[Mesa-stable] [Mesa-dev] [PATCH] r600: fix texture gradients instruction emission (v2)
Emil Velikov
emil.l.velikov at gmail.com
Wed Nov 26 12:46:09 PST 2014
Hello Dave,
Cherry picking this commit for the 10.3 branch resulted in a number of
non-trivial conflicts. Can you or anyone else familiar with the code
backport this for 10.3 ?
Thanks
Emil
On 18/11/14 22:53, Dave Airlie wrote:
> From: Dave Airlie <airlied at redhat.com>
>
> The piglit tests were failing, and it appeared to be SB
> optimising out things, but Glenn pointed out the gradients
> are meant to be clause local, so we should emit the texture
> instructions in the same clause. This moves things around
> to always copy to a temp and then emit the texture clauses
> for H/V.
>
> v2: Glenn pointed out we could get another ALU fetch in
> the wrong place, so load the src gpr earlier as well.
>
> Fixes at least:
> ./bin/tex-miplevel-selection textureGrad 2D
>
> Signed-off-by: Dave Airlie <airlied at redhat.com>
> ---
> src/gallium/drivers/r600/r600_shader.c | 59 ++++++++++++++++++----------------
> 1 file changed, 31 insertions(+), 28 deletions(-)
>
> diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
> index 76daf2c..41caac3 100644
> --- a/src/gallium/drivers/r600/r600_shader.c
> +++ b/src/gallium/drivers/r600/r600_shader.c
> @@ -5110,11 +5110,37 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
> }
>
> if (inst->Instruction.Opcode == TGSI_OPCODE_TXD) {
> + int temp_h, temp_v;
> /* TGSI moves the sampler to src reg 3 for TXD */
> sampler_src_reg = 3;
>
> sampler_index_mode = inst->Src[sampler_src_reg].Indirect.Index == 2 ? 2 : 0; // CF_INDEX_1 : CF_INDEX_NONE
>
> + src_loaded = TRUE;
> + for (i = 0; i < 3; i++) {
> + int treg = r600_get_temp(ctx);
> +
> + if (i == 0)
> + src_gpr = treg;
> + else if (i == 1)
> + temp_h = treg;
> + else
> + temp_v = treg;
> +
> + for (j = 0; j < 4; j++) {
> + memset(&alu, 0, sizeof(struct r600_bytecode_alu));
> + alu.op = ALU_OP1_MOV;
> + r600_bytecode_src(&alu.src[0], &ctx->src[i], j);
> + alu.dst.sel = treg;
> + alu.dst.chan = j;
> + if (j == 3)
> + alu.last = 1;
> + alu.dst.write = 1;
> + r = r600_bytecode_add_alu(ctx->bc, &alu);
> + if (r)
> + return r;
> + }
> + }
> for (i = 1; i < 3; i++) {
> /* set gradients h/v */
> memset(&tex, 0, sizeof(struct r600_bytecode_tex));
> @@ -5125,35 +5151,12 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
> tex.resource_id = tex.sampler_id + R600_MAX_CONST_BUFFERS;
> tex.resource_index_mode = sampler_index_mode;
>
> - if (tgsi_tex_src_requires_loading(ctx, i)) {
> - tex.src_gpr = r600_get_temp(ctx);
> - tex.src_sel_x = 0;
> - tex.src_sel_y = 1;
> - tex.src_sel_z = 2;
> - tex.src_sel_w = 3;
> + tex.src_gpr = (i == 1) ? temp_h : temp_v;
> + tex.src_sel_x = 0;
> + tex.src_sel_y = 1;
> + tex.src_sel_z = 2;
> + tex.src_sel_w = 3;
>
> - for (j = 0; j < 4; j++) {
> - memset(&alu, 0, sizeof(struct r600_bytecode_alu));
> - alu.op = ALU_OP1_MOV;
> - r600_bytecode_src(&alu.src[0], &ctx->src[i], j);
> - alu.dst.sel = tex.src_gpr;
> - alu.dst.chan = j;
> - if (j == 3)
> - alu.last = 1;
> - alu.dst.write = 1;
> - r = r600_bytecode_add_alu(ctx->bc, &alu);
> - if (r)
> - return r;
> - }
> -
> - } else {
> - tex.src_gpr = tgsi_tex_get_src_gpr(ctx, i);
> - tex.src_sel_x = ctx->src[i].swizzle[0];
> - tex.src_sel_y = ctx->src[i].swizzle[1];
> - tex.src_sel_z = ctx->src[i].swizzle[2];
> - tex.src_sel_w = ctx->src[i].swizzle[3];
> - tex.src_rel = ctx->src[i].rel;
> - }
> tex.dst_gpr = ctx->temp_reg; /* just to avoid confusing the asm scheduler */
> tex.dst_sel_x = tex.dst_sel_y = tex.dst_sel_z = tex.dst_sel_w = 7;
> if (inst->Texture.Texture != TGSI_TEXTURE_RECT) {
>
More information about the mesa-stable
mailing list