint32_t reg_txid asm("a0");  // 10
            asm volatile(
                ".word (0b0000010 << 25) |                         (  0b00010 << 20) |                         (     (14) << 15) |                         (    0b000 << 12) |                         (     (10) <<  7) |                         (0b0101011 <<  0)

" : "=r"(reg_txid) : "r"(reg_size));

            return reg_txid;
        } else {
            return -1;
        }
    }
Source code in xdsl/transforms/inline_snrt.py 
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.DmaStart2DWideptrOp, rewriter: PatternRewriter, /
):
    """
    Lowers to the equivalent of the snitch_runtime implementation:

    inline snrt_dma_txid_t snrt_dma_start_2d_wideptr(uint64_t dst, uint64_t src,
                                                     size_t size, size_t dst_stride,
                                                     size_t src_stride,
                                                     size_t repeat) {
        // Current DMA does not allow transfers with size == 0 (blocks)
        // TODO(colluca) remove this check once new DMA is integrated
        if (size > 0) {
            register uint32_t reg_dst_low asm("a0") = dst >> 0;       // 10
            register uint32_t reg_dst_high asm("a1") = dst >> 32;     // 11
            register uint32_t reg_src_low asm("a2") = src >> 0;       // 12
            register uint32_t reg_src_high asm("a3") = src >> 32;     // 13
            register uint32_t reg_size asm("a4") = size;              // 14
            register uint32_t reg_dst_stride asm("a5") = dst_stride;  // 15
            register uint32_t reg_src_stride asm("a6") = src_stride;  // 16
            register uint32_t reg_repeat asm("a7") = repeat;          // 17

            // dmsrc a0, a1
            asm volatile(
                ".word (0b0000000 << 25) | \
                    (     (13) << 20) | \
                    (     (12) << 15) | \
                    (    0b000 << 12) | \
                    (0b0101011 <<  0)   \n" ::"r"(reg_src_high),
                "r"(reg_src_low));

            // dmdst a0, a1
            asm volatile(
                ".word (0b0000001 << 25) | \
                    (     (11) << 20) | \
                    (     (10) << 15) | \
                    (    0b000 << 12) | \
                    (0b0101011 <<  0)   \n" ::"r"(reg_dst_high),
                "r"(reg_dst_low));

            // dmstr a5, a6
            asm volatile(
                ".word (0b0000110 << 25) | \
                    (     (15) << 20) | \
                    (     (16) << 15) | \
                    (    0b000 << 12) | \
                    (0b0101011 <<  0)   \n"
                :
                : "r"(reg_dst_stride), "r"(reg_src_stride));

            // dmrep a7
            asm volatile(
                ".word (0b0000111 << 25) | \
                    (     (17) << 15) | \
                    (    0b000 << 12) | \
                    (0b0101011 <<  0)   \n"
                :
                : "r"(reg_repeat));

            // dmcpyi a0, a4, 0b10
            register uint32_t reg_txid asm("a0");  // 10
            asm volatile(
                ".word (0b0000010 << 25) | \
                    (  0b00010 << 20) | \
                    (     (14) << 15) | \
                    (    0b000 << 12) | \
                    (     (10) <<  7) | \
                    (0b0101011 <<  0)   \n"
                : "=r"(reg_txid)
                : "r"(reg_size));

            return reg_txid;
        } else {
            return -1;
        }
    }
    """
    rewriter.insert_op(
        [
            dst := self.cast_i64(op.dst),
            src := self.cast_i64(op.src),
            src_stride := self.cast_i32(op.src_stride),
            dst_stride := self.cast_i32(op.dst_stride),
            size := self.cast_i32(op.size),
            repeat := self.cast_i32(op.size),
        ]
    )
    rewriter.replace_op(
        op,
        *self.generate_dma_instructions(
            dst.results[0],
            dst.results[1],
            src.results[0],
            src.results[1],
            size,
            dst_stride,
            src_stride,
            repeat,
        ),
    )

LowerDMAStart2D

Bases: LowerDMAStart2DBase

Source code in xdsl/transforms/inline_snrt.py 
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
class LowerDMAStart2D(LowerDMAStart2DBase):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.DmaStart2DOp, rewriter: PatternRewriter, /
    ):
        """
        Lower to the equivalent snitch_runtime implementation:

