i965: Port gen6+ blend state code to genxml.
authorRafael Antognolli <rafael.antognolli@intel.com>
Mon, 27 Mar 2017 22:52:10 +0000 (15:52 -0700)
committerKenneth Graunke <kenneth@whitecape.org>
Thu, 4 May 2017 01:57:52 +0000 (18:57 -0700)
Upload blend states using GENX(BLEND_STATE_ENTRY_pack), generated from
genxml.

v3:
   - style fixes (Ken)
   - cleanup to remove excessive #ifdef's (Ken)
   - remove memset (Ken)
   - disable blend.AlphaToCoverageDitherEnable on gen6 (Ken)

Signed-off-by: Rafael Antognolli <rafael.antognolli@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
src/mesa/drivers/dri/i965/Makefile.sources
src/mesa/drivers/dri/i965/brw_state.h
src/mesa/drivers/dri/i965/gen6_cc.c
src/mesa/drivers/dri/i965/gen8_blend_state.c [deleted file]
src/mesa/drivers/dri/i965/genX_state_upload.c

index 3f0c66a72211b81f1d49d3fb2693906189f85486..0c67170734590516b1f68a0de7abb2ebf9068332 100644 (file)
@@ -99,7 +99,6 @@ i965_FILES = \
        gen7_te_state.c \
        gen7_urb.c \
        gen7_wm_surface_state.c \
-       gen8_blend_state.c \
        gen8_depth_state.c \
        gen8_draw_upload.c \
        gen8_multisample_state.c \
index b6e8abc4f66b88da1338810e3d1b77c571fbda97..cf043a027a49eb5783ca078c27bed3d691b6b4f8 100644 (file)
@@ -108,7 +108,6 @@ extern const struct brw_tracked_state brw_index_buffer;
 extern const struct brw_tracked_state brw_cs_state;
 extern const struct brw_tracked_state gen7_cs_push_constants;
 extern const struct brw_tracked_state gen6_binding_table_pointers;
-extern const struct brw_tracked_state gen6_blend_state;
 extern const struct brw_tracked_state gen6_color_calc_state;
 extern const struct brw_tracked_state gen6_gs_push_constants;
 extern const struct brw_tracked_state gen6_gs_binding_table;
@@ -130,11 +129,9 @@ extern const struct brw_tracked_state gen7_te_state;
 extern const struct brw_tracked_state gen7_tes_push_constants;
 extern const struct brw_tracked_state gen7_urb;
 extern const struct brw_tracked_state haswell_cut_index;
-extern const struct brw_tracked_state gen8_blend_state;
 extern const struct brw_tracked_state gen8_index_buffer;
 extern const struct brw_tracked_state gen8_multisample_state;
 extern const struct brw_tracked_state gen8_pma_fix;
-extern const struct brw_tracked_state gen8_ps_blend;
 extern const struct brw_tracked_state gen8_vertices;
 extern const struct brw_tracked_state gen8_vf_topology;
 extern const struct brw_tracked_state brw_cs_work_groups_surface;
index 0e0d05e81465055194443afcb475dd1e7e93fb68..688362fcc59064e64b422c6378a05550ae6b1afc 100644 (file)
 #include "main/glformats.h"
 #include "main/stencil.h"
 
-static void
-gen6_upload_blend_state(struct brw_context *brw)
-{
-   bool is_buffer_zero_integer_format = false;
-   struct gl_context *ctx = &brw->ctx;
-   struct gen6_blend_state *blend;
-   int b;
-   int nr_draw_buffers = ctx->DrawBuffer->_NumColorDrawBuffers;
-   int size;
-
-   /* We need at least one BLEND_STATE written, because we might do
-    * thread dispatch even if _NumColorDrawBuffers is 0 (for example
-    * for computed depth or alpha test), which will do an FB write
-    * with render target 0, which will reference BLEND_STATE[0] for
-    * alpha test enable.
-    */
-   if (nr_draw_buffers == 0)
-      nr_draw_buffers = 1;
-
-   size = sizeof(*blend) * nr_draw_buffers;
-   blend = brw_state_batch(brw, size, 64, &brw->cc.blend_state_offset);
-
-   memset(blend, 0, size);
-
-   for (b = 0; b < nr_draw_buffers; b++) {
-      /* _NEW_BUFFERS */
-      struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[b];
-      GLenum rb_type;
-      bool integer;
-
-      if (rb)
-        rb_type = _mesa_get_format_datatype(rb->Format);
-      else
-        rb_type = GL_UNSIGNED_NORMALIZED;
-
-      /* Used for implementing the following bit of GL_EXT_texture_integer:
-       *     "Per-fragment operations that require floating-point color
-       *      components, including multisample alpha operations, alpha test,
-       *      blending, and dithering, have no effect when the corresponding
-       *      colors are written to an integer color buffer."
-      */
-      integer = (rb_type == GL_INT || rb_type == GL_UNSIGNED_INT);
-
-      if(b == 0 && integer)
-         is_buffer_zero_integer_format = true;
-
-      /* _NEW_COLOR */
-      if (ctx->Color.ColorLogicOpEnabled) {
-        /* Floating point RTs should have no effect from LogicOp,
-         * except for disabling of blending, but other types should.
-         *
-         * However, from the Sandy Bridge PRM, Vol 2 Par 1, Section 8.1.11,
-         * "Logic Ops",
-         *
-         *     "Logic Ops are only supported on *_UNORM surfaces (excluding
-         *      _SRGB variants), otherwise Logic Ops must be DISABLED."
