From: Rob Clark <robcl...@freedesktop.org>
The first pass would be used if you are not using nir_lower_indirect_deref's
(which would be the case for ir3, since we do have indirect register access).
This simply ensure there are no indirected input/outputs.
The second pass splits up array/matrix/struct inputs/outputs into primitive
types.
In the short term, we need this for gallium, since unread elements of a VS
shader input (at granularity of vec4) will not be assigned space in the VBO.
But Jason mentioned that it would also be useful for implementing linking
in NIR.
There are a few open TODO's in the 2nd patch, mostly because I wasn't really
sure how i965 would like things to look in the type_size_scalar case.
For an example of what this looks like in practice:
varying vec4 m1[4];
varying vec4 m2[4];
attribute vec4 a1[4];
uniform int index;
void main()
{
gl_Position = a1[0];
m1[0] = a1[0];
m1[1] = a1[1];
m1[2] = a1[2];
m1[3] = a1[0];
m2[0] = a1[index];
m2[1] = a1[index];
m2[2] = a1[index];
m2[3] = a1[index];
}
becomes:
shader: MESA_SHADER_VERTEX
decl_var uniform INTERP_QUALIFIER_NONE int index (0, 0)
decl_var shader_in INTERP_QUALIFIER_NONE vec4[4] a1
(VERT_ATTRIB_GENERIC0, 0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 gl_Position
(VARYING_SLOT_POS, 0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4[4] m1
(VARYING_SLOT_VAR0, 0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4[4] m2
(VARYING_SLOT_VAR4, 0)
decl_function main returning void
impl main {
block block_0:
/* preds: */
vec4 ssa_0 = intrinsic load_var () (a1[0]) ()
vec4 ssa_1 = intrinsic load_var () (a1[0]) ()
vec4 ssa_2 = intrinsic load_var () (a1[1]) ()
vec4 ssa_3 = intrinsic load_var () (a1[2]) ()
vec4 ssa_4 = intrinsic load_var () (a1[0]) ()
vec1 ssa_5 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_6 = intrinsic load_uniform (ssa_5) () (0) /* base=0 */
/* index */
vec4 ssa_7 = intrinsic load_var () (a1[ssa_6]) ()
vec1 ssa_8 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_9 = intrinsic load_uniform (ssa_8) () (0) /* base=0 */
/* index */
vec4 ssa_10 = intrinsic load_var () (a1[ssa_9]) ()
vec1 ssa_11 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_12 = intrinsic load_uniform (ssa_11) () (0) /* base=0
*/ /* index */
vec4 ssa_13 = intrinsic load_var () (a1[ssa_12]) ()
vec1 ssa_14 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_15 = intrinsic load_uniform (ssa_14) () (0) /* base=0
*/ /* index */
vec4 ssa_16 = intrinsic load_var () (a1[ssa_15]) ()
intrinsic store_var (ssa_0) (gl_Position) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_1) (m1[0]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_2) (m1[1]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_3) (m1[2]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_4) (m1[3]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_7) (m2[0]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_10) (m2[1]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_13) (m2[2]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_16) (m2[3]) (15) /* wrmask=xyzw */
/* succs: block_1 */
block block_1:
}
and then after lower_indirect_io:
shader: MESA_SHADER_VERTEX
decl_var uniform INTERP_QUALIFIER_NONE int index (0, 0)
decl_var shader_in INTERP_QUALIFIER_NONE vec4[4] a1
(VERT_ATTRIB_GENERIC0, 0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 gl_Position
(VARYING_SLOT_POS, 0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4[4] m1
(VARYING_SLOT_VAR0, 0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4[4] m2
(VARYING_SLOT_VAR4, 0)
decl_function main returning void
impl main {
decl_var INTERP_QUALIFIER_NONE vec4[4] shadow_a1
block block_0:
/* preds: */
intrinsic copy_var () (shadow_a1, a1) ()
vec4 ssa_0 = intrinsic load_var () (shadow_a1[0]) ()
vec4 ssa_1 = intrinsic load_var () (shadow_a1[0]) ()
vec4 ssa_2 = intrinsic load_var () (shadow_a1[1]) ()
vec4 ssa_3 = intrinsic load_var () (shadow_a1[2]) ()
vec4 ssa_4 = intrinsic load_var () (shadow_a1[0]) ()
vec1 ssa_5 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_6 = intrinsic load_uniform (ssa_5) () (0) /* base=0 */
/* index */
vec4 ssa_7 = intrinsic load_var () (shadow_a1[ssa_6]) ()
vec1 ssa_8 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_9 = intrinsic load_uniform (ssa_8) () (0) /* base=0 */
/* index */
vec4 ssa_10 = intrinsic load_var () (shadow_a1[ssa_9]) ()
vec1 ssa_11 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_12 = intrinsic load_uniform (ssa_11) () (0) /* base=0
*/ /* index */
vec4 ssa_13 = intrinsic load_var () (shadow_a1[ssa_12]) ()
