[Mesa-dev] [PATCH 6/6] r600: add support for PK2H/UP2H
Ilia Mirkin
imirkin at alum.mit.edu
Sun Jan 3 13:25:56 PST 2016
Glenn (or anyone else interested in r600): feel free to take this
patch over. I've pushed the core (and nvc0) support, but due to
Roland's feedback I changed the return values to be replicated(ish),
not 100% sure how to do that and it seems best to leave it to you.
On Sat, Jan 2, 2016 at 7:38 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
> ---
> src/gallium/drivers/r600/r600_pipe.c | 2 +-
> src/gallium/drivers/r600/r600_shader.c | 102 +++++++++++++++++++++++++++++++--
> 2 files changed, 99 insertions(+), 5 deletions(-)
>
> diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
> index 70c1ec1..359fe41 100644
> --- a/src/gallium/drivers/r600/r600_pipe.c
> +++ b/src/gallium/drivers/r600/r600_pipe.c
> @@ -328,6 +328,7 @@ static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
> case PIPE_CAP_TEXTURE_QUERY_LOD:
> case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
> case PIPE_CAP_SAMPLER_VIEW_TARGET:
> + case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
> return family >= CHIP_CEDAR ? 1 : 0;
> case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
> return family >= CHIP_CEDAR ? 4 : 0;
> @@ -351,7 +352,6 @@ static int r600_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
> case PIPE_CAP_DRAW_PARAMETERS:
> case PIPE_CAP_MULTI_DRAW_INDIRECT:
> case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
> - case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
> return 0;
>
> case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
> diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
> index d411b0b..23ea34e 100644
> --- a/src/gallium/drivers/r600/r600_shader.c
> +++ b/src/gallium/drivers/r600/r600_shader.c
> @@ -8959,6 +8959,100 @@ static int tgsi_umad(struct r600_shader_ctx *ctx)
> return 0;
> }
>
> +static int tgsi_pk2h(struct r600_shader_ctx *ctx)
> +{
> + struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
> + struct r600_bytecode_alu alu;
> + int r;
> +
> + /* temp.xy = f32_to_f16(src) */
> + memset(&alu, 0, sizeof(struct r600_bytecode_alu));
> + alu.op = ALU_OP1_FLT32_TO_FLT16;
> + alu.dst.chan = 0;
> + alu.dst.sel = ctx->temp_reg;
> + alu.dst.write = 1;
> + r600_bytecode_src(&alu.src[0], &ctx->src[0], 0);
> + r = r600_bytecode_add_alu(ctx->bc, &alu);
> + if (r)
> + return r;
> + alu.dst.chan = 1;
> + r600_bytecode_src(&alu.src[0], &ctx->src[0], 1);
> + alu.last = 1;
> + r = r600_bytecode_add_alu(ctx->bc, &alu);
> + if (r)
> + return r;
> +
> + /* dst.x = temp.y * 0x10000 + temp.x */
> + memset(&alu, 0, sizeof(struct r600_bytecode_alu));
> + alu.op = ALU_OP3_MULADD_UINT24;
> + alu.is_op3 = 1;
> + tgsi_dst(ctx, &inst->Dst[0], 0, &alu.dst);
> + alu.last = 1;
> + alu.src[0].sel = ctx->temp_reg;
> + alu.src[0].chan = 1;
> + alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
> + alu.src[1].value = 0x10000;
> + alu.src[2].sel = ctx->temp_reg;
> + alu.src[2].chan = 0;
> + r = r600_bytecode_add_alu(ctx->bc, &alu);
> + if (r)
> + return r;
> +
> + return 0;
> +}
> +
> +static int tgsi_up2h(struct r600_shader_ctx *ctx)
> +{
> + struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
> + struct r600_bytecode_alu alu;
> + int r;
> + int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
> +
> + /* temp.x = src.x */
> + /* note: no need to mask out the high bits */
> + memset(&alu, 0, sizeof(struct r600_bytecode_alu));
> + alu.op = ALU_OP1_MOV;
> + alu.dst.chan = 0;
> + alu.dst.sel = ctx->temp_reg;
> + alu.dst.write = 1;
