xref: /aosp_15_r20/external/mesa3d/src/panfrost/lib/pan_blend.c (revision 6104692788411f58d303aa86923a9ff6ecaded22)

/*
 * Copyright (C) 2018 Alyssa Rosenzweig
 * Copyright (C) 2019-2021 Collabora, Ltd.
 *
 * 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 "pan_blend.h"
#include "util/blend.h"

#ifdef PAN_ARCH
#include "pan_shader.h"
#endif

#include "compiler/nir/nir.h"
#include "compiler/nir/nir_builder.h"
#include "compiler/nir/nir_conversion_builder.h"
#include "compiler/nir/nir_lower_blend.h"
#include "panfrost/util/pan_lower_framebuffer.h"
#include "util/format/u_format.h"
#include "pan_texture.h"

#ifndef PAN_ARCH

/* Fixed function blending */

static bool
factor_is_supported(enum pipe_blendfactor factor)
{
   factor = util_blendfactor_without_invert(factor);

   return factor != PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE &&
          factor != PIPE_BLENDFACTOR_SRC1_COLOR &&
          factor != PIPE_BLENDFACTOR_SRC1_ALPHA;
}

/* OpenGL allows encoding (src*dest + dest*src) which is incompatiblle with
 * Midgard style blending since there are two multiplies. However, it may be
 * factored as 2*src*dest = dest*(2*src), which can be encoded on Bifrost as 0
 * + dest * (2*src) wih the new source_2 value of C. Detect this case. */

static bool
is_2srcdest(enum pipe_blend_func blend_func, enum pipe_blendfactor src_factor,
            enum pipe_blendfactor dest_factor, bool is_alpha)
{
   return (blend_func == PIPE_BLEND_ADD) &&
          ((src_factor == PIPE_BLENDFACTOR_DST_COLOR) ||
           ((src_factor == PIPE_BLENDFACTOR_DST_ALPHA) && is_alpha)) &&
          ((dest_factor == PIPE_BLENDFACTOR_SRC_COLOR) ||
           ((dest_factor == PIPE_BLENDFACTOR_SRC_ALPHA) && is_alpha));
}

static bool
can_fixed_function_equation(enum pipe_blend_func blend_func,
                            enum pipe_blendfactor src_factor,
                            enum pipe_blendfactor dest_factor, bool is_alpha,
                            bool supports_2src)
{
   if (is_2srcdest(blend_func, src_factor, dest_factor, is_alpha))
      return supports_2src;

   if (blend_func != PIPE_BLEND_ADD && blend_func != PIPE_BLEND_SUBTRACT &&
       blend_func != PIPE_BLEND_REVERSE_SUBTRACT)
      return false;

   if (!factor_is_supported(src_factor) || !factor_is_supported(dest_factor))
      return false;

   /* Fixed function requires src/dest factors to match (up to invert) or be
    * zero/one.
    */
   enum pipe_blendfactor src = util_blendfactor_without_invert(src_factor);
   enum pipe_blendfactor dest = util_blendfactor_without_invert(dest_factor);

   return (src == dest) || (src == PIPE_BLENDFACTOR_ONE) ||
          (dest == PIPE_BLENDFACTOR_ONE);
}

static unsigned
blend_factor_constant_mask(enum pipe_blendfactor factor)
{
   factor = util_blendfactor_without_invert(factor);

   if (factor == PIPE_BLENDFACTOR_CONST_COLOR)
      return 0b0111; /* RGB */
   else if (factor == PIPE_BLENDFACTOR_CONST_ALPHA)
      return 0b1000; /* A */
   else
      return 0b0000; /* - */
}

unsigned
pan_blend_constant_mask(const struct pan_blend_equation eq)
{
   return blend_factor_constant_mask(eq.rgb_src_factor) |
          blend_factor_constant_mask(eq.rgb_dst_factor) |
          blend_factor_constant_mask(eq.alpha_src_factor) |
          blend_factor_constant_mask(eq.alpha_dst_factor);
}