-         */
-         WARN_ONCE(ctx->Color.LogicOp != GL_COPY &&
-                   rb_type != GL_UNSIGNED_NORMALIZED &&
-                   rb_type != GL_FLOAT, "Ignoring %s logic op on %s "
-                   "renderbuffer\n",
-                   _mesa_enum_to_string(ctx->Color.LogicOp),
-                   _mesa_enum_to_string(rb_type));
-        if (rb_type == GL_UNSIGNED_NORMALIZED) {
-           blend[b].blend1.logic_op_enable = 1;
-           blend[b].blend1.logic_op_func =
-              intel_translate_logic_op(ctx->Color.LogicOp);
-        }
-      } else if (ctx->Color.BlendEnabled & (1 << b) && !integer &&
-                 !ctx->Color._AdvancedBlendMode) {
-        GLenum eqRGB = ctx->Color.Blend[b].EquationRGB;
-        GLenum eqA = ctx->Color.Blend[b].EquationA;
-        GLenum srcRGB = ctx->Color.Blend[b].SrcRGB;
-        GLenum dstRGB = ctx->Color.Blend[b].DstRGB;
-        GLenum srcA = ctx->Color.Blend[b].SrcA;
-        GLenum dstA = ctx->Color.Blend[b].DstA;
-
-        if (eqRGB == GL_MIN || eqRGB == GL_MAX) {
-           srcRGB = dstRGB = GL_ONE;
-        }
-
-        if (eqA == GL_MIN || eqA == GL_MAX) {
-           srcA = dstA = GL_ONE;
-        }
-
-         /* Due to hardware limitations, the destination may have information
-          * in an alpha channel even when the format specifies no alpha
-          * channel. In order to avoid getting any incorrect blending due to
-          * that alpha channel, coerce the blend factors to values that will
-          * not read the alpha channel, but will instead use the correct
-          * implicit value for alpha.
-          */
-         if (rb && !_mesa_base_format_has_channel(rb->_BaseFormat, GL_TEXTURE_ALPHA_TYPE))
-         {
-            srcRGB = brw_fix_xRGB_alpha(srcRGB);
-            srcA = brw_fix_xRGB_alpha(srcA);
-            dstRGB = brw_fix_xRGB_alpha(dstRGB);
-            dstA = brw_fix_xRGB_alpha(dstA);
-         }
-
-        blend[b].blend0.dest_blend_factor = brw_translate_blend_factor(dstRGB);
-        blend[b].blend0.source_blend_factor = brw_translate_blend_factor(srcRGB);
-        blend[b].blend0.blend_func = brw_translate_blend_equation(eqRGB);
-
-        blend[b].blend0.ia_dest_blend_factor = brw_translate_blend_factor(dstA);
-        blend[b].blend0.ia_source_blend_factor = brw_translate_blend_factor(srcA);
-        blend[b].blend0.ia_blend_func = brw_translate_blend_equation(eqA);
-
-        blend[b].blend0.blend_enable = 1;
-        blend[b].blend0.ia_blend_enable = (srcA != srcRGB ||
-                                        dstA != dstRGB ||
-                                        eqA != eqRGB);
-      }
-
-      /* See section 8.1.6 "Pre-Blend Color Clamping" of the
-       * SandyBridge PRM Volume 2 Part 1 for HW requirements.
-       *
-       * We do our ARB_color_buffer_float CLAMP_FRAGMENT_COLOR
-       * clamping in the fragment shader.  For its clamping of
-       * blending, the spec says:
-       *
-       *     "RESOLVED: For fixed-point color buffers, the inputs and
-       *      the result of the blending equation are clamped.  For
-       *      floating-point color buffers, no clamping occurs."
-       *
-       * So, generally, we want clamping to the render target's range.
-       * And, good news, the hardware tables for both pre- and
-       * post-blend color clamping are either ignored, or any are
-       * allowed, or clamping is required but RT range clamping is a
-       * valid option.
-       */
-      blend[b].blend1.pre_blend_clamp_enable = 1;
-      blend[b].blend1.post_blend_clamp_enable = 1;
-      blend[b].blend1.clamp_range = BRW_RENDERTARGET_CLAMPRANGE_FORMAT;
-
-      /* _NEW_COLOR */
-      if (ctx->Color.AlphaEnabled && !integer) {
-        blend[b].blend1.alpha_test_enable = 1;
-        blend[b].blend1.alpha_test_func =
-           intel_translate_compare_func(ctx->Color.AlphaFunc);
-
-      }
-
-      /* _NEW_COLOR */
-      if (ctx->Color.DitherFlag && !integer) {
-        blend[b].blend1.dither_enable = 1;
-        blend[b].blend1.y_dither_offset = 0;
-        blend[b].blend1.x_dither_offset = 0;
-      }
-
-      blend[b].blend1.write_disable_r = !ctx->Color.ColorMask[b][0];
-      blend[b].blend1.write_disable_g = !ctx->Color.ColorMask[b][1];
-      blend[b].blend1.write_disable_b = !ctx->Color.ColorMask[b][2];
-      blend[b].blend1.write_disable_a = !ctx->Color.ColorMask[b][3];
-
-      /* OpenGL specification 3.3 (page 196), section 4.1.3 says:
-       * "If drawbuffer zero is not NONE and the buffer it references has an
-       * integer format, the SAMPLE_ALPHA_TO_COVERAGE and SAMPLE_ALPHA_TO_ONE
-       * operations are skipped."