vec1 ssa_14 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_15 = intrinsic load_uniform (ssa_14) () (0) /* base=0
*/ /* index */
vec4 ssa_16 = intrinsic load_var () (shadow_a1[ssa_15]) ()
intrinsic store_var (ssa_0) (gl_Position) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_1) (m1[0]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_2) (m1[1]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_3) (m1[2]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_4) (m1[3]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_7) (m2[0]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_10) (m2[1]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_13) (m2[2]) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_16) (m2[3]) (15) /* wrmask=xyzw */
/* succs: block_1 */
block block_1:
}
and then after lower_io_types:
shader: MESA_SHADER_VERTEX
decl_var uniform INTERP_QUALIFIER_NONE int index (0, 0)
decl_var shader_in INTERP_QUALIFIER_NONE vec4 a1_0
(VERT_ATTRIB_GENERIC0, 0)
decl_var shader_in INTERP_QUALIFIER_NONE vec4 a1_1
(VERT_ATTRIB_GENERIC1, 0)
decl_var shader_in INTERP_QUALIFIER_NONE vec4 a1_2
(VERT_ATTRIB_GENERIC2, 0)
decl_var shader_in INTERP_QUALIFIER_NONE vec4 a1_3
(VERT_ATTRIB_GENERIC3, 0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 gl_Position
(VARYING_SLOT_POS, 0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 m1_0 (VARYING_SLOT_VAR0,
0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 m1_1 (VARYING_SLOT_VAR1,
0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 m1_2 (VARYING_SLOT_VAR2,
0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 m1_3 (VARYING_SLOT_VAR3,
0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 m2_0 (VARYING_SLOT_VAR4,
0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 m2_1 (VARYING_SLOT_VAR5,
0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 m2_2 (VARYING_SLOT_VAR6,
0)
decl_var shader_out INTERP_QUALIFIER_NONE vec4 m2_3 (VARYING_SLOT_VAR7,
0)
decl_function main returning void
impl main {
decl_var INTERP_QUALIFIER_NONE vec4[4] shadow_a1
block block_0:
/* preds: */
vec4 ssa_17 = intrinsic load_var () (a1_0) ()
intrinsic store_var (ssa_17) (shadow_a1[0]) (15) /* wrmask=xyzw
*/
vec4 ssa_18 = intrinsic load_var () (a1_1) ()
intrinsic store_var (ssa_18) (shadow_a1[1]) (15) /* wrmask=xyzw
*/
vec4 ssa_19 = intrinsic load_var () (a1_2) ()
intrinsic store_var (ssa_19) (shadow_a1[2]) (15) /* wrmask=xyzw
*/
vec4 ssa_20 = intrinsic load_var () (a1_3) ()
intrinsic store_var (ssa_20) (shadow_a1[3]) (15) /* wrmask=xyzw
*/
vec4 ssa_0 = intrinsic load_var () (shadow_a1[0]) ()
vec4 ssa_1 = intrinsic load_var () (shadow_a1[0]) ()
vec4 ssa_2 = intrinsic load_var () (shadow_a1[1]) ()
vec4 ssa_3 = intrinsic load_var () (shadow_a1[2]) ()
vec4 ssa_4 = intrinsic load_var () (shadow_a1[0]) ()
vec1 ssa_5 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_6 = intrinsic load_uniform (ssa_5) () (0) /* base=0 */
/* index */
vec4 ssa_7 = intrinsic load_var () (shadow_a1[ssa_6]) ()
vec1 ssa_8 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_9 = intrinsic load_uniform (ssa_8) () (0) /* base=0 */
/* index */
vec4 ssa_10 = intrinsic load_var () (shadow_a1[ssa_9]) ()
vec1 ssa_11 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_12 = intrinsic load_uniform (ssa_11) () (0) /* base=0
*/ /* index */
vec4 ssa_13 = intrinsic load_var () (shadow_a1[ssa_12]) ()
vec1 ssa_14 = load_const (0x00000000 /* 0.000000 */)
vec1 ssa_15 = intrinsic load_uniform (ssa_14) () (0) /* base=0
*/ /* index */
vec4 ssa_16 = intrinsic load_var () (shadow_a1[ssa_15]) ()
intrinsic store_var (ssa_0) (gl_Position) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_1) (m1_0) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_2) (m1_1) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_3) (m1_2) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_4) (m1_3) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_7) (m2_0) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_10) (m2_1) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_13) (m2_2) (15) /* wrmask=xyzw */
intrinsic store_var (ssa_16) (m2_3) (15) /* wrmask=xyzw */
/* succs: block_1 */
block block_1:
}
Rob Clark (2):
nir: lower-indirect-io pass
nir: lower-io-types pass
src/compiler/Makefile.sources | 2 +
src/compiler/nir/nir.h | 3 +
src/compiler/nir/nir_lower_indirect_io.c | 151 ++++++++++++++++++++++++++
src/compiler/nir/nir_lower_io_types.c | 178 +++++++++++++++++++++++++++++++
4 files changed, 334 insertions(+)
create mode 100644 src/compiler/nir/nir_lower_indirect_io.c
create mode 100644 src/compiler/nir/nir_lower_io_types.c
--
2.5.5
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
