ac/gpu_info: set cu_mask correctly for Arcturus

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14122>

diff --git a/src/amd/common/ac_gpu_info.c b/src/amd/common/ac_gpu_info.c
index e5b8f64..a1cc9e1 100644 (file)
--- a/src/amd/common/ac_gpu_info.c
+++ b/src/amd/common/ac_gpu_info.c
@@ -932,19 +932,23 @@ bool ac_query_gpu_info(int fd, void *dev_p, struct radeon_info *info,
    info->num_good_compute_units = 0;
    for (i = 0; i < info->max_se; i++) {
       for (j = 0; j < info->max_sa_per_se; j++) {
-         /*
-          * The cu bitmap in amd gpu info structure is
-          * 4x4 size array, and it's usually suitable for Vega
-          * ASICs which has 4*2 SE/SH layout.
-          * But for Arcturus, SE/SH layout is changed to 8*1.
-          * To mostly reduce the impact, we make it compatible
-          * with current bitmap array as below:
-          *    SE4,SH0 --> cu_bitmap[0][1]
-          *    SE5,SH0 --> cu_bitmap[1][1]
-          *    SE6,SH0 --> cu_bitmap[2][1]
-          *    SE7,SH0 --> cu_bitmap[3][1]
-          */
-         info->cu_mask[i % 4][j + i / 4] = amdinfo->cu_bitmap[i % 4][j + i / 4];
+         if (info->family == CHIP_ARCTURUS) {
+            /* The CU bitmap in amd gpu info structure is
+             * 4x4 size array, and it's usually suitable for Vega
+             * ASICs which has 4*2 SE/SA layout.
+             * But for Arcturus, SE/SA layout is changed to 8*1.
+             * To mostly reduce the impact, we make it compatible
+             * with current bitmap array as below:
+             *    SE4 --> cu_bitmap[0][1]
+             *    SE5 --> cu_bitmap[1][1]
+             *    SE6 --> cu_bitmap[2][1]
+             *    SE7 --> cu_bitmap[3][1]
+             */
+            assert(info->max_sa_per_se == 1);
+            info->cu_mask[i][0] = amdinfo->cu_bitmap[i % 4][i / 4];
+         } else {
+            info->cu_mask[i][j] = amdinfo->cu_bitmap[i][j];
+         }
          info->num_good_compute_units += util_bitcount(info->cu_mask[i][j]);
       }
    }
@@ -1247,6 +1251,12 @@ void ac_print_gpu_info(struct radeon_info *info, FILE *f)
    fprintf(f, "    has_tmz_support = %u\n", info->has_tmz_support);
 
    fprintf(f, "Shader core info:\n");
+   for (unsigned i = 0; i < info->max_se; i++) {
+      for (unsigned j = 0; j < info->max_sa_per_se; j++) {
+         fprintf(f, "    cu_mask[SE%u][SA%u] = 0x%x \t(%u)\n",
+                 i, j, info->cu_mask[i][j], util_bitcount(info->cu_mask[i][j]));
+      }
+   }
    fprintf(f, "    max_shader_clock = %i\n", info->max_shader_clock);
    fprintf(f, "    num_good_compute_units = %i\n", info->num_good_compute_units);
    fprintf(f, "    max_good_cu_per_sa = %i\n", info->max_good_cu_per_sa);

diff --git a/src/amd/common/ac_gpu_info.h b/src/amd/common/ac_gpu_info.h
index e31b670..fd89af0 100644 (file)
--- a/src/amd/common/ac_gpu_info.h
+++ b/src/amd/common/ac_gpu_info.h
@@ -37,6 +37,9 @@
 extern "C" {
 #endif
 
+#define AMD_MAX_SE         8
+#define AMD_MAX_SA_PER_SE  2
+
 struct amdgpu_gpu_info;
 
 struct radeon_info {
@@ -190,7 +193,7 @@ struct radeon_info {
    bool kernel_has_modifiers;
 
    /* Shader cores. */
-   uint32_t cu_mask[4][2];
+   uint32_t cu_mask[AMD_MAX_SE][AMD_MAX_SA_PER_SE];
    uint32_t r600_max_quad_pipes; /* wave size / 16 */
    uint32_t max_shader_clock;
    uint32_t num_good_compute_units;

diff --git a/src/amd/common/ac_rgp.c b/src/amd/common/ac_rgp.c
index 3cd81b4..5b8be63 100644 (file)
--- a/src/amd/common/ac_rgp.c
+++ b/src/amd/common/ac_rgp.c
@@ -512,8 +512,8 @@ static void ac_sqtt_fill_asic_info(struct radeon_info *rad_info,
    chunk->memory_chip_type = ac_vram_type_to_sqtt_memory_type(rad_info->vram_type);
    chunk->lds_granularity = rad_info->lds_encode_granularity;
 
-   for (unsigned se = 0; se < 4; se++) {
-      for (unsigned sa = 0; sa < 2; sa++) {
+   for (unsigned se = 0; se < AMD_MAX_SE; se++) {
+      for (unsigned sa = 0; sa < AMD_MAX_SA_PER_SE; sa++) {
          chunk->cu_mask[se][sa] = rad_info->cu_mask[se][sa];
       }
    }