/* Only "homogenous" (scalar or vector with all components equal) constants are
 * valid for fixed-function, so check for this condition */

bool
pan_blend_is_homogenous_constant(unsigned mask, const float *constants)
{
   float constant = pan_blend_get_constant(mask, constants);

   u_foreach_bit(i, mask) {
      if (constants[i] != constant)
         return false;
   }

   return true;
}

/* Determines if an equation can run in fixed function */

bool
pan_blend_can_fixed_function(const struct pan_blend_equation equation,
                             bool supports_2src)
{
   return !equation.blend_enable ||
          (can_fixed_function_equation(
              equation.rgb_func, equation.rgb_src_factor,
              equation.rgb_dst_factor, false, supports_2src) &&
           can_fixed_function_equation(
              equation.alpha_func, equation.alpha_src_factor,
              equation.alpha_dst_factor, true, supports_2src));
}

static enum mali_blend_operand_c
to_c_factor(enum pipe_blendfactor factor)
{
   switch (util_blendfactor_without_invert(factor)) {
   case PIPE_BLENDFACTOR_ONE:
      /* Extra invert to flip back in caller */
      return MALI_BLEND_OPERAND_C_ZERO;

   case PIPE_BLENDFACTOR_SRC_ALPHA:
      return MALI_BLEND_OPERAND_C_SRC_ALPHA;

   case PIPE_BLENDFACTOR_DST_ALPHA:
      return MALI_BLEND_OPERAND_C_DEST_ALPHA;

   case PIPE_BLENDFACTOR_SRC_COLOR:
      return MALI_BLEND_OPERAND_C_SRC;

   case PIPE_BLENDFACTOR_DST_COLOR:
      return MALI_BLEND_OPERAND_C_DEST;

   case PIPE_BLENDFACTOR_CONST_COLOR:
   case PIPE_BLENDFACTOR_CONST_ALPHA:
      return MALI_BLEND_OPERAND_C_CONSTANT;

   default:
      unreachable("Unsupported blend factor");
   }
}

static void
to_panfrost_function(enum pipe_blend_func blend_func,
                     enum pipe_blendfactor src_factor,
                     enum pipe_blendfactor dest_factor, bool is_alpha,
                     struct MALI_BLEND_FUNCTION *function)
{
   assert(can_fixed_function_equation(blend_func, src_factor, dest_factor,
                                      is_alpha, true));

   /* We handle ZERO/ONE specially since it's the hardware has 0 and can invert
    * to 1 but Gallium has 0 as the uninverted version.
    */
   bool src_inverted =
      util_blendfactor_is_inverted(src_factor) ^
      (util_blendfactor_without_invert(src_factor) == PIPE_BLENDFACTOR_ONE);

   bool dest_inverted =
      util_blendfactor_is_inverted(dest_factor) ^
      (util_blendfactor_without_invert(dest_factor) == PIPE_BLENDFACTOR_ONE);

   if (src_factor == PIPE_BLENDFACTOR_ZERO) {
      function->a = MALI_BLEND_OPERAND_A_ZERO;
      function->b = MALI_BLEND_OPERAND_B_DEST;
      if (blend_func == PIPE_BLEND_SUBTRACT)
         function->negate_b = true;
      function->invert_c = dest_inverted;
      function->c = to_c_factor(dest_factor);
   } else if (src_factor == PIPE_BLENDFACTOR_ONE) {
      function->a = MALI_BLEND_OPERAND_A_SRC;
      function->b = MALI_BLEND_OPERAND_B_DEST;
      if (blend_func == PIPE_BLEND_SUBTRACT)
         function->negate_b = true;
      else if (blend_func == PIPE_BLEND_REVERSE_SUBTRACT)
         function->negate_a = true;
      function->invert_c = dest_inverted;
      function->c = to_c_factor(dest_factor);
   } else if (dest_factor == PIPE_BLENDFACTOR_ZERO) {
      function->a = MALI_BLEND_OPERAND_A_ZERO;
      function->b = MALI_BLEND_OPERAND_B_SRC;
      if (blend_func == PIPE_BLEND_REVERSE_SUBTRACT)
         function->negate_b = true;
      function->invert_c = src_inverted;
      function->c = to_c_factor(src_factor);
   } else if (dest_factor == PIPE_BLENDFACTOR_ONE) {
      function->a = MALI_BLEND_OPERAND_A_DEST;
      function->b = MALI_BLEND_OPERAND_B_SRC;
      if (blend_func == PIPE_BLEND_SUBTRACT)
         function->negate_a = true;
      else if (blend_func == PIPE_BLEND_REVERSE_SUBTRACT)
         function->negate_b = true;
      function->invert_c = src_inverted;
      function->c = to_c_factor(src_factor);
   } else if (src_factor == dest_factor) {
      function->a = MALI_BLEND_OPERAND_A_ZERO;
      function->invert_c = src_inverted;
      function->c = to_c_factor(src_factor);