-       */
-      if(!is_buffer_zero_integer_format) {
-         /* _NEW_MULTISAMPLE */
-         blend[b].blend1.alpha_to_coverage =
-            _mesa_is_multisample_enabled(ctx) && ctx->Multisample.SampleAlphaToCoverage;
-
-       /* From SandyBridge PRM, volume 2 Part 1, section 8.2.3, BLEND_STATE:
-        * DWord 1, Bit 30 (AlphaToOne Enable):
-        * "If Dual Source Blending is enabled, this bit must be disabled"
-        */
-         WARN_ONCE(ctx->Color.Blend[b]._UsesDualSrc &&
-                   _mesa_is_multisample_enabled(ctx) &&
-                   ctx->Multisample.SampleAlphaToOne,
-                   "HW workaround: disabling alpha to one with dual src "
-                   "blending\n");
-        if (ctx->Color.Blend[b]._UsesDualSrc)
-            blend[b].blend1.alpha_to_one = false;
-        else
-           blend[b].blend1.alpha_to_one =
-              _mesa_is_multisample_enabled(ctx) && ctx->Multisample.SampleAlphaToOne;
-
-         blend[b].blend1.alpha_to_coverage_dither = (brw->gen >= 7);
-      }
-      else {
-         blend[b].blend1.alpha_to_coverage = false;
-         blend[b].blend1.alpha_to_one = false;
-      }
-   }
-
-   /* Point the GPU at the new indirect state. */
-   if (brw->gen == 6) {
-      BEGIN_BATCH(4);
-      OUT_BATCH(_3DSTATE_CC_STATE_POINTERS << 16 | (4 - 2));
-      OUT_BATCH(brw->cc.blend_state_offset | 1);
-      OUT_BATCH(0);
-      OUT_BATCH(0);
-      ADVANCE_BATCH();
-   } else {
-      BEGIN_BATCH(2);
-      OUT_BATCH(_3DSTATE_BLEND_STATE_POINTERS << 16 | (2 - 2));
-      OUT_BATCH(brw->cc.blend_state_offset | 1);
-      ADVANCE_BATCH();
-   }
-}
-
-const struct brw_tracked_state gen6_blend_state = {
-   .dirty = {
-      .mesa = _NEW_BUFFERS |
-              _NEW_COLOR |
-              _NEW_MULTISAMPLE,
-      .brw = BRW_NEW_BATCH |
-             BRW_NEW_BLORP |
-             BRW_NEW_STATE_BASE_ADDRESS,
-   },
-   .emit = gen6_upload_blend_state,
-};
-
 static void
 gen6_upload_color_calc_state(struct brw_context *brw)
 {
diff --git a/src/mesa/drivers/dri/i965/gen8_blend_state.c b/src/mesa/drivers/dri/i965/gen8_blend_state.c
deleted file mode 100644 (file)
index 1fa8ba2..0000000
+++ /dev/null
@@ -1,298 +0,0 @@
-/*
- * Copyright © 2012 Intel Corporation
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
- * IN THE SOFTWARE.
- */
-
-#include "brw_context.h"
-#include "brw_state.h"
-#include "brw_defines.h"
-#include "brw_util.h"
-#include "brw_wm.h"
-#include "intel_batchbuffer.h"
-#include "main/macros.h"
-#include "main/enums.h"
-#include "main/glformats.h"
-
-#define blend_factor(x) brw_translate_blend_factor(x)
-#define blend_eqn(x) brw_translate_blend_equation(x)
-
-static void
-gen8_upload_blend_state(struct brw_context *brw)
-{
-   struct gl_context *ctx = &brw->ctx;
-
-   /* We need at least one BLEND_STATE written, because we might do
-    * thread dispatch even if _NumColorDrawBuffers is 0 (for example
-    * for computed depth or alpha test), which will do an FB write
-    * with render target 0, which will reference BLEND_STATE[0] for
-    * alpha test enable.
-    */
-   int nr_draw_buffers = ctx->DrawBuffer->_NumColorDrawBuffers;
-   if (nr_draw_buffers == 0 && ctx->Color.AlphaEnabled)
-      nr_draw_buffers = 1;
-
-   int size = 4 + 8 * nr_draw_buffers;
-   uint32_t *blend =
-      brw_state_batch(brw, size, 64, &brw->cc.blend_state_offset);
-   memset(blend, 0, size);
-
-   /* OpenGL specification 3.3 (page 196), section 4.1.3 says:
-    * "If drawbuffer zero is not NONE and the buffer it references has an
-    * integer format, the SAMPLE_ALPHA_TO_COVERAGE and SAMPLE_ALPHA_TO_ONE
-    * operations are skipped."
-    */
-   if (!(ctx->DrawBuffer->_IntegerBuffers & 0x1)) {
-      /* _NEW_MULTISAMPLE */
-      if (_mesa_is_multisample_enabled(ctx)) {
-         if (ctx->Multisample.SampleAlphaToCoverage) {
-            blend[0] |= GEN8_BLEND_ALPHA_TO_COVERAGE_ENABLE;
-            blend[0] |= GEN8_BLEND_ALPHA_TO_COVERAGE_DITHER_ENABLE;
-         }
-         if (ctx->Multisample.SampleAlphaToOne)
-            blend[0] |= GEN8_BLEND_ALPHA_TO_ONE_ENABLE;
-      }
-
-      /* _NEW_COLOR */
-      if (ctx->Color.AlphaEnabled) {
-         blend[0] |=
-            GEN8_BLEND_ALPHA_TEST_ENABLE |
-            SET_FIELD(intel_translate_compare_func(ctx->Color.AlphaFunc),
-                      GEN8_BLEND_ALPHA_TEST_FUNCTION);
-      }
-
-      if (ctx->Color.DitherFlag) {
-         blend[0] |= GEN8_BLEND_COLOR_DITHER_ENABLE;
-      }
-   }
-
-   for (int i = 0; i < nr_draw_buffers; i++) {
-      /* _NEW_BUFFERS */
-      struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i];
-
-      /* Used for implementing the following bit of GL_EXT_texture_integer:
-       * "Per-fragment operations that require floating-point color
-       *  components, including multisample alpha operations, alpha test,
-       *  blending, and dithering, have no effect when the corresponding
-       *  colors are written to an integer color buffer."
