swr: Enable MSAA in OpenSWR software renderer

This patch enables multisample antialiasing in the OpenSWR software renderer.

MSAA is a proof-of-concept/work-in-progress with bug fixes and performance
on the way.  We wanted to get the changes out now to allow several customers
to begin experimenting with MSAA in a software renderer.  So as not to
impact current customers, MSAA is turned off by default - previous
functionality and performance remain intact.  It is easily enabled via
environment variables, as described below.

It has only been tested with the glx-lib winsys.  The intention is to
enable other state-trackers, both Windows and Linux and more fully support
FBOs.

There are 2 environment variables that affect behavior:

* SWR_MSAA_FORCE_ENABLE - force MSAA on, for apps that are not designed
  for MSAA... Beware, results will vary.  This is mainly for testing.

* SWR_MSAA_MAX_SAMPLE_COUNT - sets maximum supported number of
  samples (1,2,4,8,16), or 0 to disable MSAA altogether.
  (The default is currently 0.)

Reviewed-by: George Kyriazis <[email protected]>

6 files changed, 313 insertions, 25 deletions

diff --git a/src/gallium/drivers/swr/swr_context.cpp b/src/gallium/drivers/swr/swr_context.cpp
index 6f46d666ac5..aa5cca8e653 100644
--- a/src/gallium/drivers/swr/swr_context.cpp
+++ b/src/gallium/drivers/swr/swr_context.cpp
@@ -267,20 +267,104 @@ swr_resource_copy(struct pipe_context *pipe,
 }
 
 
+/* XXX: This resolve is incomplete and suboptimal. It will be removed once the
+ * pipelined resolve blit works. */
+void
+swr_do_msaa_resolve(struct pipe_resource *src_resource,
+                    struct pipe_resource *dst_resource)
+{
+   /* This is a pretty dumb inline resolve.  It only supports 8-bit formats
+    * (ex RGBA8/BGRA8) - which are most common display formats anyway.
+    */
+
+   /* quick check for 8-bit and number of components */
+   uint8_t bits_per_component =
+      util_format_get_component_bits(src_resource->format,
+            UTIL_FORMAT_COLORSPACE_RGB, 0);
+
+   /* Unsupported resolve format */
+   assert(src_resource->format == dst_resource->format);
+   assert(bits_per_component == 8);
+   if ((src_resource->format != dst_resource->format) ||
+       (bits_per_component != 8)) {
+      return;
+   }
+
+   uint8_t src_num_comps = util_format_get_nr_components(src_resource->format);
+
+   SWR_SURFACE_STATE *src_surface = &swr_resource(src_resource)->swr;
+   SWR_SURFACE_STATE *dst_surface = &swr_resource(dst_resource)->swr;
+
+   uint32_t *src, *dst, offset;
+   uint32_t num_samples = src_surface->numSamples;
+   float recip_num_samples = 1.0f / num_samples;
+   for (uint32_t y = 0; y < src_surface->height; y++) {
+      for (uint32_t x = 0; x < src_surface->width; x++) {
+         float r = 0.0f;
+         float g = 0.0f;
+         float b = 0.0f;
+         float a = 0.0f;
+         for (uint32_t sampleNum = 0;  sampleNum < num_samples; sampleNum++) {
+            offset = ComputeSurfaceOffset<false>(x, y, 0, 0, sampleNum, 0, src_surface);
+            src = (uint32_t *) src_surface->pBaseAddress + offset/src_num_comps;
+            const uint32_t sample = *src;
+            r += (float)((sample >> 24) & 0xff) / 255.0f * recip_num_samples;
+            g += (float)((sample >> 16) & 0xff) / 255.0f * recip_num_samples;
+            b += (float)((sample >>  8) & 0xff) / 255.0f * recip_num_samples;
+            a += (float)((sample      ) & 0xff) / 255.0f * recip_num_samples;
+         }
+         uint32_t result = 0;
+         result  = ((uint8_t)(r * 255.0f) & 0xff) << 24;
+         result |= ((uint8_t)(g * 255.0f) & 0xff) << 16;
+         result |= ((uint8_t)(b * 255.0f) & 0xff) <<  8;
+         result |= ((uint8_t)(a * 255.0f) & 0xff);
+         offset = ComputeSurfaceOffset<false>(x, y, 0, 0, 0, 0, src_surface);
+         dst = (uint32_t *) dst_surface->pBaseAddress + offset/src_num_comps;
+         *dst = result;
+      }
+   }
+}
+
+
 static void
 swr_blit(struct pipe_context *pipe, const struct pipe_blit_info *blit_info)
 {
    struct swr_context *ctx = swr_context(pipe);
+   /* Make a copy of the const blit_info, so we can modify it */
    struct pipe_blit_info info = *blit_info;
 