      switch (blend_func) {
      case PIPE_BLEND_ADD:
         function->b = MALI_BLEND_OPERAND_B_SRC_PLUS_DEST;
         break;
      case PIPE_BLEND_REVERSE_SUBTRACT:
         function->negate_b = true;
         FALLTHROUGH;
      case PIPE_BLEND_SUBTRACT:
         function->b = MALI_BLEND_OPERAND_B_SRC_MINUS_DEST;
         break;
      default:
         unreachable("Invalid blend function");
      }
   } else if (is_2srcdest(blend_func, src_factor, dest_factor, is_alpha)) {
      /* src*dest + dest*src = 2*src*dest = 0 + dest*(2*src) */
      function->a = MALI_BLEND_OPERAND_A_ZERO;
      function->b = MALI_BLEND_OPERAND_B_DEST;
      function->c = MALI_BLEND_OPERAND_C_SRC_X_2;
   } else {
      assert(util_blendfactor_without_invert(src_factor) ==
                util_blendfactor_without_invert(dest_factor) &&
             src_inverted != dest_inverted);

      function->a = MALI_BLEND_OPERAND_A_DEST;
      function->invert_c = src_inverted;
      function->c = to_c_factor(src_factor);

      switch (blend_func) {
      case PIPE_BLEND_ADD:
         function->b = MALI_BLEND_OPERAND_B_SRC_MINUS_DEST;
         break;
      case PIPE_BLEND_REVERSE_SUBTRACT:
         function->b = MALI_BLEND_OPERAND_B_SRC_PLUS_DEST;
         function->negate_b = true;
         break;
      case PIPE_BLEND_SUBTRACT:
         function->b = MALI_BLEND_OPERAND_B_SRC_PLUS_DEST;
         function->negate_a = true;
         break;
      default:
         unreachable("Invalid blend function\n");
      }
   }
}

bool
pan_blend_is_opaque(const struct pan_blend_equation equation)
{
   /* If a channel is masked out, we can't use opaque mode even if
    * blending is disabled, since we need a tilebuffer read in there */
   if (equation.color_mask != 0xF)
      return false;

   /* With nothing masked out, disabled bledning is opaque */
   if (!equation.blend_enable)
      return true;

   /* Also detect open-coded opaque blending */
   return equation.rgb_src_factor == PIPE_BLENDFACTOR_ONE &&
          equation.rgb_dst_factor == PIPE_BLENDFACTOR_ZERO &&
          (equation.rgb_func == PIPE_BLEND_ADD ||
           equation.rgb_func == PIPE_BLEND_SUBTRACT) &&
          equation.alpha_src_factor == PIPE_BLENDFACTOR_ONE &&
          equation.alpha_dst_factor == PIPE_BLENDFACTOR_ZERO &&
          (equation.alpha_func == PIPE_BLEND_ADD ||
           equation.alpha_func == PIPE_BLEND_SUBTRACT);
}

/* Check if a factor represents a constant value of val, assuming src_alpha is
 * the given constant.
 */

static inline bool
is_factor_01(enum pipe_blendfactor factor, unsigned val, unsigned srca)
{
   assert(val == 0 || val == 1);
   assert(srca == 0 || srca == 1);

   switch (factor) {
   case PIPE_BLENDFACTOR_ZERO:
      return (val == 0);

   case PIPE_BLENDFACTOR_ONE:
      return (val == 1);

   case PIPE_BLENDFACTOR_SRC_ALPHA:
      return (val == srca);