-      */
-      bool integer = ctx->DrawBuffer->_IntegerBuffers & (0x1 << i);
-
-      /* _NEW_COLOR */
-      if (ctx->Color.ColorLogicOpEnabled) {
-         blend[1 + 2*i+1] |=
-            GEN8_BLEND_LOGIC_OP_ENABLE |
-            SET_FIELD(intel_translate_logic_op(ctx->Color.LogicOp),
-                      GEN8_BLEND_LOGIC_OP_FUNCTION);
-      } else if (ctx->Color.BlendEnabled & (1 << i) && !integer &&
-                 !ctx->Color._AdvancedBlendMode) {
-         GLenum eqRGB = ctx->Color.Blend[i].EquationRGB;
-         GLenum eqA = ctx->Color.Blend[i].EquationA;
-         GLenum srcRGB = ctx->Color.Blend[i].SrcRGB;
-         GLenum dstRGB = ctx->Color.Blend[i].DstRGB;
-         GLenum srcA = ctx->Color.Blend[i].SrcA;
-         GLenum dstA = ctx->Color.Blend[i].DstA;
-
-         if (eqRGB == GL_MIN || eqRGB == GL_MAX)
-            srcRGB = dstRGB = GL_ONE;
-
-         if (eqA == GL_MIN || eqA == GL_MAX)
-            srcA = dstA = GL_ONE;
-
-         /* Due to hardware limitations, the destination may have information
-          * in an alpha channel even when the format specifies no alpha
-          * channel. In order to avoid getting any incorrect blending due to
-          * that alpha channel, coerce the blend factors to values that will
-          * not read the alpha channel, but will instead use the correct
-          * implicit value for alpha.
-          */
-         if (rb && !_mesa_base_format_has_channel(rb->_BaseFormat, GL_TEXTURE_ALPHA_TYPE)) {
-            srcRGB = brw_fix_xRGB_alpha(srcRGB);
-            srcA = brw_fix_xRGB_alpha(srcA);
-            dstRGB = brw_fix_xRGB_alpha(dstRGB);
-            dstA = brw_fix_xRGB_alpha(dstA);
-         }
-
-         blend[1 + 2*i] |=
-            GEN8_BLEND_COLOR_BUFFER_BLEND_ENABLE |
-            SET_FIELD(blend_factor(dstRGB), GEN8_BLEND_DST_BLEND_FACTOR) |
-            SET_FIELD(blend_factor(srcRGB), GEN8_BLEND_SRC_BLEND_FACTOR) |
-            SET_FIELD(blend_factor(dstA), GEN8_BLEND_DST_ALPHA_BLEND_FACTOR) |
-            SET_FIELD(blend_factor(srcA), GEN8_BLEND_SRC_ALPHA_BLEND_FACTOR) |
-            SET_FIELD(blend_eqn(eqRGB), GEN8_BLEND_COLOR_BLEND_FUNCTION) |
-            SET_FIELD(blend_eqn(eqA), GEN8_BLEND_ALPHA_BLEND_FUNCTION);
-
-         if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB)
-            blend[0] |= GEN8_BLEND_INDEPENDENT_ALPHA_BLEND_ENABLE;
-      }
-
-      /* See section 8.1.6 "Pre-Blend Color Clamping" of the
-       * SandyBridge PRM Volume 2 Part 1 for HW requirements.
-       *
-       * We do our ARB_color_buffer_float CLAMP_FRAGMENT_COLOR
-       * clamping in the fragment shader.  For its clamping of
-       * blending, the spec says:
-       *
-       *     "RESOLVED: For fixed-point color buffers, the inputs and
-       *      the result of the blending equation are clamped.  For
-       *      floating-point color buffers, no clamping occurs."
-       *
-       * So, generally, we want clamping to the render target's range.
-       * And, good news, the hardware tables for both pre- and
-       * post-blend color clamping are either ignored, or any are
-       * allowed, or clamping is required but RT range clamping is a
-       * valid option.
-       */
-      blend[1 + 2*i+1] |=
-         GEN8_BLEND_PRE_BLEND_COLOR_CLAMP_ENABLE |
-         GEN8_BLEND_POST_BLEND_COLOR_CLAMP_ENABLE |
-         GEN8_BLEND_COLOR_CLAMP_RANGE_RTFORMAT;
-
-      if (!ctx->Color.ColorMask[i][0])
-         blend[1 + 2*i] |= GEN8_BLEND_WRITE_DISABLE_RED;
-      if (!ctx->Color.ColorMask[i][1])
-         blend[1 + 2*i] |= GEN8_BLEND_WRITE_DISABLE_GREEN;
-      if (!ctx->Color.ColorMask[i][2])
-         blend[1 + 2*i] |= GEN8_BLEND_WRITE_DISABLE_BLUE;
-      if (!ctx->Color.ColorMask[i][3])
-         blend[1 + 2*i] |= GEN8_BLEND_WRITE_DISABLE_ALPHA;
-
-     /* From the BLEND_STATE docs, DWord 0, Bit 29 (AlphaToOne Enable):
-      * "If Dual Source Blending is enabled, this bit must be disabled."
