[Mesa-dev] [PATCH] i965/vec4: remove the generator hack for dual instanced GS
Francisco Jerez
currojerez at riseup.net
Fri Aug 26 19:39:43 UTC 2016
Iago Toral Quiroga <itoral at igalia.com> writes:
> This hack was introduced in commit 03ac2c7223f7645e3:
> i965/gs: Fix up gl_PointSize input swizzling for DUAL_INSTANCED gs
>
> Specifically to fixup the code we emitted to deal with gl_PointSize inputs
> in dual instance mode, where we were emitting a MOV to copy the point
> size from .w (where the hardware delivers it) to .x (because code will
> expect this to be a float). This meant that we were emitting a MOV
> to an ATTR destination that could have a width of 4 (in dual instanced
> mode) so it was necessary to fix the execution size and regioning of the
> instruction.
>
> Fortunately, Ken fixed this in 67c5d00273ca2:
> i965/vec4/gs: Stop munging the ATTR containing gl_PointSize.
>
> by using a WWWW swizzle instead of a MOV, and as the commit log in that
> patch states, we no longer emit instructions with ATTR destinations, so
> that makes the fixup code in the generator unnecessary.
>
> ---
>
> We tested this on the following platforms without regressions in piglit:
>
> - SandyBridge (full piglit run with our fp64 branch)
> - IvyBridge (full piglit run with master)
> - Haswell (shader.py with master)
> - Haswell (shader.py and forcing dual-instanced mode for GS with master)
>
> It would still probably be a good idea to run it through Jenkins just in case.
>
Ran through Jenkins and:
Reviewed-by: Francisco Jerez <currojerez at riseup.net>
> ---
> src/mesa/drivers/dri/i965/brw_vec4_generator.cpp | 31 ------------------------
> 1 file changed, 31 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
> index 7ad4f86..128eb2d 100644
> --- a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
> @@ -1499,34 +1499,6 @@ generate_code(struct brw_codegen *p,
> assert(inst->mlen <= BRW_MAX_MSG_LENGTH);
>
> unsigned pre_emit_nr_insn = p->nr_insn;
> - bool fix_exec_size = false;
> -
> - if (dst.width == BRW_WIDTH_4) {
> - /* This happens in attribute fixups for "dual instanced" geometry
> - * shaders, since they use attributes that are vec4's. Since the exec
> - * width is only 4, it's essential that the caller set
> - * force_writemask_all in order to make sure the instruction is executed
> - * regardless of which channels are enabled.
> - */
> - assert(inst->force_writemask_all);
> -
> - /* Fix up any <8;8,1> or <0;4,1> source registers to <4;4,1> to satisfy
> - * the following register region restrictions (from Graphics BSpec:
> - * 3D-Media-GPGPU Engine > EU Overview > Registers and Register Regions
> - * > Register Region Restrictions)
> - *
> - * 1. ExecSize must be greater than or equal to Width.
> - *
> - * 2. If ExecSize = Width and HorzStride != 0, VertStride must be set
> - * to Width * HorzStride."
> - */
> - for (int i = 0; i < 3; i++) {
> - if (src[i].file == BRW_GENERAL_REGISTER_FILE)
> - src[i] = stride(src[i], 4, 4, 1);
> - }
> - brw_set_default_exec_size(p, BRW_EXECUTE_4);
> - fix_exec_size = true;
> - }
>
> switch (inst->opcode) {
> case VEC4_OPCODE_UNPACK_UNIFORM:
> @@ -2028,9 +2000,6 @@ generate_code(struct brw_codegen *p,
> unreachable("Unsupported opcode");
> }
>
> - if (fix_exec_size)
> - brw_set_default_exec_size(p, BRW_EXECUTE_8);
> -
> if (inst->opcode == VEC4_OPCODE_PACK_BYTES) {
> /* Handled dependency hints in the generator. */
>
> --
> 2.7.4
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 212 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160826/7045537a/attachment.sig>
More information about the mesa-dev
mailing list