   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
      return (val == (1 - srca));

   default:
      return false;
   }
}

/* Returns if src alpha = 0 implies the blended colour equals the destination
 * colour. Suppose source alpha = 0 and consider cases.
 *
 * Additive blending: Equivalent to D = S * f_s + D * f_d for all D and all S
 * with S_a = 0, for each component. For the alpha component (if it unmasked),
 * we have S_a = 0 so this reduces to D = D * f_d <===> f_d = 1. For RGB
 * components (if unmasked), we need f_s = 0 and f_d = 1.
 *
 * Subtractive blending: Fails in general (D = S * f_S - D * f_D). We
 * would need f_S = 0 and f_D = -1, which is not valid in the APIs.
 *
 * Reverse subtractive blending (D = D * f_D - S * f_S), we need f_D = 1
 * and f_S = 0 up to masking. This is the same as additive blending.
 *
 * Min/max: Fails in general on the RGB components.
 */

bool
pan_blend_alpha_zero_nop(const struct pan_blend_equation eq)
{
   if (eq.rgb_func != PIPE_BLEND_ADD &&
       eq.rgb_func != PIPE_BLEND_REVERSE_SUBTRACT)
      return false;

   if (eq.color_mask & 0x8) {
      if (!is_factor_01(eq.alpha_dst_factor, 1, 0))
         return false;
   }

   if (eq.color_mask & 0x7) {
      if (!is_factor_01(eq.rgb_dst_factor, 1, 0))
         return false;

      if (!is_factor_01(eq.rgb_src_factor, 0, 0))
         return false;
   }

   return true;
}

/* Returns if src alpha = 1 implies the blended colour equals the source
 * colour. Suppose source alpha = 1 and consider cases.
 *
 * Additive blending: S = S * f_s + D * f_d. We need f_s = 1 and f_d = 0.
 *
 * Subtractive blending: S = S * f_s - D * f_d. Same as additive blending.
 *
 * Reverse subtractive blending: S = D * f_d - S * f_s. Fails in general since
 * it would require f_s = -1, which is not valid in the APIs.
 *
 * Min/max: Fails in general on the RGB components.
 *
 * Note if any component is masked, we can't use a store.
 */

bool
pan_blend_alpha_one_store(const struct pan_blend_equation eq)
{
   if (eq.rgb_func != PIPE_BLEND_ADD && eq.rgb_func != PIPE_BLEND_SUBTRACT)
      return false;

   if (eq.color_mask != 0xf)
      return false;

   return is_factor_01(eq.rgb_src_factor, 1, 1) &&
          is_factor_01(eq.alpha_src_factor, 1, 1) &&
          is_factor_01(eq.rgb_dst_factor, 0, 1) &&
          is_factor_01(eq.alpha_dst_factor, 0, 1);
}

static bool
is_dest_factor(enum pipe_blendfactor factor, bool alpha)
{
   factor = util_blendfactor_without_invert(factor);

   return factor == PIPE_BLENDFACTOR_DST_ALPHA ||
          factor == PIPE_BLENDFACTOR_DST_COLOR ||
          (factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE && !alpha);
}

/* Determines if a blend equation reads back the destination. This can occur by
 * explicitly referencing the destination in the blend equation, or by using a
 * partial writemask. */

bool
pan_blend_reads_dest(const struct pan_blend_equation equation)
{
   if (equation.color_mask && equation.color_mask != 0xF)
      return true;

   if (!equation.blend_enable)
      return false;

   return is_dest_factor(equation.rgb_src_factor, false) ||
          is_dest_factor(equation.alpha_src_factor, true) ||
          equation.rgb_dst_factor != PIPE_BLENDFACTOR_ZERO ||
          equation.alpha_dst_factor != PIPE_BLENDFACTOR_ZERO;
}

/* Create the descriptor for a fixed blend mode given the corresponding API
 * state. Assumes the equation can be represented as fixed-function. */

void
pan_blend_to_fixed_function_equation(const struct pan_blend_equation equation,
                                     struct MALI_BLEND_EQUATION *out)
{
   /* If no blending is enabled, default back on `replace` mode */
   if (!equation.blend_enable) {
      out->color_mask = equation.color_mask;
      out->rgb.a = MALI_BLEND_OPERAND_A_SRC;
      out->rgb.b = MALI_BLEND_OPERAND_B_SRC;
      out->rgb.c = MALI_BLEND_OPERAND_C_ZERO;
      out->alpha.a = MALI_BLEND_OPERAND_A_SRC;
      out->alpha.b = MALI_BLEND_OPERAND_B_SRC;
      out->alpha.c = MALI_BLEND_OPERAND_C_ZERO;
      return;
   }