-      */
-      WARN_ONCE(ctx->Color.Blend[i]._UsesDualSrc &&
-                _mesa_is_multisample_enabled(ctx) &&
-                ctx->Multisample.SampleAlphaToOne,
-                "HW workaround: disabling alpha to one with dual src "
-                "blending\n");
-      if (ctx->Color.Blend[i]._UsesDualSrc)
-         blend[0] &= ~GEN8_BLEND_ALPHA_TO_ONE_ENABLE;
-   }
-
-   BEGIN_BATCH(2);
-   OUT_BATCH(_3DSTATE_BLEND_STATE_POINTERS << 16 | (2 - 2));
-   OUT_BATCH(brw->cc.blend_state_offset | 1);
-   ADVANCE_BATCH();
-}
-
-const struct brw_tracked_state gen8_blend_state = {
-   .dirty = {
-      .mesa = _NEW_BUFFERS |
-              _NEW_COLOR |
-              _NEW_MULTISAMPLE,
-      .brw = BRW_NEW_BATCH |
-             BRW_NEW_BLORP |
-             BRW_NEW_STATE_BASE_ADDRESS,
-   },
-   .emit = gen8_upload_blend_state,
-};
-
-static void
-gen8_upload_ps_blend(struct brw_context *brw)
-{
-   struct gl_context *ctx = &brw->ctx;
-   uint32_t dw1 = 0;
-
-   /* _NEW_BUFFERS */
-   struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[0];
-   const bool buffer0_is_integer = ctx->DrawBuffer->_IntegerBuffers & 0x1;
-
-   /* BRW_NEW_FRAGMENT_PROGRAM | _NEW_BUFFERS | _NEW_COLOR */
-   if (brw_color_buffer_write_enabled(brw))
-      dw1 |= GEN8_PS_BLEND_HAS_WRITEABLE_RT;
-
-   if (!buffer0_is_integer) {
-      /* _NEW_COLOR */
-      if (ctx->Color.AlphaEnabled)
-         dw1 |= GEN8_PS_BLEND_ALPHA_TEST_ENABLE;
-
-      /* _NEW_MULTISAMPLE */
-      if (_mesa_is_multisample_enabled(ctx) &&
-          ctx->Multisample.SampleAlphaToCoverage)
-         dw1 |= GEN8_PS_BLEND_ALPHA_TO_COVERAGE_ENABLE;
-   }
-
-   /* Used for implementing the following bit of GL_EXT_texture_integer:
-    * "Per-fragment operations that require floating-point color
-    *  components, including multisample alpha operations, alpha test,
-    *  blending, and dithering, have no effect when the corresponding
-    *  colors are written to an integer color buffer."
-    *
-    * The OpenGL specification 3.3 (page 196), section 4.1.3 says:
-    * "If drawbuffer zero is not NONE and the buffer it references has an
-    *  integer format, the SAMPLE_ALPHA_TO_COVERAGE and SAMPLE_ALPHA_TO_ONE
-    *  operations are skipped."
-    */
-   if (rb && !buffer0_is_integer && (ctx->Color.BlendEnabled & 1)) {
-      GLenum eqRGB = ctx->Color.Blend[0].EquationRGB;
-      GLenum eqA = ctx->Color.Blend[0].EquationA;
-      GLenum srcRGB = ctx->Color.Blend[0].SrcRGB;
-      GLenum dstRGB = ctx->Color.Blend[0].DstRGB;
-      GLenum srcA = ctx->Color.Blend[0].SrcA;
-      GLenum dstA = ctx->Color.Blend[0].DstA;
-
-      if (eqRGB == GL_MIN || eqRGB == GL_MAX)
-         srcRGB = dstRGB = GL_ONE;
-
-      if (eqA == GL_MIN || eqA == GL_MAX)
-         srcA = dstA = GL_ONE;
-
-      /* Due to hardware limitations, the destination may have information
-       * in an alpha channel even when the format specifies no alpha
-       * channel. In order to avoid getting any incorrect blending due to
-       * that alpha channel, coerce the blend factors to values that will
-       * not read the alpha channel, but will instead use the correct
-       * implicit value for alpha.
-       */
-      if (!_mesa_base_format_has_channel(rb->_BaseFormat, GL_TEXTURE_ALPHA_TYPE)) {
-         srcRGB = brw_fix_xRGB_alpha(srcRGB);
-         srcA = brw_fix_xRGB_alpha(srcA);
-         dstRGB = brw_fix_xRGB_alpha(dstRGB);
-         dstA = brw_fix_xRGB_alpha(dstA);
-      }
-
-      dw1 |=
-         GEN8_PS_BLEND_COLOR_BUFFER_BLEND_ENABLE |
-         SET_FIELD(blend_factor(dstRGB), GEN8_PS_BLEND_DST_BLEND_FACTOR) |
-         SET_FIELD(blend_factor(srcRGB), GEN8_PS_BLEND_SRC_BLEND_FACTOR) |
-         SET_FIELD(blend_factor(dstA), GEN8_PS_BLEND_DST_ALPHA_BLEND_FACTOR) |
-         SET_FIELD(blend_factor(srcA), GEN8_PS_BLEND_SRC_ALPHA_BLEND_FACTOR);
-
-      if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB)
-         dw1 |= GEN8_PS_BLEND_INDEPENDENT_ALPHA_BLEND_ENABLE;
-   }
-
-   BEGIN_BATCH(2);
-   OUT_BATCH(_3DSTATE_PS_BLEND << 16 | (2 - 2));
-   OUT_BATCH(dw1);
-   ADVANCE_BATCH();
-}
-
-const struct brw_tracked_state gen8_ps_blend = {
-   .dirty = {
-      .mesa = _NEW_BUFFERS |
-              _NEW_COLOR |
-              _NEW_MULTISAMPLE,
-      .brw = BRW_NEW_BLORP |
-             BRW_NEW_CONTEXT |
-             BRW_NEW_FRAGMENT_PROGRAM,
-   },
-   .emit = gen8_upload_ps_blend
-};
index dd7db7ad57cd6c315a759295fcc73efe95ca8c72..f51036d41fd8e9cd3c7eb1e1f3ad579b27bea232 100644 (file)
 #include "intel_buffer_objects.h"
 #include "intel_fbo.h"
 
+#include "main/enums.h"
 #include "main/fbobject.h"
 #include "main/framebuffer.h"
+#include "main/glformats.h"
 #include "main/stencil.h"
 #include "main/transformfeedback.h"
 #include "main/viewport.h"
@@ -1449,6 +1451,221 @@ static const struct brw_tracked_state genX(gs_state) = {
    .emit = genX(upload_gs_state),
 };
 
+/* ---------------------------------------------------------------------- */
+
+#define blend_factor(x) brw_translate_blend_factor(x)
+#define blend_eqn(x) brw_translate_blend_equation(x)
+
+static void
+genX(upload_blend_state)(struct brw_context *brw)
+{
+   struct gl_context *ctx = &brw->ctx;
+   int size;
+
+   /* We need at least one BLEND_STATE written, because we might do
+    * thread dispatch even if _NumColorDrawBuffers is 0 (for example
+    * for computed depth or alpha test), which will do an FB write
+    * with render target 0, which will reference BLEND_STATE[0] for
+    * alpha test enable.
