nvk: Use a contiguous shader heap pre-Volta

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24326>

src/nouveau/vulkan/nvk_device.c

@@ -180,7 +180,8 @@ nvk_CreateDevice(VkPhysicalDevice physicalDevice,
     */
    result = nvk_heap_init(device, &device->shader_heap,
                           NOUVEAU_WS_BO_LOCAL, NOUVEAU_WS_BO_WR,
-                          4096 /* overalloc */, false /* contiguous */);
+                          4096 /* overalloc */,
+                          device->ctx->eng3d.cls < VOLTA_A);
    if (result != VK_SUCCESS)
       goto fail_samplers;
 

src/nouveau/vulkan/nvk_queue.c

@@ -28,6 +28,8 @@ nvk_queue_state_finish(struct nvk_device *dev,
       nouveau_ws_bo_destroy(qs->images.bo);
    if (qs->samplers.bo)
       nouveau_ws_bo_destroy(qs->samplers.bo);
+   if (qs->shaders.bo)
+      nouveau_ws_bo_destroy(qs->shaders.bo);
    if (qs->slm.bo)
       nouveau_ws_bo_destroy(qs->slm.bo);
    if (qs->push.bo) {
@@ -81,6 +83,19 @@ nvk_queue_state_update(struct nvk_device *dev,
          nouveau_ws_bo_destroy(bo);
    }
 
+   if (dev->shader_heap.contiguous) {
+      bo = nvk_heap_get_contiguous_bo_ref(&dev->shader_heap);
+      if (qs->shaders.bo != bo) {
+         if (qs->shaders.bo)
+            nouveau_ws_bo_destroy(qs->shaders.bo);
+         qs->shaders.bo = bo;
+         dirty = true;
+      } else {
+         if (bo)
+            nouveau_ws_bo_destroy(bo);
+      }
+   }
+
    bo = nvk_slm_area_get_bo_ref(&dev->slm, &bytes_per_warp, &bytes_per_mp);
    if (qs->slm.bo != bo || qs->slm.bytes_per_warp != bytes_per_warp ||
        qs->slm.bytes_per_mp != bytes_per_mp) {
@@ -158,6 +173,20 @@ nvk_queue_state_update(struct nvk_device *dev,
       });
    }
 
+   if (qs->shaders.bo) {
+      /* Compute */
+      assert(dev->ctx->compute.cls < VOLTA_COMPUTE_A);
+      P_MTHD(p, NVA0C0, SET_PROGRAM_REGION_A);
+      P_NVA0C0_SET_PROGRAM_REGION_A(p, qs->shaders.bo->offset >> 32);
+      P_NVA0C0_SET_PROGRAM_REGION_B(p, qs->shaders.bo->offset);
+
+      /* 3D */
+      assert(dev->ctx->eng3d.cls < VOLTA_A);
+      P_MTHD(p, NV9097, SET_PROGRAM_REGION_A);
+      P_NV9097_SET_PROGRAM_REGION_A(p, qs->shaders.bo->offset >> 32);
+      P_NV9097_SET_PROGRAM_REGION_B(p, qs->shaders.bo->offset);
+   }
+
    if (qs->slm.bo) {
       const uint64_t slm_addr = qs->slm.bo->offset;
       const uint64_t slm_size = qs->slm.bo->size;

src/nouveau/vulkan/nvk_queue.h

@@ -20,6 +20,10 @@ struct nvk_queue_state {
       uint32_t alloc_count;
    } samplers;
 
+   struct {
+      struct nouveau_ws_bo *bo;
+   } shaders;
+
    struct {
       struct nouveau_ws_bo *bo;
       uint32_t bytes_per_warp;