   /* Compile the fixed-function blend */
   to_panfrost_function(equation.rgb_func, equation.rgb_src_factor,
                        equation.rgb_dst_factor, false, &out->rgb);
   to_panfrost_function(equation.alpha_func, equation.alpha_src_factor,
                        equation.alpha_dst_factor, true, &out->alpha);

   out->color_mask = equation.color_mask;
}

uint32_t
pan_pack_blend(const struct pan_blend_equation equation)
{
   STATIC_ASSERT(sizeof(uint32_t) == MALI_BLEND_EQUATION_LENGTH);

   uint32_t out = 0;

   pan_pack(&out, BLEND_EQUATION, cfg) {
      pan_blend_to_fixed_function_equation(equation, &cfg);
   }

   return out;
}

DERIVE_HASH_TABLE(pan_blend_shader_key);

void
pan_blend_shader_cache_init(struct pan_blend_shader_cache *cache,
                            unsigned gpu_id)
{
   cache->gpu_id = gpu_id;
   cache->shaders = pan_blend_shader_key_table_create(NULL);
   pthread_mutex_init(&cache->lock, NULL);
}

void
pan_blend_shader_cache_cleanup(struct pan_blend_shader_cache *cache)
{
   _mesa_hash_table_destroy(cache->shaders, NULL);
   pthread_mutex_destroy(&cache->lock);
}

#else /* ifndef PAN_ARCH */

static const char *
logicop_str(enum pipe_logicop logicop)
{
   switch (logicop) {
   case PIPE_LOGICOP_CLEAR:
      return "clear";
   case PIPE_LOGICOP_NOR:
      return "nor";
   case PIPE_LOGICOP_AND_INVERTED:
      return "and-inverted";
   case PIPE_LOGICOP_COPY_INVERTED:
      return "copy-inverted";
   case PIPE_LOGICOP_AND_REVERSE:
      return "and-reverse";
   case PIPE_LOGICOP_INVERT:
      return "invert";
   case PIPE_LOGICOP_XOR:
      return "xor";
   case PIPE_LOGICOP_NAND:
      return "nand";
   case PIPE_LOGICOP_AND:
      return "and";
   case PIPE_LOGICOP_EQUIV:
      return "equiv";
   case PIPE_LOGICOP_NOOP:
      return "noop";
   case PIPE_LOGICOP_OR_INVERTED:
      return "or-inverted";
   case PIPE_LOGICOP_COPY:
      return "copy";
   case PIPE_LOGICOP_OR_REVERSE:
      return "or-reverse";
   case PIPE_LOGICOP_OR:
      return "or";
   case PIPE_LOGICOP_SET:
      return "set";
   default:
      unreachable("Invalid logicop\n");
   }
}

static void
get_equation_str(const struct pan_blend_rt_state *rt_state, char *str,
                 unsigned len)
{
   const char *funcs[] = {
      "add", "sub", "reverse_sub", "min", "max",
   };
   const char *factors[] = {
      "",           "one",           "src_color",   "src_alpha",   "dst_alpha",
      "dst_color",  "src_alpha_sat", "const_color", "const_alpha", "src1_color",
      "src1_alpha",
   };
   int ret;

   if (!rt_state->equation.blend_enable) {
      ret = snprintf(str, len, "replace(%s%s%s%s)",
                     (rt_state->equation.color_mask & 1) ? "R" : "",
                     (rt_state->equation.color_mask & 2) ? "G" : "",
                     (rt_state->equation.color_mask & 4) ? "B" : "",
                     (rt_state->equation.color_mask & 8) ? "A" : "");
      assert(ret > 0);
      return;
   }

   if (rt_state->equation.color_mask & 7) {
      assert(rt_state->equation.rgb_func < ARRAY_SIZE(funcs));
      ret = snprintf(
         str, len, "%s%s%s(func=%s,src_factor=%s%s,dst_factor=%s%s)%s",
         (rt_state->equation.color_mask & 1) ? "R" : "",
         (rt_state->equation.color_mask & 2) ? "G" : "",
         (rt_state->equation.color_mask & 4) ? "B" : "",
         funcs[rt_state->equation.rgb_func],
         util_blendfactor_is_inverted(rt_state->equation.rgb_src_factor) ? "-"
                                                                         : "",
         factors[util_blendfactor_without_invert(
            rt_state->equation.rgb_src_factor)],
         util_blendfactor_is_inverted(rt_state->equation.rgb_dst_factor) ? "-"
                                                                         : "",
         factors[util_blendfactor_without_invert(
            rt_state->equation.rgb_dst_factor)],
         rt_state->equation.color_mask & 8 ? ";" : "");
      assert(ret > 0);
      str += ret;
      len -= ret;
   }