        /// Initiate an asynchronous 2D DMA transfer.
        inline snrt_dma_txid_t snrt_dma_start_2d(void *dst, const void *src,
                                                 size_t size, size_t dst_stride,
                                                 size_t src_stride, size_t repeat) {
            return snrt_dma_start_2d_wideptr((size_t)dst, (size_t)src, size, dst_stride,
                                             src_stride, repeat);
        }
        """
        rewriter.insert_op(
            [
                # we use zero register for the ptr_high registers
                zero := riscv.GetRegisterOp(riscv.Registers.ZERO),
                dst := self.cast_i32(op.dst),
                src := self.cast_i32(op.src),
                src_stride := self.cast_i32(op.src_stride),
                dst_stride := self.cast_i32(op.dst_stride),
                size := self.cast_i32(op.size),
                repeat := self.cast_i32(op.size),
            ]
        )
        # generate the dma setup instructions with `zero` for the ptr_high values
        rewriter.replace_op(
            op,
            *self.generate_dma_instructions(
                dst,
                zero,
                src,
                zero,
                size,
                dst_stride,
                src_stride,
                repeat,
            ),
        )

match_and_rewrite(op: snitch_runtime.DmaStart2DOp, rewriter: PatternRewriter)

Lower to the equivalent snitch_runtime implementation:

/// Initiate an asynchronous 2D DMA transfer. inline snrt_dma_txid_t snrt_dma_start_2d(void dst, const void src, size_t size, size_t dst_stride, size_t src_stride, size_t repeat) { return snrt_dma_start_2d_wideptr((size_t)dst, (size_t)src, size, dst_stride, src_stride, repeat); }

Source code in xdsl/transforms/inline_snrt.py 
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.DmaStart2DOp, rewriter: PatternRewriter, /
):
    """
    Lower to the equivalent snitch_runtime implementation:

    /// Initiate an asynchronous 2D DMA transfer.
    inline snrt_dma_txid_t snrt_dma_start_2d(void *dst, const void *src,
                                             size_t size, size_t dst_stride,
                                             size_t src_stride, size_t repeat) {
        return snrt_dma_start_2d_wideptr((size_t)dst, (size_t)src, size, dst_stride,
                                         src_stride, repeat);
    }
    """
    rewriter.insert_op(
        [
            # we use zero register for the ptr_high registers
            zero := riscv.GetRegisterOp(riscv.Registers.ZERO),
            dst := self.cast_i32(op.dst),
            src := self.cast_i32(op.src),
            src_stride := self.cast_i32(op.src_stride),
            dst_stride := self.cast_i32(op.dst_stride),
            size := self.cast_i32(op.size),
            repeat := self.cast_i32(op.size),
        ]
    )
    # generate the dma setup instructions with `zero` for the ptr_high values
    rewriter.replace_op(
        op,
        *self.generate_dma_instructions(
            dst,
            zero,
            src,
            zero,
            size,
            dst_stride,
            src_stride,
            repeat,
        ),
    )

LowerGlobalCoreBaseHartid dataclass

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
@dataclass
class LowerGlobalCoreBaseHartid(RewritePattern):
    constants: SnrtConstants

    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.GlobalCoreBaseHartidOp, rewriter: PatternRewriter, /
    ):
        rewriter.replace_op(
            op,
            [
                arith.ConstantOp.from_int_and_width(
                    self.constants.base_hartid, builtin.i32
                )
            ],
        )

constants: SnrtConstants instance-attribute

__init__(constants: SnrtConstants) -> None

match_and_rewrite(op: snitch_runtime.GlobalCoreBaseHartidOp, rewriter: PatternRewriter)

Source code in xdsl/transforms/inline_snrt.py 
437
438
439
440
441
442
443
444
445
446
447
448
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.GlobalCoreBaseHartidOp, rewriter: PatternRewriter, /
):
    rewriter.replace_op(
        op,
        [
            arith.ConstantOp.from_int_and_width(
                self.constants.base_hartid, builtin.i32
            )
        ],
    )

LowerGlobalCoreNum dataclass

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
@dataclass
class LowerGlobalCoreNum(RewritePattern):
    constants: SnrtConstants

    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.GlobalCoreNumOp, rewriter: PatternRewriter, /
    ):
        rewriter.replace_op(
            op,
            [
                arith.ConstantOp.from_int_and_width(
                    self.constants.cluster_num * self.constants.cluster_core_num,
                    builtin.i32,
                )
            ],
        )

constants: SnrtConstants instance-attribute

__init__(constants: SnrtConstants) -> None

match_and_rewrite(op: snitch_runtime.GlobalCoreNumOp, rewriter: PatternRewriter)