-   if (blit_info->render_condition_enable && !swr_check_render_cond(pipe))
+   if (info.render_condition_enable && !swr_check_render_cond(pipe))
       return;
 
    if (info.src.resource->nr_samples > 1 && info.dst.resource->nr_samples <= 1
        && !util_format_is_depth_or_stencil(info.src.resource->format)
        && !util_format_is_pure_integer(info.src.resource->format)) {
-      debug_printf("swr: color resolve unimplemented\n");
-      return;
+      debug_printf("swr_blit: color resolve : %d -> %d\n",
+            info.src.resource->nr_samples, info.dst.resource->nr_samples);
+
+      /* Because the resolve is being done inline (not pipelined),
+       * resources need to be stored out of hottiles and the pipeline empty.
+       *
+       * Resources are marked unused following fence finish because all
+       * pipeline operations are complete.  Validation of the blit will mark
+       * them are read/write again.
+       */
+      swr_store_dirty_resource(pipe, info.src.resource, SWR_TILE_RESOLVED);
+      swr_store_dirty_resource(pipe, info.dst.resource, SWR_TILE_RESOLVED);
+      swr_fence_finish(pipe->screen, NULL, swr_screen(pipe->screen)->flush_fence, 0);
+      swr_resource_unused(info.src.resource);
+      swr_resource_unused(info.dst.resource);
+
+      struct pipe_resource *src_resource = info.src.resource;
+      struct pipe_resource *resolve_target =
+         swr_resource(src_resource)->resolve_target;
+
+      /* Inline resolve samples into resolve target resource, then continue
+       * the blit. */
+      swr_do_msaa_resolve(src_resource, resolve_target);
+
+      /* The resolve target becomes the new source for the blit.  */
+      info.src.resource = resolve_target;
    }
 
    if (util_try_blit_via_copy_region(pipe, &info)) {
diff --git a/src/gallium/drivers/swr/swr_context.h b/src/gallium/drivers/swr/swr_context.h
index be65a20c85b..233d95ab167 100644
--- a/src/gallium/drivers/swr/swr_context.h
+++ b/src/gallium/drivers/swr/swr_context.h
@@ -196,4 +196,7 @@ void swr_clear_init(struct pipe_context *pipe);
 void swr_draw_init(struct pipe_context *pipe);
 
 void swr_finish(struct pipe_context *pipe);
+
+void swr_do_msaa_resolve(struct pipe_resource *src_resource,
+                         struct pipe_resource *dst_resource);
 #endif
diff --git a/src/gallium/drivers/swr/swr_resource.h b/src/gallium/drivers/swr/swr_resource.h
index 41abd77cba0..ae9954c1e7d 100644
--- a/src/gallium/drivers/swr/swr_resource.h
+++ b/src/gallium/drivers/swr/swr_resource.h
@@ -46,6 +46,10 @@ struct swr_resource {
 
    struct sw_displaytarget *display_target;
 
+   /* If resource is multisample, then this points to a alternate resource
+    * containing the resolved multisample surface, otherwise null */
+   struct pipe_resource *resolve_target;
+
    size_t mip_offsets[PIPE_MAX_TEXTURE_LEVELS];
    size_t secondary_mip_offsets[PIPE_MAX_TEXTURE_LEVELS];
 
diff --git a/src/gallium/drivers/swr/swr_screen.cpp b/src/gallium/drivers/swr/swr_screen.cpp
index 87fd898eb11..f88989b0e21 100644
--- a/src/gallium/drivers/swr/swr_screen.cpp
+++ b/src/gallium/drivers/swr/swr_screen.cpp
@@ -61,6 +61,11 @@
 #define SWR_MAX_TEXTURE_CUBE_LEVELS 14  /* 8K x 8K for now */
 #define SWR_MAX_TEXTURE_ARRAY_LAYERS 512 /* 8K x 512 / 8K x 8K x 512 */
 