   if (rt_state->equation.color_mask & 8) {
      assert(rt_state->equation.alpha_func < ARRAY_SIZE(funcs));
      ret = snprintf(
         str, len, "A(func=%s,src_factor=%s%s,dst_factor=%s%s)",
         funcs[rt_state->equation.alpha_func],
         util_blendfactor_is_inverted(rt_state->equation.alpha_src_factor) ? "-"
                                                                           : "",
         factors[util_blendfactor_without_invert(
            rt_state->equation.alpha_src_factor)],
         util_blendfactor_is_inverted(rt_state->equation.alpha_dst_factor) ? "-"
                                                                           : "",
         factors[util_blendfactor_without_invert(
            rt_state->equation.alpha_dst_factor)]);
      assert(ret > 0);
      str += ret;
      len -= ret;
   }
}

static bool
pan_inline_blend_constants(nir_builder *b, nir_intrinsic_instr *intr,
                           void *data)
{
   if (intr->intrinsic != nir_intrinsic_load_blend_const_color_rgba)
      return false;

   float *floats = data;
   const nir_const_value constants[4] = {
      nir_const_value_for_float(floats[0], 32),
      nir_const_value_for_float(floats[1], 32),
      nir_const_value_for_float(floats[2], 32),
      nir_const_value_for_float(floats[3], 32)};

   b->cursor = nir_after_instr(&intr->instr);
   nir_def *constant = nir_build_imm(b, 4, 32, constants);
   nir_def_replace(&intr->def, constant);
   return true;
}

nir_shader *
GENX(pan_blend_create_shader)(const struct pan_blend_state *state,
                              nir_alu_type src0_type, nir_alu_type src1_type,
                              unsigned rt)
{
   const struct pan_blend_rt_state *rt_state = &state->rts[rt];
   char equation_str[128] = {0};

   get_equation_str(rt_state, equation_str, sizeof(equation_str));

   nir_builder b = nir_builder_init_simple_shader(
      MESA_SHADER_FRAGMENT, GENX(pan_shader_get_compiler_options)(),
      "pan_blend(rt=%d,fmt=%s,nr_samples=%d,%s=%s)", rt,
      util_format_name(rt_state->format), rt_state->nr_samples,
      state->logicop_enable ? "logicop" : "equation",
      state->logicop_enable ? logicop_str(state->logicop_func) : equation_str);

   const struct util_format_description *format_desc =
      util_format_description(rt_state->format);
   nir_alu_type nir_type = pan_unpacked_type_for_format(format_desc);

   /* Bifrost/Valhall support 16-bit and 32-bit register formats for
    * LD_TILE/ST_TILE/BLEND, but do not support 8-bit. Rather than making
    * the fragment output 8-bit and inserting extra conversions in the
    * compiler, promote the output to 16-bit. The larger size is still
    * compatible with correct conversion semantics.
    */
   if (PAN_ARCH >= 6 && nir_alu_type_get_type_size(nir_type) == 8)
      nir_type = nir_alu_type_get_base_type(nir_type) | 16;

   nir_lower_blend_options options = {
      .logicop_enable = state->logicop_enable,
      .logicop_func = state->logicop_func,
   };

   options.rt[rt].colormask = rt_state->equation.color_mask;
   options.format[rt] = rt_state->format;

   if (!rt_state->equation.blend_enable) {
      static const nir_lower_blend_channel replace = {
         .func = PIPE_BLEND_ADD,
         .src_factor = PIPE_BLENDFACTOR_ONE,
         .dst_factor = PIPE_BLENDFACTOR_ZERO,
      };

      options.rt[rt].rgb = replace;
      options.rt[rt].alpha = replace;
   } else {
      options.rt[rt].rgb.func = rt_state->equation.rgb_func;
      options.rt[rt].rgb.src_factor = rt_state->equation.rgb_src_factor;
      options.rt[rt].rgb.dst_factor = rt_state->equation.rgb_dst_factor;
      options.rt[rt].alpha.func = rt_state->equation.alpha_func;
      options.rt[rt].alpha.src_factor = rt_state->equation.alpha_src_factor;
      options.rt[rt].alpha.dst_factor = rt_state->equation.alpha_dst_factor;
   }

   nir_def *pixel = nir_load_barycentric_pixel(&b, 32, .interp_mode = 1);
   nir_def *zero = nir_imm_int(&b, 0);

   for (unsigned i = 0; i < 2; ++i) {
      nir_alu_type src_type =
         (i == 1 ? src1_type : src0_type) ?: nir_type_float32;