Source code in xdsl/transforms/inline_snrt.py 
455
456
457
458
459
460
461
462
463
464
465
466
467
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.GlobalCoreNumOp, rewriter: PatternRewriter, /
):
    rewriter.replace_op(
        op,
        [
            arith.ConstantOp.from_int_and_width(
                self.constants.cluster_num * self.constants.cluster_core_num,
                builtin.i32,
            )
        ],
    )

LowerClusterCoreNum dataclass

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
@dataclass
class LowerClusterCoreNum(RewritePattern):
    constants: SnrtConstants

    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.ClusterCoreNumOp, rewriter: PatternRewriter, /
    ):
        rewriter.replace_op(
            op,
            [
                arith.ConstantOp.from_int_and_width(
                    self.constants.cluster_core_num, builtin.i32
                )
            ],
        )

constants: SnrtConstants instance-attribute

__init__(constants: SnrtConstants) -> None

match_and_rewrite(op: snitch_runtime.ClusterCoreNumOp, rewriter: PatternRewriter)

Source code in xdsl/transforms/inline_snrt.py 
474
475
476
477
478
479
480
481
482
483
484
485
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.ClusterCoreNumOp, rewriter: PatternRewriter, /
):
    rewriter.replace_op(
        op,
        [
            arith.ConstantOp.from_int_and_width(
                self.constants.cluster_core_num, builtin.i32
            )
        ],
    )

LowerClusterNum dataclass

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
@dataclass
class LowerClusterNum(RewritePattern):
    constants: SnrtConstants

    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.ClusterNumOp, rewriter: PatternRewriter, /
    ):
        rewriter.replace_op(
            op,
            [
                arith.ConstantOp.from_int_and_width(
                    self.constants.cluster_num, builtin.i32
                )
            ],
        )

constants: SnrtConstants instance-attribute

__init__(constants: SnrtConstants) -> None

match_and_rewrite(op: snitch_runtime.ClusterNumOp, rewriter: PatternRewriter)

Source code in xdsl/transforms/inline_snrt.py 
492
493
494
495
496
497
498
499
500
501
502
503
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.ClusterNumOp, rewriter: PatternRewriter, /
):
    rewriter.replace_op(
        op,
        [
            arith.ConstantOp.from_int_and_width(
                self.constants.cluster_num, builtin.i32
            )
        ],
    )

LowerClusterDmCoreNum dataclass

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
@dataclass
class LowerClusterDmCoreNum(RewritePattern):
    constants: SnrtConstants

    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.ClusterDmCoreNumOp, rewriter: PatternRewriter, /
    ):
        rewriter.replace_op(
            op,
            [
                arith.ConstantOp.from_int_and_width(
                    self.constants.cluster_dm_core_num, builtin.i32
                )
            ],
        )

constants: SnrtConstants instance-attribute

__init__(constants: SnrtConstants) -> None

match_and_rewrite(op: snitch_runtime.ClusterDmCoreNumOp, rewriter: PatternRewriter)

Source code in xdsl/transforms/inline_snrt.py 
510
511
512
513
514
515
516
517
518
519
520
521
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.ClusterDmCoreNumOp, rewriter: PatternRewriter, /
):
    rewriter.replace_op(
        op,
        [
            arith.ConstantOp.from_int_and_width(
                self.constants.cluster_dm_core_num, builtin.i32
            )
        ],
    )

LowerIsComputeCore

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
class LowerIsComputeCore(RewritePattern):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.IsComputeCoreOp, rewriter: PatternRewriter, /
    ):
        """
        inline int __attribute__((const)) snrt_is_compute_core() {
            return snrt_cluster_core_idx() < snrt_cluster_compute_core_num();
        }
        """
        rewriter.replace_op(
            op,
            [
                cluster_core_idx := snitch_runtime.ClusterCoreIdxOp(),
                compute_core_num := snitch_runtime.ClusterComputeCoreNumOp(),
                arith.CmpiOp(cluster_core_idx, compute_core_num, "slt"),
            ],
        )

match_and_rewrite(op: snitch_runtime.IsComputeCoreOp, rewriter: PatternRewriter)

inline int attribute((const)) snrt_is_compute_core() { return snrt_cluster_core_idx() < snrt_cluster_compute_core_num(); }

