Hi Brian,
On Friday, August 05, 2016 09:27:34 Brian Paul wrote:
> > As requested with the initial creation of util/bitscan.h
> > now move other bitscan related functions into util.
> > Make use of win32 intrinsics for util_last_bit/fls if present.
> >
> > Signed-off-by: Mathias Fröhlich <mathias.froehl...@web.de>
> >
> >
> > Any testing especially on win32 is apprechiated.
>
> We need to include "c99_compat.h" in u_bitcast.h in order to get the
> "inline" macro.
Confused now: The file src/util/bitscan.h already contains the
mentioned include. I think the patch should safely have 'inline'
present when it needs?
Oh, I see you did test that for me and failed on win32 by the recent
introduction of u_bitcast.h that does not include c99_compat.h - right?
best
Mathias
>
> -Brian
>
>
> >
> > Please review
> >
> > Mathias
> >
> >
> > ---
> > src/compiler/glsl/glsl_to_nir.cpp | 2 +-
> > src/gallium/auxiliary/util/u_math.h | 64 ------------------
> > src/mesa/drivers/dri/i965/brw_cs.c | 2 +-
> > src/mesa/drivers/dri/i965/brw_draw.c | 10 +--
> > src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 6 +-
> > src/mesa/drivers/dri/i965/brw_program.c | 2 +-
> > src/mesa/drivers/dri/i965/brw_shader.cpp | 2 +-
> > src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp | 2 +-
> > src/mesa/drivers/dri/i965/brw_wm_surface_state.c | 2 +-
> > src/mesa/main/imports.h | 45 -------------
> > src/mesa/program/prog_to_nir.c | 6 +-
> > src/util/bitscan.h | 80
> > +++++++++++++++++++++++
> > 12 files changed, 97 insertions(+), 126 deletions(-)
> >
> > diff --git a/src/compiler/glsl/glsl_to_nir.cpp
> > b/src/compiler/glsl/glsl_to_nir.cpp
> > index 20302e3..d3cc5b4 100644
> > --- a/src/compiler/glsl/glsl_to_nir.cpp
> > +++ b/src/compiler/glsl/glsl_to_nir.cpp
> > @@ -146,7 +146,7 @@ glsl_to_nir(const struct gl_shader_program *shader_prog,
> > shader->info.name = ralloc_asprintf(shader, "GLSL%d",
> > shader_prog->Name);
> > if (shader_prog->Label)
> > shader->info.label = ralloc_strdup(shader, shader_prog->Label);
> > - shader->info.num_textures = _mesa_fls(sh->Program->SamplersUsed);
> > + shader->info.num_textures = util_last_bit(sh->Program->SamplersUsed);
> > shader->info.num_ubos = sh->NumUniformBlocks;
> > shader->info.num_abos = shader_prog->NumAtomicBuffers;
> > shader->info.num_ssbos = sh->NumShaderStorageBlocks;
> > diff --git a/src/gallium/auxiliary/util/u_math.h
> > b/src/gallium/auxiliary/util/u_math.h
> > index 1661e63..a923271 100644
> > --- a/src/gallium/auxiliary/util/u_math.h
> > +++ b/src/gallium/auxiliary/util/u_math.h
> > @@ -347,70 +347,6 @@ util_half_inf_sign(int16_t x)
> >
> >
> > /**
> > - * Find last bit set in a word. The least significant bit is 1.
> > - * Return 0 if no bits are set.
> > - */
> > -static inline unsigned
> > -util_last_bit(unsigned u)
> > -{
> > -#if defined(HAVE___BUILTIN_CLZ)
> > - return u == 0 ? 0 : 32 - __builtin_clz(u);
> > -#else
> > - unsigned r = 0;
> > - while (u) {
> > - r++;
> > - u >>= 1;
> > - }
> > - return r;
> > -#endif
> > -}
> > -
> > -/**
> > - * Find last bit set in a word. The least significant bit is 1.
> > - * Return 0 if no bits are set.
> > - */
> > -static inline unsigned
> > -util_last_bit64(uint64_t u)
> > -{
> > -#if defined(HAVE___BUILTIN_CLZLL)
> > - return u == 0 ? 0 : 64 - __builtin_clzll(u);
> > -#else
> > - unsigned r = 0;
> > - while (u) {
> > - r++;
> > - u >>= 1;
> > - }
> > - return r;
> > -#endif
> > -}
> > -
> > -/**
> > - * Find last bit in a word that does not match the sign bit. The least
> > - * significant bit is 1.
> > - * Return 0 if no bits are set.
> > - */
> > -static inline unsigned
> > -util_last_bit_signed(int i)
> > -{
> > - if (i >= 0)
> > - return util_last_bit(i);
> > - else
> > - return util_last_bit(~(unsigned)i);
> > -}
> > -
> > -/* Returns a bitfield in which the first count bits starting at start are
> > - * set.