+    */
+   int nr_draw_buffers = ctx->DrawBuffer->_NumColorDrawBuffers;
+   if (nr_draw_buffers == 0 && ctx->Color.AlphaEnabled)
+      nr_draw_buffers = 1;
+
+   size = GENX(BLEND_STATE_ENTRY_length) * 4 * nr_draw_buffers;
+#if GEN_GEN >= 8
+   size += GENX(BLEND_STATE_length) * 4;
+#endif
+
+   uint32_t *blend_map;
+   blend_map = brw_state_batch(brw, size, 64, &brw->cc.blend_state_offset);
+
+#if GEN_GEN >= 8
+   struct GENX(BLEND_STATE) blend = { 0 };
+   {
+#else
+   for (int i = 0; i < nr_draw_buffers; i++) {
+      struct GENX(BLEND_STATE_ENTRY) entry = { 0 };
+#define blend entry
+#endif
+      /* OpenGL specification 3.3 (page 196), section 4.1.3 says:
+       * "If drawbuffer zero is not NONE and the buffer it references has an
+       * integer format, the SAMPLE_ALPHA_TO_COVERAGE and SAMPLE_ALPHA_TO_ONE
+       * operations are skipped."
+       */
+      if (!(ctx->DrawBuffer->_IntegerBuffers & 0x1)) {
+         /* _NEW_MULTISAMPLE */
+         if (_mesa_is_multisample_enabled(ctx)) {
+            if (ctx->Multisample.SampleAlphaToCoverage) {
+               blend.AlphaToCoverageEnable = true;
+               blend.AlphaToCoverageDitherEnable = GEN_GEN >= 7;
+            }
+            if (ctx->Multisample.SampleAlphaToOne)
+               blend.AlphaToOneEnable = true;
+         }
+
+         /* _NEW_COLOR */
+         if (ctx->Color.AlphaEnabled) {
+            blend.AlphaTestEnable = true;
+            blend.AlphaTestFunction =
+               intel_translate_compare_func(ctx->Color.AlphaFunc);
+         }
+
+         if (ctx->Color.DitherFlag) {
+            blend.ColorDitherEnable = true;
+         }
+      }
+
+#if GEN_GEN >= 8
+      for (int i = 0; i < nr_draw_buffers; i++) {
+         struct GENX(BLEND_STATE_ENTRY) entry = { 0 };
+#else
+      {
+#endif
+
+         /* _NEW_BUFFERS */
+         struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i];
+
+         /* Used for implementing the following bit of GL_EXT_texture_integer:
+          * "Per-fragment operations that require floating-point color
+          *  components, including multisample alpha operations, alpha test,
+          *  blending, and dithering, have no effect when the corresponding
+          *  colors are written to an integer color buffer."
+          */
+         bool integer = ctx->DrawBuffer->_IntegerBuffers & (0x1 << i);
+
+         /* _NEW_COLOR */
+         if (ctx->Color.ColorLogicOpEnabled) {
+            GLenum rb_type = rb ? _mesa_get_format_datatype(rb->Format)
+                                : GL_UNSIGNED_NORMALIZED;
+            WARN_ONCE(ctx->Color.LogicOp != GL_COPY &&
+                      rb_type != GL_UNSIGNED_NORMALIZED &&
+                      rb_type != GL_FLOAT, "Ignoring %s logic op on %s "
+                      "renderbuffer\n",
+                      _mesa_enum_to_string(ctx->Color.LogicOp),
+                      _mesa_enum_to_string(rb_type));
+            if (GEN_GEN >= 8 || rb_type == GL_UNSIGNED_NORMALIZED) {
+               entry.LogicOpEnable = true;
+               entry.LogicOpFunction =
+                  intel_translate_logic_op(ctx->Color.LogicOp);
+            }
+         } else if (ctx->Color.BlendEnabled & (1 << i) && !integer &&
+                    !ctx->Color._AdvancedBlendMode) {
+            GLenum eqRGB = ctx->Color.Blend[i].EquationRGB;
+            GLenum eqA = ctx->Color.Blend[i].EquationA;
+            GLenum srcRGB = ctx->Color.Blend[i].SrcRGB;
+            GLenum dstRGB = ctx->Color.Blend[i].DstRGB;
+            GLenum srcA = ctx->Color.Blend[i].SrcA;
+            GLenum dstA = ctx->Color.Blend[i].DstA;
+
+            if (eqRGB == GL_MIN || eqRGB == GL_MAX)
+               srcRGB = dstRGB = GL_ONE;
+
+            if (eqA == GL_MIN || eqA == GL_MAX)
+               srcA = dstA = GL_ONE;
+
+            /* Due to hardware limitations, the destination may have information
+             * in an alpha channel even when the format specifies no alpha
+             * channel. In order to avoid getting any incorrect blending due to
+             * that alpha channel, coerce the blend factors to values that will
+             * not read the alpha channel, but will instead use the correct
+             * implicit value for alpha.