Source code in xdsl/transforms/inline_snrt.py 
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.IsComputeCoreOp, rewriter: PatternRewriter, /
):
    """
    inline int __attribute__((const)) snrt_is_compute_core() {
        return snrt_cluster_core_idx() < snrt_cluster_compute_core_num();
    }
    """
    rewriter.replace_op(
        op,
        [
            cluster_core_idx := snitch_runtime.ClusterCoreIdxOp(),
            compute_core_num := snitch_runtime.ClusterComputeCoreNumOp(),
            arith.CmpiOp(cluster_core_idx, compute_core_num, "slt"),
        ],
    )

LowerIsDmCore

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
class LowerIsDmCore(RewritePattern):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.IsDmCoreOp, rewriter: PatternRewriter, /
    ):
        """
        inline int __attribute__((const)) snrt_is_compute_core() {
            return snrt_cluster_core_idx() < snrt_cluster_compute_core_num();
        }

        inline int __attribute__((const)) snrt_is_dm_core() {
            return !snrt_is_compute_core();
        }
        """
        rewriter.replace_op(
            op,
            [
                cluster_core_idx := snitch_runtime.ClusterCoreIdxOp(),
                compute_core_num := snitch_runtime.ClusterComputeCoreNumOp(),
                arith.CmpiOp(cluster_core_idx, compute_core_num, "sge"),
            ],
        )

match_and_rewrite(op: snitch_runtime.IsDmCoreOp, rewriter: PatternRewriter)

inline int attribute((const)) snrt_is_compute_core() { return snrt_cluster_core_idx() < snrt_cluster_compute_core_num(); }

inline int attribute((const)) snrt_is_dm_core() { return !snrt_is_compute_core(); }

Source code in xdsl/transforms/inline_snrt.py 
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.IsDmCoreOp, rewriter: PatternRewriter, /
):
    """
    inline int __attribute__((const)) snrt_is_compute_core() {
        return snrt_cluster_core_idx() < snrt_cluster_compute_core_num();
    }

    inline int __attribute__((const)) snrt_is_dm_core() {
        return !snrt_is_compute_core();
    }
    """
    rewriter.replace_op(
        op,
        [
            cluster_core_idx := snitch_runtime.ClusterCoreIdxOp(),
            compute_core_num := snitch_runtime.ClusterComputeCoreNumOp(),
            arith.CmpiOp(cluster_core_idx, compute_core_num, "sge"),
        ],
    )

LowerClusterCoreIdx

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
class LowerClusterCoreIdx(RewritePattern):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.ClusterCoreIdxOp, rewriter: PatternRewriter, /
    ):
        """
        inline uint32_t __attribute__((const)) snrt_cluster_core_idx() {
            return snrt_global_core_idx() % snrt_cluster_core_num();
        }
        """
        rewriter.replace_op(
            op,
            [
                global_core_idx := snitch_runtime.GlobalCoreIdxOp(),
                cluster_core_num := snitch_runtime.ClusterCoreNumOp(),
                arith.RemSIOp(global_core_idx, cluster_core_num),
            ],
        )

match_and_rewrite(op: snitch_runtime.ClusterCoreIdxOp, rewriter: PatternRewriter)

inline uint32_t attribute((const)) snrt_cluster_core_idx() { return snrt_global_core_idx() % snrt_cluster_core_num(); }

Source code in xdsl/transforms/inline_snrt.py 
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.ClusterCoreIdxOp, rewriter: PatternRewriter, /
):
    """
    inline uint32_t __attribute__((const)) snrt_cluster_core_idx() {
        return snrt_global_core_idx() % snrt_cluster_core_num();
    }
    """
    rewriter.replace_op(
        op,
        [
            global_core_idx := snitch_runtime.GlobalCoreIdxOp(),
            cluster_core_num := snitch_runtime.ClusterCoreNumOp(),
            arith.RemSIOp(global_core_idx, cluster_core_num),
        ],
    )

LowerClusterComputeCoreNum dataclass

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
@dataclass
class LowerClusterComputeCoreNum(RewritePattern):
    constants: SnrtConstants

    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.ClusterComputeCoreNumOp, rewriter: PatternRewriter, /
    ):
        """
        Implementation:

        inline uint32_t __attribute__((const)) snrt_cluster_compute_core_num() {
            return snrt_cluster_core_num() - snrt_cluster_dm_core_num();
        }

        inline uint32_t __attribute__((const)) snrt_cluster_core_num() {
            return SNRT_CLUSTER_CORE_NUM;
        }

        inline uint32_t __attribute__((const)) snrt_cluster_dm_core_num() {
            return SNRT_CLUSTER_DM_CORE_NUM;
        }
        """
        rewriter.replace_op(
            op,
            [
                arith.ConstantOp.from_int_and_width(
                    self.constants.cluster_core_num
                    - self.constants.cluster_dm_core_num,
                    builtin.i32,
                )
            ],
        )

constants: SnrtConstants instance-attribute

__init__(constants: SnrtConstants) -> None

match_and_rewrite(op: snitch_runtime.ClusterComputeCoreNumOp, rewriter: PatternRewriter)

