Mesa (main): radeonsi: add SI_PROFILE_CLAMP_DIV_BY_ZERO

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: 9685b5785b0a3862fb3eb28cd8f5ffc1a1ccf984
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=9685b5785b0a3862fb3eb28cd8f5ffc1a1ccf984

Author: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer at amd.com>
Date:   Mon Feb  7 18:45:50 2022 +0100

radeonsi: add SI_PROFILE_CLAMP_DIV_BY_ZERO

To enable divide by zero clamping per shader, instead of per app.

Reviewed-by: Marek Olšák <marek.olsak at amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14931>

---

 src/gallium/drivers/radeonsi/si_shader.h      | 1 +
 src/gallium/drivers/radeonsi/si_shader_llvm.c | 4 +++-
 2 files changed, 4 insertions(+), 1 deletion(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 62283444f25..24f2e829ed6 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -293,6 +293,7 @@ enum
 #define SI_PROFILE_IGNORE_LLVM_DISCARD_BUG   (1 << 2)
 #define SI_PROFILE_VS_NO_BINNING             (1 << 3)
 #define SI_PROFILE_PS_NO_BINNING             (1 << 4)
+#define SI_PROFILE_CLAMP_DIV_BY_ZERO         (1 << 5)
 
 /**
  * For VS shader keys, describe any fixups required for vertex fetch.
diff --git a/src/gallium/drivers/radeonsi/si_shader_llvm.c b/src/gallium/drivers/radeonsi/si_shader_llvm.c
index c6161858c1f..400d1b2230f 100644
--- a/src/gallium/drivers/radeonsi/si_shader_llvm.c
+++ b/src/gallium/drivers/radeonsi/si_shader_llvm.c
@@ -519,7 +519,6 @@ static bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *
    ctx->abi.clamp_shadow_reference = true;
    ctx->abi.robust_buffer_access = true;
    ctx->abi.convert_undef_to_zero = true;
-   ctx->abi.clamp_div_by_zero = ctx->screen->options.clamp_div_by_zero;
    ctx->abi.adjust_frag_coord_z = false;
    ctx->abi.disable_aniso_single_level = true;
 
@@ -536,6 +535,9 @@ static bool si_nir_build_llvm(struct si_shader_context *ctx, struct nir_shader *
          ctx->abi.outputs[i * 4 + j] = ac_build_alloca_undef(&ctx->ac, type, "");
    }
 
+   ctx->abi.clamp_div_by_zero = ctx->screen->options.clamp_div_by_zero ||
+                                info->options & SI_PROFILE_CLAMP_DIV_BY_ZERO;
+
    ac_nir_translate(&ctx->ac, &ctx->abi, &ctx->args, nir);
 
    return true;