freedreno,computerator: support initialization of buffers

The following syntax can now be used to set the initial content of
buffers:

@buf size (reg) val0, val1, ...

If the buffer is not fully initialized, remaining values will be set to
zero.

Signed-off-by: Job Noorman <jnoorman@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28625>

src/freedreno/computerator/examples/buf-init.asm

@@ -0,0 +1,9 @@
+@localsize 8, 1, 1
+@buf 8 1, 2, 3, 4, 5, 6, 7, 8
+@invocationid(r0.x)
+
+ldib.b.untyped.1d.u32.4.imm r1.x, r0.x, 0
+(sy)add.u r1.x, r1.x, 5
+(rpt5)nop
+stib.b.untyped.1d.u32.4.imm r1.x, r0.x, 0
+end

src/freedreno/computerator/ir3_asm.cc

@@ -46,6 +46,8 @@ ir3_asm_assemble(struct ir3_compiler *c, FILE *in)
           sizeof(kernel->base.buf_sizes));
    memcpy(kernel->base.buf_addr_regs, kernel->info.buf_addr_regs,
           sizeof(kernel->base.buf_addr_regs));
+   memcpy(kernel->base.buf_init_data, kernel->info.buf_init_data,
+          sizeof(kernel->base.buf_init_data));
 
    unsigned sz = v->info.size;
 

src/freedreno/computerator/main.cc

@@ -269,6 +269,13 @@ main(int argc, char **argv)
    for (int i = 0; i < kernel->num_bufs; i++) {
       printf("buf[%d]: size=%u\n", i, kernel->buf_sizes[i]);
       kernel->bufs[i] = fd_bo_new(dev, kernel->buf_sizes[i] * 4, 0, "buf[%d]", i);
+
+      if (kernel->buf_init_data[i]) {
+         fd_bo_cpu_prep(kernel->bufs[i], pipe, FD_BO_PREP_WRITE);
+         void *map = fd_bo_map(kernel->bufs[i]);
+         memcpy(map, kernel->buf_init_data[i], kernel->buf_sizes[i] * 4);
+         free(kernel->buf_init_data[i]);
+      }
    }
 
    if (disasm)

src/freedreno/computerator/main.h

@@ -42,6 +42,7 @@ struct kernel {
    uint32_t num_bufs;
    uint32_t buf_sizes[MAX_BUFS]; /* size in dwords */
    uint32_t buf_addr_regs[MAX_BUFS];
+   uint32_t *buf_init_data[MAX_BUFS];
 
    /* filled in by frontend before launching grid: */
    struct fd_bo *bufs[MAX_BUFS];

src/freedreno/ir3/ir3_assembler.h

@@ -37,6 +37,8 @@ struct ir3_kernel_info {
    uint32_t num_bufs;
    uint32_t buf_sizes[MAX_BUFS]; /* size in dwords */
    uint32_t buf_addr_regs[MAX_BUFS];
+   uint32_t *buf_init_data[MAX_BUFS];
+   uint32_t buf_init_data_sizes[MAX_BUFS];
 
    uint64_t shader_print_buffer_iova;
 

src/freedreno/ir3/ir3_parser.y

@@ -236,6 +236,21 @@ static void add_const(unsigned reg, unsigned c0, unsigned c1, unsigned c2, unsig
 	const_state->immediates[idx * 4 + 3] = c3;
 }
 
+static void add_buf_init_val(uint32_t val)
+{
+	assert(info->num_bufs > 0);
+	unsigned idx = info->num_bufs - 1;
+
+	if (!info->buf_init_data[idx]) {
+		unsigned sz = info->buf_sizes[idx] * 4;
+		info->buf_init_data[idx] = malloc(sz);
+		memset(info->buf_init_data[idx], 0, sz);
+	}
+
+	assert(info->buf_init_data_sizes[idx] < info->buf_sizes[idx]);
+	info->buf_init_data[idx][info->buf_init_data_sizes[idx]++] = val;
+}
+
 static void add_sysval(unsigned reg, unsigned compmask, gl_system_value sysval)
 {
 	unsigned n = variant->inputs_count++;
@@ -752,6 +767,11 @@ const_header:      T_A_CONST '(' T_CONSTANT ')' const_val ',' const_val ',' cons
                        add_const($3, $5, $7, $9, $11);
 }
 
+buf_header_init_val:  const_val { add_buf_init_val($1); }
+buf_header_init_vals: buf_header_init_val
+|                     buf_header_init_val ',' buf_header_init_vals
+|
+
 buf_header_addr_reg:
                    '(' T_CONSTANT ')' {
                        assert(($2 & 0x1) == 0);  /* half-reg not allowed */
@@ -767,7 +787,7 @@ buf_header:        T_A_BUF const_val {
                        int idx = info->num_bufs++;
                        assert(idx < MAX_BUFS);
                        info->buf_sizes[idx] = $2;
-} buf_header_addr_reg
+} buf_header_addr_reg buf_header_init_vals
 
 invocationid_header: T_A_INVOCATIONID '(' T_REGISTER ')' {
                        assert(($3 & 0x1) == 0);  /* half-reg not allowed */