diff options
| author | Eric Anholt <[email protected]> | 2018-07-20 14:27:09 -0700 |
|---|---|---|
| committer | Eric Anholt <[email protected]> | 2018-07-23 10:21:43 -0700 |
| commit | 6b73a97f84f86f4c3d3bbbbadf963c20b8e52b57 (patch) | |
| tree | 6c8701d6c84c1478f7e829a08e5bb141777846e5 /src/broadcom/compiler/vir_dump.c | |
| parent | 79e0f042bcd6a1956015cf332c2232ade2c2321f (diff) | |
v3d: Implement a small immediates optimization, based on VC4's.
We can do one per instruction, and we have to be careful not to overwrite
raddr_b, but this greatly reduces the pressure on uniform loads
(particularly around ldvpm/stvpm instructions).
total instructions in shared programs: 90768 -> 88220 (-2.81%)
instructions in affected programs: 82711 -> 80163 (-3.08%)
Diffstat (limited to 'src/broadcom/compiler/vir_dump.c')
| -rw-r--r-- | src/broadcom/compiler/vir_dump.c | 27 |
1 files changed, 18 insertions, 9 deletions
diff --git a/src/broadcom/compiler/vir_dump.c b/src/broadcom/compiler/vir_dump.c index 88b5dc90ac7..c43578302b5 100644 --- a/src/broadcom/compiler/vir_dump.c +++ b/src/broadcom/compiler/vir_dump.c @@ -25,7 +25,8 @@ #include "v3d_compiler.h" static void -vir_print_reg(struct v3d_compile *c, struct qreg reg) +vir_print_reg(struct v3d_compile *c, const struct qinst *inst, + struct qreg reg) { static const char *files[] = { [QFILE_TEMP] = "t", @@ -58,12 +59,20 @@ vir_print_reg(struct v3d_compile *c, struct qreg reg) fprintf(stderr, "%s", v3d_qpu_magic_waddr_name(reg.index)); break; - case QFILE_SMALL_IMM: - if ((int)reg.index >= -16 && (int)reg.index <= 15) - fprintf(stderr, "%d", reg.index); + case QFILE_SMALL_IMM: { + uint32_t unpacked; + bool ok = v3d_qpu_small_imm_unpack(c->devinfo, + inst->qpu.raddr_b, + &unpacked); + assert(ok); (void) ok; + + if ((int)inst->qpu.raddr_b >= -16 && + (int)inst->qpu.raddr_b <= 15) + fprintf(stderr, "%d", unpacked); else - fprintf(stderr, "%f", uif(reg.index)); + fprintf(stderr, "%f", uif(unpacked)); break; + } case QFILE_VPM: fprintf(stderr, "vpm%d.%d", @@ -220,7 +229,7 @@ vir_dump_alu(struct v3d_compile *c, struct qinst *inst) fprintf(stderr, "%s", v3d_qpu_uf_name(instr->flags.auf)); fprintf(stderr, " "); - vir_print_reg(c, inst->dst); + vir_print_reg(c, inst, inst->dst); fprintf(stderr, "%s", v3d_qpu_pack_name(instr->alu.add.output_pack)); unpack[0] = instr->alu.add.a_unpack; @@ -232,7 +241,7 @@ vir_dump_alu(struct v3d_compile *c, struct qinst *inst) fprintf(stderr, "%s", v3d_qpu_uf_name(instr->flags.muf)); fprintf(stderr, " "); - vir_print_reg(c, inst->dst); + vir_print_reg(c, inst, inst->dst); fprintf(stderr, "%s", v3d_qpu_pack_name(instr->alu.mul.output_pack)); unpack[0] = instr->alu.mul.a_unpack; @@ -241,7 +250,7 @@ vir_dump_alu(struct v3d_compile *c, struct qinst *inst) for (int i = 0; i < sideband_nsrc; i++) { fprintf(stderr, ", "); - vir_print_reg(c, inst->src[i]); + vir_print_reg(c, inst, inst->src[i]); if (i < nsrc) fprintf(stderr, "%s", v3d_qpu_unpack_name(unpack[i])); } @@ -307,7 +316,7 @@ vir_dump_inst(struct v3d_compile *c, struct qinst *inst) if (vir_has_implicit_uniform(inst)) { fprintf(stderr, " "); - vir_print_reg(c, inst->src[vir_get_implicit_uniform_src(inst)]); + vir_print_reg(c, inst, inst->src[vir_get_implicit_uniform_src(inst)]); } break; |