      /* HACK: workaround buggy TGSI shaders (u_blitter) */
      src_type = nir_alu_type_get_base_type(nir_type) |
                 nir_alu_type_get_type_size(src_type);

      nir_def *src = nir_load_interpolated_input(
         &b, 4, nir_alu_type_get_type_size(src_type), pixel, zero,
         .io_semantics.location = i ? VARYING_SLOT_VAR0 : VARYING_SLOT_COL0,
         .io_semantics.num_slots = 1, .base = i, .dest_type = src_type);

      /* On Midgard, the blend shader is responsible for format conversion.
       * As the OpenGL spec requires integer conversions to saturate, we must
       * saturate ourselves here. On Bifrost and later, the conversion
       * hardware handles this automatically.
       */
      nir_alu_type T = nir_alu_type_get_base_type(nir_type);
      bool should_saturate = (PAN_ARCH <= 5) && (T != nir_type_float);
      src = nir_convert_with_rounding(&b, src, T, nir_type,
                                      nir_rounding_mode_undef, should_saturate);

      nir_store_output(&b, src, zero, .write_mask = BITFIELD_MASK(4),
                       .src_type = nir_type,
                       .io_semantics.location = FRAG_RESULT_DATA0 + rt,
                       .io_semantics.num_slots = 1,
                       .io_semantics.dual_source_blend_index = i);
   }

   b.shader->info.io_lowered = true;

   NIR_PASS_V(b.shader, nir_lower_blend, &options);

   return b.shader;
}

#if PAN_ARCH >= 6
uint64_t
GENX(pan_blend_get_internal_desc)(enum pipe_format fmt, unsigned rt,
                                  unsigned force_size, bool dithered)
{
   const struct util_format_description *desc = util_format_description(fmt);
   uint64_t res;

   pan_pack(&res, INTERNAL_BLEND, cfg) {
      cfg.mode = MALI_BLEND_MODE_OPAQUE;
      cfg.fixed_function.num_comps = desc->nr_channels;
      cfg.fixed_function.rt = rt;

      nir_alu_type T = pan_unpacked_type_for_format(desc);

      if (force_size)
         T = nir_alu_type_get_base_type(T) | force_size;

      switch (T) {
      case nir_type_float16:
         cfg.fixed_function.conversion.register_format =
            MALI_REGISTER_FILE_FORMAT_F16;
         break;
      case nir_type_float32:
         cfg.fixed_function.conversion.register_format =
            MALI_REGISTER_FILE_FORMAT_F32;
         break;
      case nir_type_int8:
      case nir_type_int16:
         cfg.fixed_function.conversion.register_format =
            MALI_REGISTER_FILE_FORMAT_I16;
         break;
      case nir_type_int32:
         cfg.fixed_function.conversion.register_format =
            MALI_REGISTER_FILE_FORMAT_I32;
         break;
      case nir_type_uint8:
      case nir_type_uint16:
         cfg.fixed_function.conversion.register_format =
            MALI_REGISTER_FILE_FORMAT_U16;
         break;
      case nir_type_uint32:
         cfg.fixed_function.conversion.register_format =
            MALI_REGISTER_FILE_FORMAT_U32;
         break;
      default:
         unreachable("Invalid format");
      }

      cfg.fixed_function.conversion.memory_format =
         GENX(panfrost_dithered_format_from_pipe_format)(fmt, dithered);
   }

   return res;
}

static bool
inline_rt_conversion(nir_builder *b, nir_intrinsic_instr *intr, void *data)
{
   if (intr->intrinsic != nir_intrinsic_load_rt_conversion_pan)
      return false;

   enum pipe_format *formats = data;
   unsigned rt = nir_intrinsic_base(intr);
   unsigned size = nir_alu_type_get_type_size(nir_intrinsic_src_type(intr));
   uint64_t conversion =
      GENX(pan_blend_get_internal_desc)(formats[rt], rt, size, false);

