drm/amdgpu: implement raster configuration for gfx v6
authorHuang Rui <ray.huang@amd.com>
Fri, 9 Sep 2016 08:37:08 +0000 (16:37 +0800)
committerAlex Deucher <alexander.deucher@amd.com>
Mon, 19 Sep 2016 18:38:24 +0000 (14:38 -0400)
This patch is to implement the raster configuration and harvested
configuration of gfx v6.

Signed-off-by: Huang Rui <ray.huang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c
drivers/gpu/drm/amd/include/asic_reg/si/sid.h

index 410b29c05671809f11da50eda1642c1528a536d4..40abb6b81c095a677dd6f6992369d5db8f2ec4aa 100644 (file)
@@ -931,6 +931,123 @@ static u32 gfx_v6_0_get_rb_disabled(struct amdgpu_device *adev,
        return data & mask;
 }
 
+static void gfx_v6_0_raster_config(struct amdgpu_device *adev, u32 *rconf)
+{
+       switch (adev->asic_type) {
+       case CHIP_TAHITI:
+       case CHIP_PITCAIRN:
+               *rconf |= RB_XSEL2(2) | RB_XSEL | PKR_MAP(2) | PKR_YSEL(1) |
+                         SE_MAP(2) | SE_XSEL(2) | SE_YSEL(2);
+               break;
+       case CHIP_VERDE:
+               *rconf |= RB_XSEL | PKR_MAP(2) | PKR_YSEL(1);
+               break;
+       case CHIP_OLAND:
+               *rconf |= RB_YSEL;
+               break;
+       case CHIP_HAINAN:
+               *rconf |= 0x0;
+               break;
+       default:
+               DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
+               break;
+       }
+}
+
+static void gfx_v6_0_write_harvested_raster_configs(struct amdgpu_device *adev,
+                                                   u32 raster_config, unsigned rb_mask,
+                                                   unsigned num_rb)
+{
+       unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
+       unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
+       unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
+       unsigned rb_per_se = num_rb / num_se;
+       unsigned se_mask[4];
+       unsigned se;
+
+       se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
+       se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
+       se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
+       se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
+
+       WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
+       WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
+       WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
+
+       for (se = 0; se < num_se; se++) {
+               unsigned raster_config_se = raster_config;
+               unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
+               unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
+               int idx = (se / 2) * 2;
+
+               if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
+                       raster_config_se &= ~SE_MAP_MASK;
+
+                       if (!se_mask[idx]) {
+                               raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
+                       } else {
+                               raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
+                       }
+               }
+
+               pkr0_mask &= rb_mask;
+               pkr1_mask &= rb_mask;
+               if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
+                       raster_config_se &= ~PKR_MAP_MASK;
+
+                       if (!pkr0_mask) {
+                               raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
+                       } else {
+                               raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
+                       }
+               }
+
+               if (rb_per_se >= 2) {
+                       unsigned rb0_mask = 1 << (se * rb_per_se);
+                       unsigned rb1_mask = rb0_mask << 1;
+
+                       rb0_mask &= rb_mask;
+                       rb1_mask &= rb_mask;
+                       if (!rb0_mask || !rb1_mask) {
+                               raster_config_se &= ~RB_MAP_PKR0_MASK;
+
+                               if (!rb0_mask) {
+                                       raster_config_se |=
+                                               RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
+                               } else {
+                                       raster_config_se |=
+                                               RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
+                               }
+                       }
+
+                       if (rb_per_se > 2) {
+                               rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
+                               rb1_mask = rb0_mask << 1;
+                               rb0_mask &= rb_mask;
+                               rb1_mask &= rb_mask;
+                               if (!rb0_mask || !rb1_mask) {
+                                       raster_config_se &= ~RB_MAP_PKR1_MASK;
+
+                                       if (!rb0_mask) {
+                                               raster_config_se |=
+                                                       RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
+                                       } else {
+                                               raster_config_se |=
+                                                       RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
+                                       }
+                               }
+                       }
+               }
+
+               /* GRBM_GFX_INDEX has a different offset on SI */
+               gfx_v6_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
+               WREG32(PA_SC_RASTER_CONFIG, raster_config_se);
+       }
+
+       /* GRBM_GFX_INDEX has a different offset on SI */
+       gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
+}
+
 static void gfx_v6_0_setup_rb(struct amdgpu_device *adev,
                              u32 se_num, u32 sh_per_se,
                              u32 max_rb_num_per_se)
@@ -939,6 +1056,7 @@ static void gfx_v6_0_setup_rb(struct amdgpu_device *adev,
        u32 data, mask;
        u32 disabled_rbs = 0;
        u32 enabled_rbs = 0;
+       unsigned num_rb_pipes;
 
        mutex_lock(&adev->grbm_idx_mutex);
        for (i = 0; i < se_num; i++) {
@@ -961,6 +1079,9 @@ static void gfx_v6_0_setup_rb(struct amdgpu_device *adev,
        adev->gfx.config.backend_enable_mask = enabled_rbs;
        adev->gfx.config.num_rbs = hweight32(enabled_rbs);
 