> > - */
> > -static inline unsigned
> > -u_bit_consecutive(unsigned start, unsigned count)
> > -{
> > - assert(start + count <= 32);
> > - if (count == 32)
> > - return ~0;
> > - return ((1u << count) - 1) << start;
> > -}
> > -
> > -/**
> > * Return float bits.
> > */
> > static inline unsigned
> > diff --git a/src/mesa/drivers/dri/i965/brw_cs.c
> > b/src/mesa/drivers/dri/i965/brw_cs.c
> > index 655adc1..6685acd 100644
> > --- a/src/mesa/drivers/dri/i965/brw_cs.c
> > +++ b/src/mesa/drivers/dri/i965/brw_cs.c
> > @@ -220,7 +220,7 @@ brw_upload_cs_prog(struct brw_context *brw)
> > return;
> >
> > brw->cs.base.sampler_count =
> > - _mesa_fls(ctx->ComputeProgram._Current->Base.SamplersUsed);
> > + util_last_bit(ctx->ComputeProgram._Current->Base.SamplersUsed);
> >
> > brw_cs_populate_key(brw, &key);
> >
> > diff --git a/src/mesa/drivers/dri/i965/brw_draw.c
> > b/src/mesa/drivers/dri/i965/brw_draw.c
> > index d7a1ba3..9b1e18c 100644
> > --- a/src/mesa/drivers/dri/i965/brw_draw.c
> > +++ b/src/mesa/drivers/dri/i965/brw_draw.c
> > @@ -452,15 +452,15 @@ brw_try_draw_prims(struct gl_context *ctx,
> > * index.
> > */
> > brw->wm.base.sampler_count =
> > - _mesa_fls(ctx->FragmentProgram._Current->Base.SamplersUsed);
> > + util_last_bit(ctx->FragmentProgram._Current->Base.SamplersUsed);
> > brw->gs.base.sampler_count = ctx->GeometryProgram._Current ?
> > - _mesa_fls(ctx->GeometryProgram._Current->Base.SamplersUsed) : 0;
> > + util_last_bit(ctx->GeometryProgram._Current->Base.SamplersUsed) : 0;
> > brw->tes.base.sampler_count = ctx->TessEvalProgram._Current ?
> > - _mesa_fls(ctx->TessEvalProgram._Current->Base.SamplersUsed) : 0;
> > + util_last_bit(ctx->TessEvalProgram._Current->Base.SamplersUsed) : 0;
> > brw->tcs.base.sampler_count = ctx->TessCtrlProgram._Current ?
> > - _mesa_fls(ctx->TessCtrlProgram._Current->Base.SamplersUsed) : 0;
> > + util_last_bit(ctx->TessCtrlProgram._Current->Base.SamplersUsed) : 0;
> > brw->vs.base.sampler_count =
> > - _mesa_fls(ctx->VertexProgram._Current->Base.SamplersUsed);
> > + util_last_bit(ctx->VertexProgram._Current->Base.SamplersUsed);
> >
> > intel_prepare_render(brw);
> > brw_predraw_set_aux_buffers(brw);
> > diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > index c1f413b..117eabe 100644
> > --- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > +++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > @@ -1845,7 +1845,7 @@ fs_visitor::emit_gs_control_data_bits(const fs_reg
> > &vertex_count)
> > fs_reg prev_count = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
> > abld.ADD(prev_count, vertex_count, brw_imm_ud(0xffffffffu));
> > unsigned log2_bits_per_vertex =
> > - _mesa_fls(gs_compile->control_data_bits_per_vertex);
> > + util_last_bit(gs_compile->control_data_bits_per_vertex);
> > abld.SHR(dword_index, prev_count, brw_imm_ud(6u -
> > log2_bits_per_vertex));
> >
> > if (per_slot_offset.file != BAD_FILE) {
> > @@ -2789,7 +2789,7 @@ fs_visitor::nir_emit_tcs_intrinsic(const fs_builder
> > &bld,
> > if (mask == 0)
> > break;
> >
> > - unsigned num_components = _mesa_fls(mask);
> > + unsigned num_components = util_last_bit(mask);
> > enum opcode opcode;
> >
> > /* We can only pack two 64-bit components in a single message, so
> > send
> > @@ -4547,7 +4547,7 @@ fs_visitor::nir_emit_texture(const fs_builder &bld,
> > nir_tex_instr *instr)
> > nir_ssa_def_components_read(&instr->dest.ssa):
> > (1 << dest_size) - 1;
> > assert(write_mask != 0); /* dead code should have been eliminated */
> > - inst->regs_written = _mesa_fls(write_mask) * dispatch_width / 8;
> > + inst->regs_written = util_last_bit(write_mask) * dispatch_width / 8;