+/* Flag indicates creation of alternate surface, to prevent recursive loop
+ * in resource creation when msaa_force_enable is set. */
+#define SWR_RESOURCE_FLAG_ALT_SURFACE (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
+
+
 static const char *
 swr_get_name(struct pipe_screen *screen)
 {
@@ -78,13 +83,14 @@ swr_get_vendor(struct pipe_screen *screen)
 }
 
 static boolean
-swr_is_format_supported(struct pipe_screen *screen,
+swr_is_format_supported(struct pipe_screen *_screen,
                         enum pipe_format format,
                         enum pipe_texture_target target,
                         unsigned sample_count,
                         unsigned bind)
 {
-   struct sw_winsys *winsys = swr_screen(screen)->winsys;
+   struct swr_screen *screen = swr_screen(_screen);
+   struct sw_winsys *winsys = screen->winsys;
    const struct util_format_description *format_desc;
 
    assert(target == PIPE_BUFFER || target == PIPE_TEXTURE_1D
@@ -100,7 +106,8 @@ swr_is_format_supported(struct pipe_screen *screen,
    if (!format_desc)
       return FALSE;
 
-   if (sample_count > 1)
+   if ((sample_count > screen->msaa_max_count)
+      || !util_is_power_of_two(sample_count))
       return FALSE;
 
    if (bind & PIPE_BIND_DISPLAY_TARGET) {
@@ -235,7 +242,6 @@ swr_get_param(struct pipe_screen *screen, enum pipe_cap param)
    case PIPE_CAP_QUERY_TIMESTAMP:
    case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
    case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
-   case PIPE_CAP_FAKE_SW_MSAA:
    case PIPE_CAP_DRAW_INDIRECT:
    case PIPE_CAP_UMA:
    case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
@@ -250,6 +256,15 @@ swr_get_param(struct pipe_screen *screen, enum pipe_cap param)
    case PIPE_CAP_DOUBLES:
       return 1;
 
+   /* MSAA support
+    * If user has explicitly set max_sample_count = 0 (via SWR_MSAA_MAX_COUNT)
+    * then disable all MSAA support and go back to old caps. */
+   case PIPE_CAP_TEXTURE_MULTISAMPLE:
+   case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
+      return swr_screen(screen)->msaa_max_count ? 1 : 0;
+   case PIPE_CAP_FAKE_SW_MSAA:
+      return swr_screen(screen)->msaa_max_count ? 0 : 1;
+
       /* unsupported features */
    case PIPE_CAP_ANISOTROPIC_FILTER:
    case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
@@ -264,7 +279,6 @@ swr_get_param(struct pipe_screen *screen, enum pipe_cap param)
    case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
    case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
    case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
-   case PIPE_CAP_TEXTURE_MULTISAMPLE:
    case PIPE_CAP_TGSI_TEXCOORD:
    case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
    case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
@@ -276,7 +290,6 @@ swr_get_param(struct pipe_screen *screen, enum pipe_cap param)
    case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
    case PIPE_CAP_SAMPLER_VIEW_TARGET:
    case PIPE_CAP_VERTEXID_NOBASE:
-   case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
    case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
@@ -712,6 +725,14 @@ swr_texture_layout(struct swr_screen *screen,
    if (pt->bind & (PIPE_BIND_RENDER_TARGET | PIPE_BIND_DEPTH_STENCIL)) {
       res->swr.halign = KNOB_MACROTILE_X_DIM;
       res->swr.valign = KNOB_MACROTILE_Y_DIM;
+
+      /* If SWR_MSAA_FORCE_ENABLE is set, turn on MSAA and override requested
+       * surface sample count. */
+      if (screen->msaa_force_enable) {
+         res->swr.numSamples = screen->msaa_max_count;
+         fprintf(stderr,"swr_texture_layout: forcing sample count: %d\n",
+                 res->swr.numSamples);
+      }
    } else {
       res->swr.halign = 1;
       res->swr.valign = 1;
@@ -790,8 +811,8 @@ swr_texture_layout(struct swr_screen *screen,
          ComputeSurfaceOffset<false>(0, 0, 0, 0, 0, level, &res->swr);
    }
 