+       num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
+                            adev->gfx.config.max_shader_engines, 16);
+
        mutex_lock(&adev->grbm_idx_mutex);
        for (i = 0; i < se_num; i++) {
                gfx_v6_0_select_se_sh(adev, i, 0xffffffff, 0xffffffff);
@@ -980,7 +1101,15 @@ static void gfx_v6_0_setup_rb(struct amdgpu_device *adev,
                        }
                        enabled_rbs >>= 2;
                }
-               WREG32(PA_SC_RASTER_CONFIG, data);
+               gfx_v6_0_raster_config(adev, &data);
+
+               if (!adev->gfx.config.backend_enable_mask ||
+                               adev->gfx.config.num_rbs >= num_rb_pipes)
+                       WREG32(PA_SC_RASTER_CONFIG, data);
+               else
+                       gfx_v6_0_write_harvested_raster_configs(adev, data,
+                                                               adev->gfx.config.backend_enable_mask,
+                                                               num_rb_pipes);
        }
        gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
        mutex_unlock(&adev->grbm_idx_mutex);
index 8c5608a4d52623c74d9705ba70b028c5cf8b75b8..c57eff159374ceb7b864235fc63929be4d1227cf 100644 (file)
 #define DB_DEPTH_INFO                                   0xA00F
 
 #define PA_SC_RASTER_CONFIG                             0xA0D4
+#      define RB_MAP_PKR0(x)                           ((x) << 0)
+#      define RB_MAP_PKR0_MASK                         (0x3 << 0)
+#      define RB_MAP_PKR1(x)                           ((x) << 2)
+#      define RB_MAP_PKR1_MASK                         (0x3 << 2)
 #       define RASTER_CONFIG_RB_MAP_0                   0
 #       define RASTER_CONFIG_RB_MAP_1                   1
 #       define RASTER_CONFIG_RB_MAP_2                   2
 #       define RASTER_CONFIG_RB_MAP_3                   3
+#      define RB_XSEL2(x)                              ((x) << 4)
+#      define RB_XSEL2_MASK                            (0x3 << 4)
+#      define RB_XSEL                                  (1 << 6)
+#      define RB_YSEL                                  (1 << 7)
+#      define PKR_MAP(x)                               ((x) << 8)
+#      define PKR_MAP_MASK                             (0x3 << 8)
+#       define RASTER_CONFIG_PKR_MAP_0                 0
+#       define RASTER_CONFIG_PKR_MAP_1                 1
+#       define RASTER_CONFIG_PKR_MAP_2                 2
+#       define RASTER_CONFIG_PKR_MAP_3                 3
+#      define PKR_XSEL(x)                              ((x) << 10)
+#      define PKR_XSEL_MASK                            (0x3 << 10)
+#      define PKR_YSEL(x)                              ((x) << 12)
+#      define PKR_YSEL_MASK                            (0x3 << 12)
+#      define SC_MAP(x)                                ((x) << 16)
+#      define SC_MAP_MASK                              (0x3 << 16)
+#      define SC_XSEL(x)                               ((x) << 18)
+#      define SC_XSEL_MASK                             (0x3 << 18)
+#      define SC_YSEL(x)                               ((x) << 20)
+#      define SC_YSEL_MASK                             (0x3 << 20)
+#      define SE_MAP(x)                                ((x) << 24)
+#      define SE_MAP_MASK                              (0x3 << 24)
+#       define RASTER_CONFIG_SE_MAP_0                  0
+#       define RASTER_CONFIG_SE_MAP_1                  1
+#       define RASTER_CONFIG_SE_MAP_2                  2
+#       define RASTER_CONFIG_SE_MAP_3                  3
+#      define SE_XSEL(x)                               ((x) << 26)
+#      define SE_XSEL_MASK                             (0x3 << 26)
+#      define SE_YSEL(x)                               ((x) << 28)
+#      define SE_YSEL_MASK                             (0x3 << 28)
+
 
 #define VGT_EVENT_INITIATOR                             0xA2A4
 #       define SAMPLE_STREAMOUTSTATS1                   (1 << 0)