OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nir: Add lowering for nir_op_bit_count.

Browse Source 
This is basically the same as the GLSL lowering path.

v2: Fix typo in the link

Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
This commit is contained in:
Eric Anholt
2018-05-08 13:04:37 -07:00
parent 7afa26d4e3
commit 73953b0713
2 changed files with 38 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/compiler/nir/nir.h
View File

@@ -1913,6 +1913,8 @@ typedef struct nir_shader_compiler_options {
bool lower_bitfield_insert_to_shifts;
/** Lowers bitfield_reverse to shifts. */
bool lower_bitfield_reverse;
/** Lowers bit_count to shifts. */
bool lower_bit_count;
/** Lowers bfm to shifts and subtracts. */
bool lower_bfm;
/** Lowers ifind_msb to compare and ufind_msb */

36
src/compiler/nir/nir_lower_alu.c
View File

@@ -94,6 +94,42 @@ lower_alu_instr(nir_alu_instr *instr, nir_builder *b)
}
break;
case nir_op_bit_count:
if (b->shader->options->lower_bit_count) {
/* For more details, see:
*
* http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel
*/
nir_ssa_def *c1 = nir_imm_int(b, 1);
nir_ssa_def *c2 = nir_imm_int(b, 2);
nir_ssa_def *c4 = nir_imm_int(b, 4);
nir_ssa_def *c24 = nir_imm_int(b, 24);
nir_ssa_def *c33333333 = nir_imm_int(b, 0x33333333);
nir_ssa_def *c55555555 = nir_imm_int(b, 0x55555555);
nir_ssa_def *c0f0f0f0f = nir_imm_int(b, 0x0f0f0f0f);
nir_ssa_def *c01010101 = nir_imm_int(b, 0x01010101);
lowered = nir_ssa_for_alu_src(b, instr, 0);
lowered = nir_isub(b, lowered,
nir_iand(b, nir_ushr(b, lowered, c1), c55555555));
lowered = nir_iadd(b,
nir_iand(b, lowered, c33333333),
nir_iand(b, nir_ushr(b, lowered, c2), c33333333));
lowered = nir_ushr(b,
nir_imul(b,
nir_iand(b,
nir_iadd(b,
lowered,
nir_ushr(b, lowered, c4)),
c0f0f0f0f),
c01010101),
c24);
}
break;
case nir_op_imul_high:
case nir_op_umul_high:
if (b->shader->options->lower_mul_high) {