-   size_t total_size =
-      (size_t)res->swr.depth * res->swr.qpitch * res->swr.pitch;
+   size_t total_size = res->swr.depth * res->swr.qpitch * res->swr.pitch *
+                       res->swr.numSamples;
    if (total_size > SWR_MAX_TEXTURE_SIZE)
       return false;
 
@@ -808,9 +829,11 @@ swr_texture_layout(struct swr_screen *screen,
                ComputeSurfaceOffset<false>(0, 0, 0, 0, 0, level, &res->secondary);
          }
 
-         res->secondary.pBaseAddress = (uint8_t *)AlignedMalloc(
-            res->secondary.depth * res->secondary.qpitch *
-            res->secondary.pitch, 64);
+         total_size = res->secondary.depth * res->secondary.qpitch *
+                      res->secondary.pitch * res->secondary.numSamples;
+
+         res->secondary.pBaseAddress = (uint8_t *) AlignedMalloc(total_size,
+                                                                 64);
       }
    }
 
@@ -827,6 +850,51 @@ swr_can_create_resource(struct pipe_screen *screen,
    return swr_texture_layout(swr_screen(screen), &res, false);
 }
 
+/* Helper function that conditionally creates a single-sample resolve resource
+ * and attaches it to main multisample resource. */
+static boolean
+swr_create_resolve_resource(struct pipe_screen *_screen,
+                            struct swr_resource *msaa_res)
+{
+   struct swr_screen *screen = swr_screen(_screen);
+
+   /* If resource is multisample, create a single-sample resolve resource */
+   if (msaa_res->base.nr_samples > 1 || (screen->msaa_force_enable &&
+            !(msaa_res->base.flags & SWR_RESOURCE_FLAG_ALT_SURFACE))) {
+
+      /* Create a single-sample copy of the resource.  Copy the original
+       * resource parameters and set flag to prevent recursion when re-calling
+       * resource_create */
+      struct pipe_resource alt_template = msaa_res->base;
+      alt_template.nr_samples = 0;
+      alt_template.flags |= SWR_RESOURCE_FLAG_ALT_SURFACE;
+
+      /* Note: Display_target is a special single-sample resource, only the
+       * display_target has been created already. */
+      if (msaa_res->base.bind & (PIPE_BIND_DISPLAY_TARGET | PIPE_BIND_SCANOUT
+               | PIPE_BIND_SHARED)) {
+         /* Allocate the multisample buffers. */
+         if (!swr_texture_layout(screen, msaa_res, true))
+            return false;
+
+         /* Alt resource will only be bound as PIPE_BIND_RENDER_TARGET
+          * remove the DISPLAY_TARGET, SCANOUT, and SHARED bindings */
+         alt_template.bind = PIPE_BIND_RENDER_TARGET;
+      }
+
+      /* Allocate single-sample resolve surface */
+      struct pipe_resource *alt;
+      alt = _screen->resource_create(_screen, &alt_template);
+      if (!alt)
+         return false;
+
+      /* Attach it to the multisample resource */
+      msaa_res->resolve_target = alt;
+   }
+
+   return true; /* success */
+}
+
 static struct pipe_resource *
 swr_resource_create(struct pipe_screen *_screen,
                     const struct pipe_resource *templat)
@@ -845,7 +913,7 @@ swr_resource_create(struct pipe_screen *_screen,
                             | PIPE_BIND_SHARED)) {
          /* displayable surface
           * first call swr_texture_layout without allocating to finish
-          * filling out the SWR_SURFAE_STATE in res */
+          * filling out the SWR_SURFACE_STATE in res */
          swr_texture_layout(screen, res, false);
          if (!swr_displaytarget_layout(screen, res))
             goto fail;
@@ -854,6 +922,12 @@ swr_resource_create(struct pipe_screen *_screen,
          if (!swr_texture_layout(screen, res, true))
             goto fail;
       }
+
+      /* If resource was multisample, create resolve resource and attach
+       * it to multisample resource. */
+      if (!swr_create_resolve_resource(_screen, res))
+            goto fail;
+
    } else {
       /* other data (vertex buffer, const buffer, etc) */
       assert(util_format_get_blocksize(templat->format) == 1);
@@ -862,7 +936,7 @@ swr_resource_create(struct pipe_screen *_screen,
       assert(templat->last_level == 0);
 