> > } else {
> > inst->regs_written = 4 * dispatch_width / 8;
> > }
> > diff --git a/src/mesa/drivers/dri/i965/brw_program.c
> > b/src/mesa/drivers/dri/i965/brw_program.c
> > index 7785490..0e55c7b 100644
> > --- a/src/mesa/drivers/dri/i965/brw_program.c
> > +++ b/src/mesa/drivers/dri/i965/brw_program.c
> > @@ -674,7 +674,7 @@ brw_setup_tex_for_precompile(struct brw_context *brw,
> > struct gl_program *prog)
> > {
> > const bool has_shader_channel_select = brw->is_haswell || brw->gen >=
> > 8;
> > - unsigned sampler_count = _mesa_fls(prog->SamplersUsed);
> > + unsigned sampler_count = util_last_bit(prog->SamplersUsed);
> > for (unsigned i = 0; i < sampler_count; i++) {
> > if (!has_shader_channel_select && (prog->ShadowSamplers & (1 <<
> > i))) {
> > /* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
> > diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp
> > b/src/mesa/drivers/dri/i965/brw_shader.cpp
> > index 559e44c..62bad9b 100644
> > --- a/src/mesa/drivers/dri/i965/brw_shader.cpp
> > +++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
> > @@ -1157,7 +1157,7 @@
> > brw_assign_common_binding_table_offsets(gl_shader_stage stage,
> > uint32_t
> > next_binding_table_offset)
> > {
> > const struct gl_linked_shader *shader = NULL;
> > - int num_textures = _mesa_fls(prog->SamplersUsed);
> > + int num_textures = util_last_bit(prog->SamplersUsed);
> >
> > if (shader_prog)
> > shader = shader_prog->_LinkedShaders[stage];
> > diff --git a/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp
> > b/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp
> > index 927438f..c5886d4 100644
> > --- a/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp
> > +++ b/src/mesa/drivers/dri/i965/brw_vec4_gs_visitor.cpp
> > @@ -334,7 +334,7 @@ vec4_gs_visitor::emit_control_data_bits()
> > emit(ADD(dst_reg(prev_count), this->vertex_count,
> > brw_imm_ud(0xffffffffu)));
> > unsigned log2_bits_per_vertex =
> > - _mesa_fls(c->control_data_bits_per_vertex);
> > + util_last_bit(c->control_data_bits_per_vertex);
> > emit(SHR(dst_reg(dword_index), prev_count,
> > brw_imm_ud(6 - log2_bits_per_vertex)));
> > }
> > diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> > b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> > index 9bee7dd..a53f9da 100644
> > --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> > +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> > @@ -998,7 +998,7 @@ update_stage_texture_surfaces(struct brw_context *brw,
> > else
> > surf_offset +=
> > stage_state->prog_data->binding_table.plane_start[plane];
> >
> > - unsigned num_samplers = _mesa_fls(prog->SamplersUsed);
> > + unsigned num_samplers = util_last_bit(prog->SamplersUsed);
> > for (unsigned s = 0; s < num_samplers; s++) {
> > surf_offset[s] = 0;
> >
> > diff --git a/src/mesa/main/imports.h b/src/mesa/main/imports.h
> > index 05cc5ca..21bcb10 100644
> > --- a/src/mesa/main/imports.h
> > +++ b/src/mesa/main/imports.h
> > @@ -339,51 +339,6 @@ extern unsigned int
> > _mesa_bitcount_64(uint64_t n);
> > #endif
> >
> > -/**
> > - * Find the last (most significant) bit set in a word.
> > - *
> > - * Essentially ffs() in the reverse direction.
> > - */
> > -static inline unsigned int
> > -_mesa_fls(unsigned int n)
> > -{
> > -#ifdef HAVE___BUILTIN_CLZ
> > - return n == 0 ? 0 : 32 - __builtin_clz(n);
> > -#else
> > - unsigned int v = 1;
> > -
> > - if (n == 0)
> > - return 0;
> > -
> > - while (n >>= 1)
> > - v++;
> > -
> > - return v;
> > -#endif
> > -}
> > -
> > -/**
> > - * Find the last (most significant) bit set in a uint64_t value.
> > - *
> > - * Essentially ffsll() in the reverse direction.
