radeonsi: increase gfx1100/gfx1101 physical vgprs

https://reviews.llvm.org/D134522

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18825>

diff --git a/src/amd/common/ac_binary.c b/src/amd/common/ac_binary.c
index a26e549..53e22e9 100644 (file)
--- a/src/amd/common/ac_binary.c
+++ b/src/amd/common/ac_binary.c
@@ -118,16 +118,6 @@ void ac_parse_shader_binary_config(const char *data, size_t nbytes, unsigned wav
    if (!conf->spi_ps_input_addr)
       conf->spi_ps_input_addr = conf->spi_ps_input_ena;
 
-   /* GFX 10.3 internally:
-    * - aligns VGPRS to 16 for Wave32 and 8 for Wave64
-    * - aligns LDS to 1024
-    *
-    * For shader-db stats, set num_vgprs that the hw actually uses.
-    */
-   if (info->gfx_level == GFX10_3) {
-      conf->num_vgprs = align(conf->num_vgprs, wave_size == 32 ? 16 : 8);
-   }
-
    /* Enable 64-bit and 16-bit denormals, because there is no performance
     * cost.
     *

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index de21a80..4797f0f 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -1058,10 +1058,25 @@ static void si_calculate_max_simd_waves(struct si_shader *shader)
    }
 
    if (conf->num_vgprs) {
+      /* GFX 10.3 internally:
+       * - aligns VGPRS to 16 for Wave32 and 8 for Wave64
+       * - aligns LDS to 1024
+       *
+       * For shader-db stats, set num_vgprs that the hw actually uses.
+       */
+      unsigned num_vgprs = conf->num_vgprs;
+      if (sscreen->info.family == CHIP_GFX1100 || sscreen->info.family == CHIP_GFX1101) {
+         num_vgprs = util_align_npot(num_vgprs, shader->wave_size == 32 ? 24 : 12);
+      } else if (sscreen->info.gfx_level == GFX10_3) {
+         num_vgprs = align(num_vgprs, shader->wave_size == 32 ? 16 : 8);
+      } else {
+         num_vgprs = align(num_vgprs, shader->wave_size == 32 ? 8 : 4);
+      }
+
       /* Always print wave limits as Wave64, so that we can compare
        * Wave32 and Wave64 with shader-db fairly. */
       unsigned max_vgprs = sscreen->info.num_physical_wave64_vgprs_per_simd;
-      max_simd_waves = MIN2(max_simd_waves, max_vgprs / conf->num_vgprs);
+      max_simd_waves = MIN2(max_simd_waves, max_vgprs / num_vgprs);
    }
 
    unsigned max_lds_per_simd = sscreen->info.lds_size_per_workgroup / 4;