#define GEN7_3DPRIM_START_INSTANCE 0x243C
#define GEN7_3DPRIM_BASE_VERTEX 0x2440
-/* MI_MATH only exists on Haswell+ */
-#if GEN_IS_HASWELL || GEN_GEN >= 8
-
-/* Emit dwords to multiply GPR0 by N */
-static void
-build_alu_multiply_gpr0(uint32_t *dw, unsigned *dw_count, uint32_t N)
-{
- VK_OUTARRAY_MAKE(out, dw, dw_count);
-
-#define append_alu(opcode, operand1, operand2) \
- vk_outarray_append(&out, alu_dw) *alu_dw = mi_alu(opcode, operand1, operand2)
-
- assert(N > 0);
- unsigned top_bit = 31 - __builtin_clz(N);
- for (int i = top_bit - 1; i >= 0; i--) {
- /* We get our initial data in GPR0 and we write the final data out to
- * GPR0 but we use GPR1 as our scratch register.
- */
- unsigned src_reg = i == top_bit - 1 ? MI_ALU_REG0 : MI_ALU_REG1;
- unsigned dst_reg = i == 0 ? MI_ALU_REG0 : MI_ALU_REG1;
-
- /* Shift the current value left by 1 */
- append_alu(MI_ALU_LOAD, MI_ALU_SRCA, src_reg);
- append_alu(MI_ALU_LOAD, MI_ALU_SRCB, src_reg);
- append_alu(MI_ALU_ADD, 0, 0);
-
- if (N & (1 << i)) {
- /* Store ACCU to R1 and add R0 to R1 */
- append_alu(MI_ALU_STORE, MI_ALU_REG1, MI_ALU_ACCU);
- append_alu(MI_ALU_LOAD, MI_ALU_SRCA, MI_ALU_REG0);
- append_alu(MI_ALU_LOAD, MI_ALU_SRCB, MI_ALU_REG1);
- append_alu(MI_ALU_ADD, 0, 0);
- }
-
- append_alu(MI_ALU_STORE, dst_reg, MI_ALU_ACCU);
- }
-
-#undef append_alu
-}
-
-static void
-emit_mul_gpr0(struct anv_batch *batch, uint32_t N)
-{
- uint32_t num_dwords;
- build_alu_multiply_gpr0(NULL, &num_dwords, N);
-
- uint32_t *dw = anv_batch_emitn(batch, 1 + num_dwords, GENX(MI_MATH));
- build_alu_multiply_gpr0(dw + 1, &num_dwords, N);
-}
-
-#endif /* GEN_IS_HASWELL || GEN_GEN >= 8 */
-
void genX(CmdDrawIndirectByteCountEXT)(
VkCommandBuffer commandBuffer,
uint32_t instanceCount,
struct anv_address addr,
bool indexed)
{
- struct anv_batch *batch = &cmd_buffer->batch;
+ struct gen_mi_builder b;
+ gen_mi_builder_init(&b, &cmd_buffer->batch);
- emit_lrm(batch, GEN7_3DPRIM_VERTEX_COUNT, anv_address_add(addr, 0));
+ gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_VERTEX_COUNT),
+ gen_mi_mem32(anv_address_add(addr, 0)));
+ struct gen_mi_value instance_count = gen_mi_mem32(anv_address_add(addr, 4));
unsigned view_count = anv_subpass_view_count(cmd_buffer->state.subpass);
if (view_count > 1) {
#if GEN_IS_HASWELL || GEN_GEN >= 8
- emit_lrm(batch, CS_GPR(0), anv_address_add(addr, 4));
- emit_mul_gpr0(batch, view_count);
- emit_lrr(batch, GEN7_3DPRIM_INSTANCE_COUNT, CS_GPR(0));
+ instance_count = gen_mi_imul_imm(&b, instance_count, view_count);
#else
anv_finishme("Multiview + indirect draw requires MI_MATH; "
"MI_MATH is not supported on Ivy Bridge");
- emit_lrm(batch, GEN7_3DPRIM_INSTANCE_COUNT, anv_address_add(addr, 4));
#endif
- } else {
- emit_lrm(batch, GEN7_3DPRIM_INSTANCE_COUNT, anv_address_add(addr, 4));
}
+ gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_INSTANCE_COUNT), instance_count);
- emit_lrm(batch, GEN7_3DPRIM_START_VERTEX, anv_address_add(addr, 8));
+ gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_START_VERTEX),
+ gen_mi_mem32(anv_address_add(addr, 8)));
if (indexed) {
- emit_lrm(batch, GEN7_3DPRIM_BASE_VERTEX, anv_address_add(addr, 12));
- emit_lrm(batch, GEN7_3DPRIM_START_INSTANCE, anv_address_add(addr, 16));
+ gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_BASE_VERTEX),
+ gen_mi_mem32(anv_address_add(addr, 12)));
+ gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_START_INSTANCE),
+ gen_mi_mem32(anv_address_add(addr, 16)));
} else {
- emit_lrm(batch, GEN7_3DPRIM_START_INSTANCE, anv_address_add(addr, 12));
- emit_lri(batch, GEN7_3DPRIM_BASE_VERTEX, 0);
+ gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_START_INSTANCE),
+ gen_mi_mem32(anv_address_add(addr, 12)));
+ gen_mi_store(&b, gen_mi_reg32(GEN7_3DPRIM_BASE_VERTEX), gen_mi_imm(0));
}
}