# please do not use this as evidence when alyssa is finally prosecuted for writing cursed things, thank you <3 def createCDM(self, output_buffer_addr): # 0: 7e0102088004 mov r0, 66 # 6: 4501080d00c01200 store r0, du4, 0, i32, x, a # e: 8800 stop shader = bytes.fromhex("7e0102088004 4501080d00c01200 8800") shader_addr = self.alloc(shader, True, "shader") # 0: 0501040d00c0f200 load r0_r1_r2_r3, du2, 0, i32, xyzw, a # 8: 3800 wait a # a: c500803d0080f000 mov du4, dr0 # 12: 8800 stop preshader = bytes.fromhex("0501040d00c0f200 3800 c500803d0080f000 8800") preshader_addr = self.alloc(preshader, True, "preshader") # uniform # two pointers: # first is mesa binding the uniform itself (for bindless?) # second is hopefully output buffer uniform_buf = ctx.gobj.new(Array(2, Int64ul), "uniform", track=False) uniform_buf_addr = uniform_buf._addr uniform_buf.val = [uniform_buf_addr, output_buffer_addr] uniform_buf.push() self.objects.append(uniform_buf) # USC (the 'pipeline'): usc = b"" # uniform: # address is from bit 26. shr by 2 (i.e., aligned on bit 24). # 1D 00 80 usc = usc + b"\x1d\x00\x80" # 00 00 84 F6 FF 77 usc = usc + struct.pack(">Q", uniform_buf_addr)[:6] # shared: # 4D 90 00 00 usc = usc + b"\x4d\x90\x00\x00" # shader: # 0D 0C # 00 C0 FA FF // pipeline address usc = usc + b"\x0d\x0c" usc = usc + struct.pack(">I", shader_addr) # registers: # 8D 01 00 01 usc = usc + b"\x8d\x01\x00\x01" # preshader: # 38 00 # 80 C0 # 80 40 FB FF // pipeline address usc = usc + b"\x38\x00\x80\xc0" usc = usc + struct.pack(">I", preshader_addr) pipeline_addr = self.alloc(usc, True, "pipeline/USC") # CDM (this is where the 'encoder' actually points): # compute (0x0 @ bit 29): # 12 10 00 00 # 00 80 F8 FF // pipeline address (start at 6) # 00 01 00 00 // global x # 01 00 00 00 // global y # 01 00 00 00 // global z # 01 00 00 00 // local x # 01 00 00 00 // local y # 01 00 00 00 // local z cdm = b"\x12\x10\x00\x00" cdm = cdm + struct.pack(">I", pipeline_addr) cdm = cdm + struct.pack(">I", 0x100) cdm = cdm + struct.pack(">I", 0x1)*5 # barrier (0x3 @ bit 29): # FF FF 0F 60 cdm = cdm + b"\xff\xff\x0f\x60" # terminate (0x2 @ bit 29): # 00 00 00 40 00 00 00 00 cdm = cdm + b"\x00\x00\x00\x40" cdm_addr = self.alloc(cdm, False, "CDM") return cdm_addr, len(cdm)