       /* Easiest to just call swr_texture_layout, as it sets up
-       * SWR_SURFAE_STATE in res */
+       * SWR_SURFACE_STATE in res */
       if (!swr_texture_layout(screen, res, true))
          goto fail;
    }
@@ -888,9 +962,24 @@ swr_resource_destroy(struct pipe_screen *p_screen, struct pipe_resource *pt)
       struct sw_winsys *winsys = screen->winsys;
       winsys->displaytarget_destroy(winsys, spr->display_target);
 
+      if (spr->swr.numSamples > 1) {
+         /* Free an attached resolve resource */
+         struct swr_resource *alt = swr_resource(spr->resolve_target);
+         swr_fence_work_free(screen->flush_fence, alt->swr.pBaseAddress, true);
+
+         /* Free multisample buffer */
+         swr_fence_work_free(screen->flush_fence, spr->swr.pBaseAddress, true);
+      }
    } else {
       /* For regular resources, defer deletion */
       swr_resource_unused(pt);
+
+      if (spr->swr.numSamples > 1) {
+         /* Free an attached resolve resource */
+         struct swr_resource *alt = swr_resource(spr->resolve_target);
+         swr_fence_work_free(screen->flush_fence, alt->swr.pBaseAddress, true);
+      }
+
       swr_fence_work_free(screen->flush_fence, spr->swr.pBaseAddress, true);
       swr_fence_work_free(screen->flush_fence,
                           spr->secondary.pBaseAddress, true);
@@ -919,6 +1008,23 @@ swr_flush_frontbuffer(struct pipe_screen *p_screen,
       SwrEndFrame(swr_context(pipe)->swrContext);
    }
 
+   /* Multisample surfaces need to be resolved before present */
+   if (pipe && spr->swr.numSamples > 1) {
+      struct pipe_resource *resolve_target = spr->resolve_target;
+
+      /* Do an inline surface resolve into the resolve target resource
+       * XXX: This works, just not optimal. Work on using a pipelined blit. */
+      swr_do_msaa_resolve(resource, resolve_target);
+
+      /* Once resolved, copy into display target */
+      SWR_SURFACE_STATE *resolve = &swr_resource(resolve_target)->swr;
+
+      void *map = winsys->displaytarget_map(winsys, spr->display_target,
+                                            PIPE_TRANSFER_WRITE);
+      memcpy(map, resolve->pBaseAddress, resolve->pitch * resolve->height);
+      winsys->displaytarget_unmap(winsys, spr->display_target);
+   }
+
    debug_assert(spr->display_target);
    if (spr->display_target)
       winsys->displaytarget_display(
@@ -986,6 +1092,33 @@ swr_create_screen_internal(struct sw_winsys *winsys)
 
    util_format_s3tc_init();
 
+   /* XXX msaa under development, disable by default for now */
+   screen->msaa_max_count = 0; /* was SWR_MAX_NUM_MULTISAMPLES; */
+
+   /* validate env override values, within range and power of 2 */
+   int msaa_max_count = debug_get_num_option("SWR_MSAA_MAX_COUNT", 0);
+   if (msaa_max_count) {
+      if ((msaa_max_count < 0) || (msaa_max_count > SWR_MAX_NUM_MULTISAMPLES)
+            || !util_is_power_of_two(msaa_max_count)) {
+         fprintf(stderr, "SWR_MSAA_MAX_COUNT invalid: %d\n", msaa_max_count);
+         fprintf(stderr, "must be power of 2 between 1 and %d" \
+                         " (or 0 to disable msaa)\n",
+               SWR_MAX_NUM_MULTISAMPLES);
+         msaa_max_count = 0;
+      }
+
+      fprintf(stderr, "SWR_MSAA_MAX_COUNT: %d\n", msaa_max_count);
+      if (!msaa_max_count)
+         fprintf(stderr, "(msaa disabled)\n");
+
+      screen->msaa_max_count = msaa_max_count;
+   }
+
+   screen->msaa_force_enable = debug_get_bool_option(
+         "SWR_MSAA_FORCE_ENABLE", false);
+   if (screen->msaa_force_enable)
+      fprintf(stderr, "SWR_MSAA_FORCE_ENABLE: true\n");
+
    return &screen->base;
 }
 
diff --git a/src/gallium/drivers/swr/swr_screen.h b/src/gallium/drivers/swr/swr_screen.h
index 0c82a2eff7a..dc1bb47f02d 100644
--- a/src/gallium/drivers/swr/swr_screen.h
+++ b/src/gallium/drivers/swr/swr_screen.h
@@ -24,10 +24,15 @@
 #ifndef SWR_SCREEN_H
 #define SWR_SCREEN_H
 