   b->cursor = nir_after_instr(&intr->instr);
   nir_def_rewrite_uses(&intr->def, nir_imm_int(b, conversion >> 32));
   return true;
}

bool
GENX(pan_inline_rt_conversion)(nir_shader *s, enum pipe_format *formats)
{
   return nir_shader_intrinsics_pass(s, inline_rt_conversion,
                                     nir_metadata_control_flow, formats);
}
#endif

struct pan_blend_shader_variant *
GENX(pan_blend_get_shader_locked)(struct pan_blend_shader_cache *cache,
                                  const struct pan_blend_state *state,
                                  nir_alu_type src0_type,
                                  nir_alu_type src1_type, unsigned rt)
{
   struct pan_blend_shader_key key = {
      .format = state->rts[rt].format,
      .src0_type = src0_type,
      .src1_type = src1_type,
      .rt = rt,
      .has_constants = pan_blend_constant_mask(state->rts[rt].equation) != 0,
      .logicop_enable = state->logicop_enable,
      .logicop_func = state->logicop_func,
      .nr_samples = state->rts[rt].nr_samples,
      .equation = state->rts[rt].equation,
   };

   /* Blend shaders should only be used for blending on Bifrost onwards */
   assert(PAN_ARCH <= 5 || state->logicop_enable ||
          !pan_blend_is_opaque(state->rts[rt].equation));
   assert(state->rts[rt].equation.color_mask != 0);

   struct hash_entry *he =
      _mesa_hash_table_search(cache->shaders, &key);
   struct pan_blend_shader *shader = he ? he->data : NULL;

   if (!shader) {
      shader = rzalloc(cache->shaders, struct pan_blend_shader);
      shader->key = key;
      list_inithead(&shader->variants);
      _mesa_hash_table_insert(cache->shaders, &shader->key, shader);
   }

   list_for_each_entry(struct pan_blend_shader_variant, iter, &shader->variants,
                       node) {
      if (!key.has_constants ||
          !memcmp(iter->constants, state->constants, sizeof(iter->constants))) {
         return iter;
      }
   }

   struct pan_blend_shader_variant *variant = NULL;

   if (shader->nvariants < PAN_BLEND_SHADER_MAX_VARIANTS) {
      variant = rzalloc(shader, struct pan_blend_shader_variant);
      util_dynarray_init(&variant->binary, variant);
      list_add(&variant->node, &shader->variants);
      shader->nvariants++;
   } else {
      variant = list_last_entry(&shader->variants,
                                struct pan_blend_shader_variant, node);
      list_del(&variant->node);
      list_add(&variant->node, &shader->variants);
      util_dynarray_clear(&variant->binary);
   }

   memcpy(variant->constants, state->constants, sizeof(variant->constants));

   nir_shader *nir =
      GENX(pan_blend_create_shader)(state, src0_type, src1_type, rt);

   nir_shader_intrinsics_pass(nir, pan_inline_blend_constants,
                              nir_metadata_control_flow,
                              (void *)state->constants);

   /* Compile the NIR shader */
   struct panfrost_compile_inputs inputs = {
      .gpu_id = cache->gpu_id,
      .is_blend = true,
      .blend.nr_samples = key.nr_samples,
   };

   enum pipe_format rt_formats[8] = {0};
   rt_formats[rt] = key.format;

#if PAN_ARCH >= 6
   inputs.blend.bifrost_blend_desc =
      GENX(pan_blend_get_internal_desc)(key.format, key.rt, 0, false);
#endif

   struct pan_shader_info info;
   pan_shader_preprocess(nir, inputs.gpu_id);

#if PAN_ARCH >= 6
   NIR_PASS_V(nir, GENX(pan_inline_rt_conversion), rt_formats);
#else
   NIR_PASS_V(nir, pan_lower_framebuffer, rt_formats,
              pan_raw_format_mask_midgard(rt_formats), MAX2(key.nr_samples, 1),
              cache->gpu_id < 0x700);
#endif

   GENX(pan_shader_compile)(nir, &inputs, &variant->binary, &info);

   variant->work_reg_count = info.work_reg_count;

#if PAN_ARCH <= 5
   variant->first_tag = info.midgard.first_tag;
#endif

   ralloc_free(nir);

   return variant;
}
#endif /* ifndef PAN_ARCH */