fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-09 06:48:06 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

r600g: implement output modifiers and use them to further optimize LRP

Browse source
This commit is contained in:
Christian König 2010-12-16 16:42:14 +01:00
parent f75578b31e
commit 08c8cd26b8
4 changed files with 32 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
src/gallium/drivers/r600/r600_asm.c
View file

@ -707,6 +707,7 @@ static int r600_bc_alu_build(struct r600_bc *bc, struct r600_bc_alu *alu, unsign
S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu->src[0].abs) | S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu->src[0].abs) |
S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu->src[1].abs) | S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu->src[1].abs) |
S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu->dst.write) | S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu->dst.write) |
S_SQ_ALU_WORD1_OP2_OMOD(alu->omod) |
S_SQ_ALU_WORD1_OP2_ALU_INST(alu->inst) | S_SQ_ALU_WORD1_OP2_ALU_INST(alu->inst) |
S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) | S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) |
S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->predicate) | S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->predicate) |

1
src/gallium/drivers/r600/r600_asm.h
View file

@ -62,6 +62,7 @@ struct r600_bc_alu {
unsigned bank_swizzle_force; unsigned bank_swizzle_force;
u32 value[4]; u32 value[4];
int hw_gpr[NUM_OF_CYCLES][NUM_OF_COMPONENTS]; int hw_gpr[NUM_OF_CYCLES][NUM_OF_COMPONENTS];
unsigned omod;
}; };
struct r600_bc_tex { struct r600_bc_tex {

29
src/gallium/drivers/r600/r600_shader.c
View file

@ -1975,6 +1975,35 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
r = tgsi_split_literal_constant(ctx, r600_src); r = tgsi_split_literal_constant(ctx, r600_src);
if (r) if (r)
return r; return r;
/* optimize if it's just an equal balance */
if(r600_src[0].sel == V_SQ_ALU_SRC_0_5) {
for (i = 0; i < lasti + 1; i++) {
if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
continue;
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD);
alu.src[0] = r600_src[1];
alu.src[0].chan = tgsi_chan(&inst->Src[1], i);
alu.src[1] = r600_src[2];
alu.src[1].chan = tgsi_chan(&inst->Src[2], i);
alu.omod = 3;
r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
if (r)
return r;
alu.dst.chan = i;
if (i == lasti) {
alu.last = 1;
}
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
}
return 0;
}
/* 1 - src0 */ /* 1 - src0 */
for (i = 0; i < lasti + 1; i++) { for (i = 0; i < lasti + 1; i++) {
if (!(inst->Dst[0].Register.WriteMask & (1 << i))) if (!(inst->Dst[0].Register.WriteMask & (1 << i)))

1
src/gallium/drivers/r600/r700_asm.c
View file

@ -61,6 +61,7 @@ int r700_bc_alu_build(struct r600_bc *bc, struct r600_bc_alu *alu, unsigned id)
S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu->src[0].abs) | S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu->src[0].abs) |
S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu->src[1].abs) | S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu->src[1].abs) |
S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu->dst.write) | S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu->dst.write) |
S_SQ_ALU_WORD1_OP2_OMOD(alu->omod) |
S_SQ_ALU_WORD1_OP2_ALU_INST(alu->inst) | S_SQ_ALU_WORD1_OP2_ALU_INST(alu->inst) |
S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) | S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) |
S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->predicate) | S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->predicate) |