+#include "swr_resource.h"
+
 #include "pipe/p_screen.h"
 #include "pipe/p_defines.h"
+#include "util/u_format.h"
 #include "api.h"
 
+#include "memory/TilingFunctions.h"
+
 struct sw_winsys;
 
 struct swr_screen {
@@ -38,6 +43,9 @@ struct swr_screen {
 
    struct sw_winsys *winsys;
 
+   boolean msaa_force_enable;
+   uint8_t msaa_max_count;
+
    HANDLE hJitMgr;
 };
 
diff --git a/src/gallium/drivers/swr/swr_state.cpp b/src/gallium/drivers/swr/swr_state.cpp
index 5cc01ddcab0..aa9fe099f0a 100644
--- a/src/gallium/drivers/swr/swr_state.cpp
+++ b/src/gallium/drivers/swr/swr_state.cpp
@@ -30,6 +30,7 @@
 #include "common/os.h"
 #include "jit_api.h"
 #include "gen_state_llvm.h"
+#include "core/multisample.h"
 
 #include "gallivm/lp_bld_tgsi.h"
 #include "util/u_format.h"
@@ -668,6 +669,9 @@ swr_set_framebuffer_state(struct pipe_context *pipe,
    if (changed) {
       util_copy_framebuffer_state(&ctx->framebuffer, fb);
 
+      /* 0 and 1 both indicate no msaa.  Core doesn't understand 0 samples */
+      ctx->framebuffer.samples = std::max((ubyte)1, ctx->framebuffer.samples);
+
       ctx->dirty |= SWR_NEW_FRAMEBUFFER;
    }
 }
@@ -685,6 +689,36 @@ swr_set_sample_mask(struct pipe_context *pipe, unsigned sample_mask)
 }
 
 /*
+ * MSAA fixed sample position table
+ * used by update_derived and get_sample_position
+ * (integer locations on a 16x16 grid)
+ */
+static const uint8_t swr_sample_positions[][2] =
+{ /* 1x*/ { 8, 8},
+  /* 2x*/ {12,12},{ 4, 4},
+  /* 4x*/ { 6, 2},{14, 6},{ 2,10},{10,14},
+  /* 8x*/ { 9, 5},{ 7,11},{13, 9},{ 5, 3},
+          { 3,13},{ 1, 7},{11,15},{15, 1},
+  /*16x*/ { 9, 9},{ 7, 5},{ 5,10},{12, 7},
+          { 3, 6},{10,13},{13,11},{11, 3},
+          { 6,14},{ 8, 1},{ 4, 2},{ 2,12},
+          { 0, 8},{15, 4},{14,15},{ 1, 0} };
+
+static void
+swr_get_sample_position(struct pipe_context *pipe,
+                        unsigned sample_count, unsigned sample_index,
+                        float *out_value)
+{
+   /* validate sample_count */
+   sample_count = GetNumSamples(GetSampleCount(sample_count));
+
+   const uint8_t *sample = swr_sample_positions[sample_count-1 + sample_index];
+   out_value[0] = sample[0] / 16.0f;
+   out_value[1] = sample[1] / 16.0f;
+}
+
+
+/*
  * Update resource in-use status
  * All resources bound to color or depth targets marked as WRITE resources.
  * VBO Vertex/index buffers and texture views marked as READ resources.
@@ -1060,9 +1094,30 @@ swr_update_derived(struct pipe_context *pipe,
       rastState->pointSpriteTopOrigin =
          rasterizer->sprite_coord_mode == PIPE_SPRITE_COORD_UPPER_LEFT;
 