> + r600_bytecode_src(&alu.src[0], &ctx->src[0], 0);
> + r = r600_bytecode_add_alu(ctx->bc, &alu);
> + if (r)
> + return r;
> +
> + /* temp.y = src.x >> 16 */
> + memset(&alu, 0, sizeof(struct r600_bytecode_alu));
> + alu.op = ALU_OP2_LSHR_INT;
> + alu.dst.chan = 1;
> + alu.dst.sel = ctx->temp_reg;
> + alu.dst.write = 1;
> + r600_bytecode_src(&alu.src[0], &ctx->src[0], 0);
> + alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
> + alu.src[1].value = 16;
> + alu.last = 1;
> + r = r600_bytecode_add_alu(ctx->bc, &alu);
> + if (r)
> + return r;
> +
> + /* dst.xy = f16_to_f32(temp.xy) */
> + for (int i = 0; i < lasti + 1; i++) {
> + if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
> + continue;
> + memset(&alu, 0, sizeof(struct r600_bytecode_alu));
> + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
> + alu.op = ALU_OP1_FLT16_TO_FLT32;
> + alu.src[0].sel = ctx->temp_reg;
> + alu.src[0].chan = i;
> + if (i == lasti)
> + alu.last = 1;
> + r = r600_bytecode_add_alu(ctx->bc, &alu);
> + if (r)
> + return r;
> + }
> +
> + return 0;
> +}
> +
> static const struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[] = {
> [TGSI_OPCODE_ARL] = { ALU_OP0_NOP, tgsi_r600_arl},
> [TGSI_OPCODE_MOV] = { ALU_OP1_MOV, tgsi_op2},
> @@ -9205,7 +9299,7 @@ static const struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] =
> [TGSI_OPCODE_DDX] = { FETCH_OP_GET_GRADIENTS_H, tgsi_tex},
> [TGSI_OPCODE_DDY] = { FETCH_OP_GET_GRADIENTS_V, tgsi_tex},
> [TGSI_OPCODE_KILL] = { ALU_OP2_KILLGT, tgsi_kill}, /* unconditional kill */
> - [TGSI_OPCODE_PK2H] = { ALU_OP0_NOP, tgsi_unsupported},
> + [TGSI_OPCODE_PK2H] = { ALU_OP0_NOP, tgsi_pk2h},
> [TGSI_OPCODE_PK2US] = { ALU_OP0_NOP, tgsi_unsupported},
> [TGSI_OPCODE_PK4B] = { ALU_OP0_NOP, tgsi_unsupported},
> [TGSI_OPCODE_PK4UB] = { ALU_OP0_NOP, tgsi_unsupported},
> @@ -9220,7 +9314,7 @@ static const struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] =
> [TGSI_OPCODE_TEX] = { FETCH_OP_SAMPLE, tgsi_tex},
> [TGSI_OPCODE_TXD] = { FETCH_OP_SAMPLE_G, tgsi_tex},
> [TGSI_OPCODE_TXP] = { FETCH_OP_SAMPLE, tgsi_tex},
> - [TGSI_OPCODE_UP2H] = { ALU_OP0_NOP, tgsi_unsupported},
> + [TGSI_OPCODE_UP2H] = { ALU_OP0_NOP, tgsi_up2h},
> [TGSI_OPCODE_UP2US] = { ALU_OP0_NOP, tgsi_unsupported},
> [TGSI_OPCODE_UP4B] = { ALU_OP0_NOP, tgsi_unsupported},
> [TGSI_OPCODE_UP4UB] = { ALU_OP0_NOP, tgsi_unsupported},
> @@ -9427,7 +9521,7 @@ static const struct r600_shader_tgsi_instruction cm_shader_tgsi_instruction[] =
> [TGSI_OPCODE_DDX] = { FETCH_OP_GET_GRADIENTS_H, tgsi_tex},
> [TGSI_OPCODE_DDY] = { FETCH_OP_GET_GRADIENTS_V, tgsi_tex},
> [TGSI_OPCODE_KILL] = { ALU_OP2_KILLGT, tgsi_kill}, /* unconditional kill */
> - [TGSI_OPCODE_PK2H] = { ALU_OP0_NOP, tgsi_unsupported},
> + [TGSI_OPCODE_PK2H] = { ALU_OP0_NOP, tgsi_pk2h},
> [TGSI_OPCODE_PK2US] = { ALU_OP0_NOP, tgsi_unsupported},
> [TGSI_OPCODE_PK4B] = { ALU_OP0_NOP, tgsi_unsupported},
> [TGSI_OPCODE_PK4UB] = { ALU_OP0_NOP, tgsi_unsupported},
> @@ -9442,7 +9536,7 @@ static const struct r600_shader_tgsi_instruction cm_shader_tgsi_instruction[] =
> [TGSI_OPCODE_TEX] = { FETCH_OP_SAMPLE, tgsi_tex},
> [TGSI_OPCODE_TXD] = { FETCH_OP_SAMPLE_G, tgsi_tex},
> [TGSI_OPCODE_TXP] = { FETCH_OP_SAMPLE, tgsi_tex},
> - [TGSI_OPCODE_UP2H] = { ALU_OP0_NOP, tgsi_unsupported},
> + [TGSI_OPCODE_UP2H] = { ALU_OP0_NOP, tgsi_up2h},
> [TGSI_OPCODE_UP2US] = { ALU_OP0_NOP, tgsi_unsupported},
> [TGSI_OPCODE_UP4B] = { ALU_OP0_NOP, tgsi_unsupported},
> [TGSI_OPCODE_UP4UB] = { ALU_OP0_NOP, tgsi_unsupported},
> --
> 2.4.10
>
More information about the mesa-dev
mailing list