On 03/12/2014 01:28 AM, Kenneth Graunke wrote:
> Ideally, we'd like to never even attempt the SIMD16 compile if we could
> know ahead of time that it won't succeed---it's purely a waste of time.
> This is especially important for state-based recompiles, which happen at
> draw time.
>
> The fragment shader compiler has a number of checks like:
>
> if (dispatch_width == 16)
> fail("...some reason...");
>
> This patch introduces a new no16() function which replaces the above
> pattern. In the SIMD8 compile, it sets a "SIMD16 will never work" flag.
> Then, brw_wm_fs_emit can check that flag, skip the SIMD16 compile, and
> issue a helpful performance warning if INTEL_DEBUG=perf is set. (In
> SIMD16 mode, no16() calls fail(), for safety's sake.)
>
> The great part is that this is not a heuristic---if the flag is set, we
> know with 100% certainty that the SIMD16 compile would fail. (It might
> fail anyway if we run out of registers, but it's always worth trying.)
>
> Signed-off-by: Kenneth Graunke <kenn...@whitecape.org>
> ---
> src/mesa/drivers/dri/i965/brw_fs.cpp | 69
> +++++++++++++++++++++++-----
> src/mesa/drivers/dri/i965/brw_fs.h | 4 ++
> src/mesa/drivers/dri/i965/brw_fs_visitor.cpp | 44 +++++++++---------
> 3 files changed, 83 insertions(+), 34 deletions(-)
>
> I forgot to send this one out...it applies on top of the previous 7 patches.
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp
> b/src/mesa/drivers/dri/i965/brw_fs.cpp
> index 62848be..9ad80c5 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
> @@ -647,9 +647,8 @@ fs_visitor::emit_shader_time_write(enum
> shader_time_shader_type type,
> }
>
> void
> -fs_visitor::fail(const char *format, ...)
> +fs_visitor::vfail(const char *format, va_list va)
> {
> - va_list va;
> char *msg;
>
> if (failed)
> @@ -657,9 +656,7 @@ fs_visitor::fail(const char *format, ...)
>
> failed = true;
>
> - va_start(va, format);
> msg = ralloc_vasprintf(mem_ctx, format, va);
> - va_end(va);
> msg = ralloc_asprintf(mem_ctx, "FS compile failed: %s\n", msg);
>
> this->fail_msg = msg;
> @@ -669,6 +666,49 @@ fs_visitor::fail(const char *format, ...)
> }
> }
>
> +void
> +fs_visitor::fail(const char *format, ...)
> +{
> + va_list va;
> +
> + va_start(va, format);
> + vfail(format, va);
> + va_end(va);
> +}
> +
> +/**
> + * Mark this program as impossible to compile in SIMD16 mode.
> + *
> + * During the SIMD8 compile (which happens first), we can detect and flag
> + * things that are unsupported in SIMD16 mode, so the compiler can skip
> + * the SIMD16 compile altogether.
> + *
> + * During a SIMD16 compile (if one happens anyway), this just calls fail().
> + */
> +void
> +fs_visitor::no16(const char *format, ...)
> +{
> + va_list va;
> +
> + va_start(va, format);
> +
> + if (dispatch_width == 16) {
> + vfail(format, va);
> + return;
> + }
> +
> + simd16_unsupported = true;
> +
> + if (INTEL_DEBUG & DEBUG_PERF) {
> + if (no16_msg)
> + ralloc_vasprintf_append(&no16_msg, format, va);
> + else
> + no16_msg = ralloc_vasprintf(mem_ctx, format, va);
> + }
> +
> + va_end(va);
> +}
> +
> fs_inst *
> fs_visitor::emit(enum opcode opcode)
> {
> @@ -1356,8 +1396,8 @@ fs_visitor::emit_math(enum opcode opcode, fs_reg dst,
> fs_reg src0, fs_reg src1)
> switch (opcode) {
> case SHADER_OPCODE_INT_QUOTIENT:
> case SHADER_OPCODE_INT_REMAINDER:
> - if (brw->gen >= 7 && dispatch_width == 16)
> - fail("SIMD16 INTDIV unsupported\n");
> + if (brw->gen >= 7)
> + no16("SIMD16 INTDIV unsupported\n");
> break;
> case SHADER_OPCODE_POW:
> break;
> @@ -3504,13 +3544,18 @@ brw_wm_fs_emit(struct brw_context *brw, struct
> brw_wm_compile *c,
> exec_list *simd16_instructions = NULL;
> fs_visitor v2(brw, c, prog, fp, 16);
> if (brw->gen >= 5 && likely(!(INTEL_DEBUG & DEBUG_NO16))) {
> - /* Try a SIMD16 compile */
> - v2.import_uniforms(&v);
> - if (!v2.run()) {
> - perf_debug("SIMD16 shader failed to compile, falling back to "
> - "SIMD8 at a 10-20%% performance cost: %s", v2.fail_msg);
> + if (!v.simd16_unsupported) {
> + /* Try a SIMD16 compile */
> + v2.import_uniforms(&v);
> + if (!v2.run()) {
> + perf_debug("SIMD16 shader failed to compile, falling back to "
> + "SIMD8 at a 10-20%% performance cost: %s",
> v2.fail_msg);
> + } else {
> + simd16_instructions = &v2.instructions;
> + }
> } else {
> - simd16_instructions = &v2.instructions;
> + perf_debug("SIMD16 shader unsupported, falling back to "
> + "SIMD8 at a 10-20%% performance cost: %s", v.no16_msg);
> }
> }
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs.h
> b/src/mesa/drivers/dri/i965/brw_fs.h
> index 9de1f3a..0d064f6 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs.h
> +++ b/src/mesa/drivers/dri/i965/brw_fs.h
> @@ -381,7 +381,9 @@ public:
> void insert_gen4_send_dependency_workarounds();
> void insert_gen4_pre_send_dependency_workarounds(fs_inst *inst);
> void insert_gen4_post_send_dependency_workarounds(fs_inst *inst);
> + void vfail(const char *msg, va_list args);
> void fail(const char *msg, ...);
> + void no16(const char *msg, ...);
> void lower_uniform_pull_constant_loads();
>
> void push_force_uncompressed();
> @@ -541,6 +543,8 @@ public:
>
> bool failed;
> char *fail_msg;
> + bool simd16_unsupported;
> + char *no16_msg;
>
> /* Result of last visit() method. */
> fs_reg result;I neglected to to initialize these, so SIMD16 compiles would randomly fail for no reason. I've fixed that in the newest version. --Ken
signature.asc
Description: OpenPGP digital signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