+             */
+            if (rb && !_mesa_base_format_has_channel(rb->_BaseFormat,
+                                                     GL_TEXTURE_ALPHA_TYPE)) {
+               srcRGB = brw_fix_xRGB_alpha(srcRGB);
+               srcA = brw_fix_xRGB_alpha(srcA);
+               dstRGB = brw_fix_xRGB_alpha(dstRGB);
+               dstA = brw_fix_xRGB_alpha(dstA);
+            }
+
+            entry.ColorBufferBlendEnable = true;
+            entry.DestinationBlendFactor = blend_factor(dstRGB);
+            entry.SourceBlendFactor = blend_factor(srcRGB);
+            entry.DestinationAlphaBlendFactor = blend_factor(dstA);
+            entry.SourceAlphaBlendFactor = blend_factor(srcA);
+            entry.ColorBlendFunction = blend_eqn(eqRGB);
+            entry.AlphaBlendFunction = blend_eqn(eqA);
+
+            if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB)
+               blend.IndependentAlphaBlendEnable = true;
+         }
+
+         /* See section 8.1.6 "Pre-Blend Color Clamping" of the
+          * SandyBridge PRM Volume 2 Part 1 for HW requirements.
+          *
+          * We do our ARB_color_buffer_float CLAMP_FRAGMENT_COLOR
+          * clamping in the fragment shader.  For its clamping of
+          * blending, the spec says:
+          *
+          *     "RESOLVED: For fixed-point color buffers, the inputs and
+          *      the result of the blending equation are clamped.  For
+          *      floating-point color buffers, no clamping occurs."
+          *
+          * So, generally, we want clamping to the render target's range.
+          * And, good news, the hardware tables for both pre- and
+          * post-blend color clamping are either ignored, or any are
+          * allowed, or clamping is required but RT range clamping is a
+          * valid option.
+          */
+         entry.PreBlendColorClampEnable = true;
+         entry.PostBlendColorClampEnable = true;
+         entry.ColorClampRange = COLORCLAMP_RTFORMAT;
+
+         entry.WriteDisableRed   = !ctx->Color.ColorMask[i][0];
+         entry.WriteDisableGreen = !ctx->Color.ColorMask[i][1];
+         entry.WriteDisableBlue  = !ctx->Color.ColorMask[i][2];
+         entry.WriteDisableAlpha = !ctx->Color.ColorMask[i][3];
+
+         /* From the BLEND_STATE docs, DWord 0, Bit 29 (AlphaToOne Enable):
+          * "If Dual Source Blending is enabled, this bit must be disabled."
+          */
+         WARN_ONCE(ctx->Color.Blend[i]._UsesDualSrc &&
+                   _mesa_is_multisample_enabled(ctx) &&
+                   ctx->Multisample.SampleAlphaToOne,
+                   "HW workaround: disabling alpha to one with dual src "
+                   "blending\n");
+         if (ctx->Color.Blend[i]._UsesDualSrc)
+            blend.AlphaToOneEnable = false;
+#if GEN_GEN >= 8
+         GENX(BLEND_STATE_ENTRY_pack)(NULL, &blend_map[1 + i * 2], &entry);
+#else
+         GENX(BLEND_STATE_ENTRY_pack)(NULL, &blend_map[i * 2], &entry);
+#endif
+      }
+   }
+
+#if GEN_GEN >= 8
+   GENX(BLEND_STATE_pack)(NULL, blend_map, &blend);
+#endif
+
+#if GEN_GEN < 7
+   brw_batch_emit(brw, GENX(3DSTATE_CC_STATE_POINTERS), ptr) {
+      ptr.PointertoBLEND_STATE = brw->cc.blend_state_offset;
+      ptr.BLEND_STATEChange = true;
+   }
+#else
+   brw_batch_emit(brw, GENX(3DSTATE_BLEND_STATE_POINTERS), ptr) {
+      ptr.BlendStatePointer = brw->cc.blend_state_offset;
+#if GEN_GEN >= 8
+      ptr.BlendStatePointerValid = true;
+#endif
+   }
+#endif
+}
+
+static const struct brw_tracked_state genX(blend_state) = {
+   .dirty = {
+      .mesa = _NEW_BUFFERS |
+              _NEW_COLOR |
+              _NEW_MULTISAMPLE,
+      .brw = BRW_NEW_BATCH |
+             BRW_NEW_BLORP |
+             BRW_NEW_STATE_BASE_ADDRESS,
+   },
+   .emit = genX(upload_blend_state),
+};
+
 #endif
 
 /* ---------------------------------------------------------------------- */
@@ -2297,6 +2514,97 @@ const struct brw_tracked_state genX(ps_extra) = {
    },
    .emit = genX(upload_ps_extra),
 };
+
+/* ---------------------------------------------------------------------- */
+
+static void
+genX(upload_ps_blend)(struct brw_context *brw)
+{
+   struct gl_context *ctx = &brw->ctx;
+
+   /* _NEW_BUFFERS */
+   struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[0];
+   const bool buffer0_is_integer = ctx->DrawBuffer->_IntegerBuffers & 0x1;
+
+   /* _NEW_COLOR */
+   struct gl_colorbuffer_attrib *color = &ctx->Color;
+
+   brw_batch_emit(brw, GENX(3DSTATE_PS_BLEND), pb) {
+      /* BRW_NEW_FRAGMENT_PROGRAM | _NEW_BUFFERS | _NEW_COLOR */
+      pb.HasWriteableRT = brw_color_buffer_write_enabled(brw);
+
+      if (!buffer0_is_integer) {
+         /* _NEW_MULTISAMPLE */
+         pb.AlphaToCoverageEnable =
+            _mesa_is_multisample_enabled(ctx) &&
+            ctx->Multisample.SampleAlphaToCoverage;
+
+         pb.AlphaTestEnable = color->AlphaEnabled;
+      }
+
+      /* Used for implementing the following bit of GL_EXT_texture_integer:
+       * "Per-fragment operations that require floating-point color
+       *  components, including multisample alpha operations, alpha test,
+       *  blending, and dithering, have no effect when the corresponding
+       *  colors are written to an integer color buffer."