Implementation:

inline uint32_t attribute((const)) snrt_cluster_compute_core_num() { return snrt_cluster_core_num() - snrt_cluster_dm_core_num(); }

inline uint32_t attribute((const)) snrt_cluster_core_num() { return SNRT_CLUSTER_CORE_NUM; }

inline uint32_t attribute((const)) snrt_cluster_dm_core_num() { return SNRT_CLUSTER_DM_CORE_NUM; }

Source code in xdsl/transforms/inline_snrt.py 
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.ClusterComputeCoreNumOp, rewriter: PatternRewriter, /
):
    """
    Implementation:

    inline uint32_t __attribute__((const)) snrt_cluster_compute_core_num() {
        return snrt_cluster_core_num() - snrt_cluster_dm_core_num();
    }

    inline uint32_t __attribute__((const)) snrt_cluster_core_num() {
        return SNRT_CLUSTER_CORE_NUM;
    }

    inline uint32_t __attribute__((const)) snrt_cluster_dm_core_num() {
        return SNRT_CLUSTER_DM_CORE_NUM;
    }
    """
    rewriter.replace_op(
        op,
        [
            arith.ConstantOp.from_int_and_width(
                self.constants.cluster_core_num
                - self.constants.cluster_dm_core_num,
                builtin.i32,
            )
        ],
    )

LowerGlobalCoreIdx dataclass

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
@dataclass
class LowerGlobalCoreIdx(RewritePattern):
    constants: SnrtConstants

    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.GlobalCoreIdxOp, rewriter: PatternRewriter, /
    ):
        """
        Implementation:

        inline uint32_t __attribute__((const)) snrt_hartid() {
            uint32_t hartid;
            asm("csrr %0, mhartid" : "=r"(hartid));
            return hartid;
        }

        inline uint32_t __attribute__((const)) snrt_global_core_idx() {
            return snrt_hartid() - snrt_global_core_base_hartid();
        }
        """
        rewriter.replace_op(
            op,
            [
                zero := riscv.GetRegisterOp(riscv.Registers.ZERO),
                hartid := riscv.CsrrsOp(zero, IntegerAttr(0xF14, 12), readonly=True),
                hartid_i32 := builtin.UnrealizedConversionCastOp.get(
                    [hartid], [builtin.i32]
                ),
                base_hartid := arith.ConstantOp.from_int_and_width(
                    self.constants.base_hartid, builtin.i32
                ),
                arith.SubiOp(hartid_i32, base_hartid),
            ],
        )

constants: SnrtConstants instance-attribute

__init__(constants: SnrtConstants) -> None

match_and_rewrite(op: snitch_runtime.GlobalCoreIdxOp, rewriter: PatternRewriter)

Implementation:

inline uint32_t attribute((const)) snrt_hartid() { uint32_t hartid; asm("csrr %0, mhartid" : "=r"(hartid)); return hartid; }

inline uint32_t attribute((const)) snrt_global_core_idx() { return snrt_hartid() - snrt_global_core_base_hartid(); }

Source code in xdsl/transforms/inline_snrt.py 
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.GlobalCoreIdxOp, rewriter: PatternRewriter, /
):
    """
    Implementation:

    inline uint32_t __attribute__((const)) snrt_hartid() {
        uint32_t hartid;
        asm("csrr %0, mhartid" : "=r"(hartid));
        return hartid;
    }

    inline uint32_t __attribute__((const)) snrt_global_core_idx() {
        return snrt_hartid() - snrt_global_core_base_hartid();
    }
    """
    rewriter.replace_op(
        op,
        [
            zero := riscv.GetRegisterOp(riscv.Registers.ZERO),
            hartid := riscv.CsrrsOp(zero, IntegerAttr(0xF14, 12), readonly=True),
            hartid_i32 := builtin.UnrealizedConversionCastOp.get(
                [hartid], [builtin.i32]
            ),
            base_hartid := arith.ConstantOp.from_int_and_width(
                self.constants.base_hartid, builtin.i32
            ),
            arith.SubiOp(hartid_i32, base_hartid),
        ],
    )