-      /* XXX TODO: Add multisample */
-      rastState->sampleCount = SWR_MULTISAMPLE_1X;
+      /* If SWR_MSAA_FORCE_ENABLE is set, turn msaa on */
+      if (screen->msaa_force_enable && !rasterizer->multisample) {
+         /* Force enable and use the value the surface was created with */
+         rasterizer->multisample = true;
+         fb->samples = swr_resource(fb->cbufs[0]->texture)->swr.numSamples;
+         fprintf(stderr,"msaa force enable: %d samples\n", fb->samples);
+      }
+
+      rastState->sampleCount = GetSampleCount(fb->samples);
       rastState->forcedSampleCount = false;
+      rastState->bIsCenterPattern = !rasterizer->multisample;
+      rastState->pixelLocation = SWR_PIXEL_LOCATION_CENTER;
+
+      /* Only initialize sample positions if msaa is enabled */
+      if (rasterizer->multisample) {
+         for (uint32_t i = 0; i < fb->samples; i++) {
+            const uint8_t *sample = swr_sample_positions[fb->samples-1 + i];
+            rastState->samplePositions.SetXi(i, sample[0] << 4);
+            rastState->samplePositions.SetYi(i, sample[1] << 4);
+            rastState->samplePositions.SetX (i, sample[0] / 16.0f);
+            rastState->samplePositions.SetY (i, sample[1] / 16.0f);
+         }
+         rastState->samplePositions.PrecalcSampleData(fb->samples);
+      }
 
       bool do_offset = false;
       switch (rasterizer->fill_front) {
@@ -1375,9 +1430,9 @@ swr_update_derived(struct pipe_context *pipe,
       psState.inputCoverage = SWR_INPUT_COVERAGE_NORMAL;
       psState.writesODepth = ctx->fs->info.base.writes_z;
       psState.usesSourceDepth = ctx->fs->info.base.reads_z;
-      psState.shadingRate = SWR_SHADING_RATE_PIXEL; // XXX
+      psState.shadingRate = SWR_SHADING_RATE_PIXEL;
       psState.numRenderTargets = ctx->framebuffer.nr_cbufs;
-      psState.posOffset = SWR_PS_POSITION_SAMPLE_NONE; // XXX msaa
+      psState.posOffset = SWR_PS_POSITION_SAMPLE_NONE;
       uint32_t barycentricsMask = 0;
 #if 0
       // when we switch to mesa-master
@@ -1507,6 +1562,7 @@ swr_update_derived(struct pipe_context *pipe,
 
    /* Blend State */
    if (ctx->dirty & (SWR_NEW_BLEND |
+                     SWR_NEW_RASTERIZER |
                      SWR_NEW_FRAMEBUFFER |
                      SWR_NEW_DEPTH_STENCIL_ALPHA)) {
       struct pipe_framebuffer_state *fb = &ctx->framebuffer;
@@ -1520,9 +1576,8 @@ swr_update_derived(struct pipe_context *pipe,
       blendState.alphaTestReference =
          *((uint32_t*)&ctx->depth_stencil->alpha.ref_value);
 
-      // XXX MSAA
-      blendState.sampleMask = 0;
-      blendState.sampleCount = SWR_MULTISAMPLE_1X;
+      blendState.sampleMask = ctx->sample_mask;
+      blendState.sampleCount = GetSampleCount(fb->samples);
 
       /* If there are no color buffers bound, disable writes on RT0
        * and skip loop */
@@ -1578,8 +1633,8 @@ swr_update_derived(struct pipe_context *pipe,
                 compileState.blendState.alphaBlendFunc);
             compileState.desc.alphaToCoverageEnable =
                ctx->blend->pipe.alpha_to_coverage;
-            compileState.desc.sampleMaskEnable = 0; // XXX
-            compileState.desc.numSamples = 1; // XXX
+            compileState.desc.sampleMaskEnable = (blendState.sampleMask != 0);
+            compileState.desc.numSamples = fb->samples;
 
             compileState.alphaTestFunction =
                swr_convert_depth_func(ctx->depth_stencil->alpha.func);
@@ -1781,6 +1836,7 @@ swr_state_init(struct pipe_context *pipe)
    pipe->set_stencil_ref = swr_set_stencil_ref;
 
    pipe->set_sample_mask = swr_set_sample_mask;
+   pipe->get_sample_position = swr_get_sample_position;
 
    pipe->create_stream_output_target = swr_create_so_target;
    pipe->stream_output_target_destroy = swr_destroy_so_target;