+       *
+       * The OpenGL specification 3.3 (page 196), section 4.1.3 says:
+       * "If drawbuffer zero is not NONE and the buffer it references has an
+       *  integer format, the SAMPLE_ALPHA_TO_COVERAGE and SAMPLE_ALPHA_TO_ONE
+       *  operations are skipped."
+       */
+      if (rb && !buffer0_is_integer && (color->BlendEnabled & 1)) {
+         GLenum eqRGB = color->Blend[0].EquationRGB;
+         GLenum eqA = color->Blend[0].EquationA;
+         GLenum srcRGB = color->Blend[0].SrcRGB;
+         GLenum dstRGB = color->Blend[0].DstRGB;
+         GLenum srcA = color->Blend[0].SrcA;
+         GLenum dstA = color->Blend[0].DstA;
+
+         if (eqRGB == GL_MIN || eqRGB == GL_MAX)
+            srcRGB = dstRGB = GL_ONE;
+
+         if (eqA == GL_MIN || eqA == GL_MAX)
+            srcA = dstA = GL_ONE;
+
+         /* Due to hardware limitations, the destination may have information
+          * in an alpha channel even when the format specifies no alpha
+          * channel. In order to avoid getting any incorrect blending due to
+          * that alpha channel, coerce the blend factors to values that will
+          * not read the alpha channel, but will instead use the correct
+          * implicit value for alpha.
+          */
+         if (!_mesa_base_format_has_channel(rb->_BaseFormat,
+                                            GL_TEXTURE_ALPHA_TYPE)) {
+            srcRGB = brw_fix_xRGB_alpha(srcRGB);
+            srcA = brw_fix_xRGB_alpha(srcA);
+            dstRGB = brw_fix_xRGB_alpha(dstRGB);
+            dstA = brw_fix_xRGB_alpha(dstA);
+         }
+
+         pb.ColorBufferBlendEnable = true;
+         pb.SourceAlphaBlendFactor = brw_translate_blend_factor(srcA);
+         pb.DestinationAlphaBlendFactor = brw_translate_blend_factor(dstA);
+         pb.SourceBlendFactor = brw_translate_blend_factor(srcRGB);
+         pb.DestinationBlendFactor = brw_translate_blend_factor(dstRGB);
+
+         pb.IndependentAlphaBlendEnable =
+            srcA != srcRGB || dstA != dstRGB || eqA != eqRGB;
+      }
+   }
+}
+
+static const struct brw_tracked_state genX(ps_blend) = {
+   .dirty = {
+      .mesa = _NEW_BUFFERS |
+              _NEW_COLOR |
+              _NEW_MULTISAMPLE,
+      .brw = BRW_NEW_BLORP |
+             BRW_NEW_CONTEXT |
+             BRW_NEW_FRAGMENT_PROGRAM,
+   },
+   .emit = genX(upload_ps_blend)
+};
 #endif
 
 /* ---------------------------------------------------------------------- */
@@ -2373,7 +2681,7 @@ genX(init_atoms)(struct brw_context *brw)
       &gen6_viewport_state,    /* must do after *_vp stages */
 
       &gen6_urb,
-      &gen6_blend_state,               /* must do before cc unit */
+      &genX(blend_state),              /* must do before cc unit */
       &gen6_color_calc_state,  /* must do before cc unit */
       &gen6_depth_stencil_state,       /* must do before cc unit */
 
@@ -2438,7 +2746,7 @@ genX(init_atoms)(struct brw_context *brw)
       &gen7_l3_state,
       &gen7_push_constant_space,
       &gen7_urb,
-      &gen6_blend_state,               /* must do before cc unit */
+      &genX(blend_state),              /* must do before cc unit */
       &gen6_color_calc_state,  /* must do before cc unit */
       &genX(depth_stencil_state),      /* must do before cc unit */
 
@@ -2526,7 +2834,7 @@ genX(init_atoms)(struct brw_context *brw)
       &gen7_l3_state,
       &gen7_push_constant_space,
       &gen7_urb,
-      &gen8_blend_state,
+      &genX(blend_state),
       &gen6_color_calc_state,
 
       &brw_vs_image_surfaces, /* Before vs push/pull constants and binding table */
@@ -2585,7 +2893,7 @@ genX(init_atoms)(struct brw_context *brw)
       &genX(raster_state),
       &genX(sbe_state),
       &genX(sf_state),
-      &gen8_ps_blend,
+      &genX(ps_blend),
       &genX(ps_extra),
       &genX(ps_state),
       &genX(depth_stencil_state),