LowerClusterIdx

Bases: RewritePattern

Source code in xdsl/transforms/inline_snrt.py 
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
class LowerClusterIdx(RewritePattern):
    @op_type_rewrite_pattern
    def match_and_rewrite(
        self, op: snitch_runtime.ClusterIdxOp, rewriter: PatternRewriter, /
    ):
        """
        Implementation:

        inline uint32_t __attribute__((const)) snrt_cluster_idx() {
            return snrt_global_core_idx() / snrt_cluster_core_num();
        }
        """
        rewriter.replace_op(
            op,
            [
                cluster_core_num := snitch_runtime.ClusterCoreNumOp(),
                core_idx := snitch_runtime.GlobalCoreIdxOp(),
                arith.DivSIOp(
                    core_idx,
                    cluster_core_num,
                ),
            ],
        )

match_and_rewrite(op: snitch_runtime.ClusterIdxOp, rewriter: PatternRewriter)

Implementation:

inline uint32_t attribute((const)) snrt_cluster_idx() { return snrt_global_core_idx() / snrt_cluster_core_num(); }

Source code in xdsl/transforms/inline_snrt.py 
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
@op_type_rewrite_pattern
def match_and_rewrite(
    self, op: snitch_runtime.ClusterIdxOp, rewriter: PatternRewriter, /
):
    """
    Implementation:

    inline uint32_t __attribute__((const)) snrt_cluster_idx() {
        return snrt_global_core_idx() / snrt_cluster_core_num();
    }
    """
    rewriter.replace_op(
        op,
        [
            cluster_core_num := snitch_runtime.ClusterCoreNumOp(),
            core_idx := snitch_runtime.GlobalCoreIdxOp(),
            arith.DivSIOp(
                core_idx,
                cluster_core_num,
            ),
        ],
    )

InlineSnrtPass dataclass

Bases: SnrtConstants, ModulePass

Inline operations of the snrt dialect to their definitions.

Uses arith operations where possible, and emit riscv.csr* and riscv_snitch ops where required.

Source code in xdsl/transforms/inline_snrt.py 
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
@dataclass(frozen=True)
class InlineSnrtPass(SnrtConstants, ModulePass):
    """
    Inline operations of the snrt dialect to their definitions.

    Uses arith operations where possible, and emit `riscv.csr*` and `riscv_snitch` ops where required.
    """

    name = "inline-snrt"

    def apply(self, ctx: Context, op: builtin.ModuleOp) -> None:
        PatternRewriteWalker(
            GreedyRewritePatternApplier(
                [
                    LowerClusterHWBarrier(),
                    LowerSSRDisable(),
                    LowerDMAStart1D(),
                    LowerDMAStart1DWidePtr(),
                    LowerDMAStart2D(),
                    LowerDMAStart2DWideptr(),
                    # information getting ops:
                    LowerClusterIdx(),
                    LowerClusterNum(self),
                    LowerClusterCoreIdx(),
                    LowerClusterCoreNum(self),
                    LowerClusterDmCoreNum(self),
                    LowerClusterComputeCoreNum(self),
                    LowerGlobalCoreNum(self),
                    LowerGlobalCoreIdx(self),
                    LowerGlobalCoreBaseHartid(self),
                    LowerIsComputeCore(),
                    LowerIsDmCore(),
                ]
            )
        ).rewrite_module(op)

name = 'inline-snrt' class-attribute instance-attribute

__init__(cluster_num: int = 4, cluster_core_num: int = 9, base_hartid: int = 0, cluster_dm_core_num: int = 1) -> None

apply(ctx: Context, op: builtin.ModuleOp) -> None

Source code in xdsl/transforms/inline_snrt.py 
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
def apply(self, ctx: Context, op: builtin.ModuleOp) -> None:
    PatternRewriteWalker(
        GreedyRewritePatternApplier(
            [
                LowerClusterHWBarrier(),
                LowerSSRDisable(),
                LowerDMAStart1D(),
                LowerDMAStart1DWidePtr(),
                LowerDMAStart2D(),
                LowerDMAStart2DWideptr(),
                # information getting ops:
                LowerClusterIdx(),
                LowerClusterNum(self),
                LowerClusterCoreIdx(),
                LowerClusterCoreNum(self),
                LowerClusterDmCoreNum(self),
                LowerClusterComputeCoreNum(self),
                LowerGlobalCoreNum(self),
                LowerGlobalCoreIdx(self),
                LowerGlobalCoreBaseHartid(self),
                LowerIsComputeCore(),
                LowerIsDmCore(),
            ]
        )
    ).rewrite_module(op)