> > - */
> > -static inline unsigned int
> > -_mesa_flsll(uint64_t n)
> > -{
> > -#ifdef HAVE___BUILTIN_CLZLL
> > - return n == 0 ? 0 : 64 - __builtin_clzll(n);
> > -#else
> > - unsigned int v = 1;
> > -
> > - if (n == 0)
> > - return 0;
> > -
> > - while (n >>= 1)
> > - v++;
> > -
> > - return v;
> > -#endif
> > -}
> >
> > static inline bool
> > _mesa_half_is_negative(GLhalfARB h)
> > diff --git a/src/mesa/program/prog_to_nir.c b/src/mesa/program/prog_to_nir.c
> > index 9e01151..1efd1a1 100644
> > --- a/src/mesa/program/prog_to_nir.c
> > +++ b/src/mesa/program/prog_to_nir.c
> > @@ -887,7 +887,7 @@ setup_registers_and_variables(struct ptn_compile *c)
> > struct nir_shader *shader = b->shader;
> >
> > /* Create input variables. */
> > - const int num_inputs = _mesa_flsll(c->prog->InputsRead);
> > + const int num_inputs = util_last_bit64(c->prog->InputsRead);
> > for (int i = 0; i < num_inputs; i++) {
> > if (!(c->prog->InputsRead & BITFIELD64_BIT(i)))
> > continue;
> > @@ -948,7 +948,7 @@ setup_registers_and_variables(struct ptn_compile *c)
> > }
> >
> > /* Create output registers and variables. */
> > - int max_outputs = _mesa_fls(c->prog->OutputsWritten);
> > + int max_outputs = util_last_bit(c->prog->OutputsWritten);
> > c->output_regs = rzalloc_array(c, nir_register *, max_outputs);
> >
> > for (int i = 0; i < max_outputs; i++) {
> > @@ -1043,7 +1043,7 @@ prog_to_nir(const struct gl_program *prog,
> > ptn_add_output_stores(c);
> >
> > s->info.name = ralloc_asprintf(s, "ARB%d", prog->Id);
> > - s->info.num_textures = _mesa_fls(prog->SamplersUsed);
> > + s->info.num_textures = util_last_bit(prog->SamplersUsed);
> > s->info.num_ubos = 0;
> > s->info.num_abos = 0;
> > s->info.num_ssbos = 0;
> > diff --git a/src/util/bitscan.h b/src/util/bitscan.h
> > index 4999b74..a5bb34e 100644
> > --- a/src/util/bitscan.h
> > +++ b/src/util/bitscan.h
> > @@ -29,6 +29,7 @@
> > #ifndef BITSCAN_H
> > #define BITSCAN_H
> >
> > +#include <assert.h>
> > #include <stdint.h>
> >
> > #if defined(_MSC_VER)
> > @@ -146,6 +147,85 @@ u_bit_scan_consecutive_range64(uint64_t *mask, int
> > *start, int *count)
> > }
> >
> >
> > +/**
> > + * Find last bit set in a word. The least significant bit is 1.
> > + * Return 0 if no bits are set.
> > + * Essentially ffs() in the reverse direction.
> > + */
> > +static inline unsigned
> > +util_last_bit(unsigned u)
> > +{
> > +#if defined(HAVE___BUILTIN_CLZ)
> > + return u == 0 ? 0 : 32 - __builtin_clz(u);
> > +#elif defined(_MSC_VER) && (_M_IX86 || _M_ARM || _M_AMD64 || _M_IA64)
> > + unsigned long index;
> > + if (_BitScanReverse(&index, u))
> > + return index;
> > + else
> > + return 0;
> > +#else
> > + unsigned r = 0;
> > + while (u) {
> > + r++;
> > + u >>= 1;
> > + }
> > + return r;
> > +#endif
> > +}
> > +
> > +/**
> > + * Find last bit set in a word. The least significant bit is 1.
> > + * Return 0 if no bits are set.
> > + * Essentially ffsll() in the reverse direction.
> > + */
> > +static inline unsigned
> > +util_last_bit64(uint64_t u)
> > +{
> > +#if defined(HAVE___BUILTIN_CLZLL)
> > + return u == 0 ? 0 : 64 - __builtin_clzll(u);
> > +#elif defined(_MSC_VER) && (_M_AMD64 || _M_ARM || _M_IA64)
> > + unsigned long index;
> > + if (_BitScanReverse64(&index, u))
> > + return index;
> > + else
> > + return 0;
> > +#else
> > + unsigned r = 0;
> > + while (u) {
> > + r++;
> > + u >>= 1;
> > + }
> > + return r;
> > +#endif
> > +}
> > +
> > +/**
> > + * Find last bit in a word that does not match the sign bit. The least
> > + * significant bit is 1.
> > + * Return 0 if no bits are set.
> > + */
> > +static inline unsigned
> > +util_last_bit_signed(int i)
> > +{
> > + if (i >= 0)
> > + return util_last_bit(i);
> > + else
> > + return util_last_bit(~(unsigned)i);
> > +}
> > +
> > +/* Returns a bitfield in which the first count bits starting at start are
> > + * set.
> > + */
> > +static inline unsigned
> > +u_bit_consecutive(unsigned start, unsigned count)
> > +{
> > + assert(start + count <= 32);
> > + if (count == 32)
> > + return ~0;
> > + return ((1u << count) - 1) << start;
> > +}
> > +
> > +
> > #ifdef __cplusplus
> > }
> > #endif
> >
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
