Create dedicated op for inverted condition masks.

This allows side-effecting ternaries and else-clauses to use one fewer instruction, and it won't add a lot of code size. Change-Id: I831ab059ba6d4c368dde682130ca0b49cce0cbb1 Reviewed-on: https://skia-review.googlesource.com/c/skia/+/684856 Reviewed-by: Jim Van Verth <[email protected]> Commit-Queue: Jim Van Verth <[email protected]> Auto-Submit: John Stiles <[email protected]>
armansito · Apr 28, 2023 · e88e5de · e88e5de
1 parent 49412cb
commit e88e5de
Show file tree

Hide file tree

Showing 11 changed files with 82 additions and 104 deletions.
diff --git a/src/core/SkRasterPipelineOpList.h b/src/core/SkRasterPipelineOpList.h
@@ -58,10 +58,11 @@
 
 #define SK_RASTER_PIPELINE_OPS_SKSL(M)                                                        \
     M(init_lane_masks) M(store_device_xy01) M(exchange_src)                                   \
-    M(load_condition_mask) M(store_condition_mask) M(merge_condition_mask)                    \
-    M(load_loop_mask)      M(store_loop_mask)      M(mask_off_loop_mask)                      \
-    M(reenable_loop_mask)  M(merge_loop_mask)      M(case_op)   M(continue_op)                \
-    M(load_return_mask)    M(store_return_mask)    M(mask_off_return_mask)                    \
+    M(load_condition_mask)  M(store_condition_mask)                                           \
+    M(merge_condition_mask) M(merge_inv_condition_mask)                                       \
+    M(load_loop_mask)       M(store_loop_mask)      M(mask_off_loop_mask)                     \
+    M(reenable_loop_mask)   M(merge_loop_mask)      M(case_op)   M(continue_op)               \
+    M(load_return_mask)     M(store_return_mask)    M(mask_off_return_mask)                   \
     M(branch_if_all_lanes_active) M(branch_if_any_lanes_active) M(branch_if_no_lanes_active)  \
     M(branch_if_no_active_lanes_eq) M(jump)                                                   \
     M(bitwise_and_n_ints)                                                                     \

diff --git a/src/opts/SkRasterPipeline_opts.h b/src/opts/SkRasterPipeline_opts.h
@@ -3356,6 +3356,12 @@ STAGE_TAIL(merge_condition_mask, I32* ptr) {
     update_execution_mask();
 }
 
+STAGE_TAIL(merge_inv_condition_mask, I32* ptr) {
+    // Set the condition-mask to the intersection of the first mask and the inverse of the second.
+    r = sk_bit_cast<F>(ptr[0] & ~ptr[1]);
+    update_execution_mask();
+}
+
 STAGE_TAIL(load_loop_mask, F* ctx) {
     g = sk_unaligned_load<F>(ctx);
     update_execution_mask();

diff --git a/src/sksl/codegen/SkSLRasterPipelineBuilder.cpp b/src/sksl/codegen/SkSLRasterPipelineBuilder.cpp
@@ -1965,9 +1965,10 @@ void Program::makeStages(TArray<Stage>* pipeline,
                 pipeline->push_back({ProgramOp::load_condition_mask, src});
                 break;
             }
-            case BuilderOp::merge_condition_mask: {
+            case BuilderOp::merge_condition_mask:
+            case BuilderOp::merge_inv_condition_mask: {
                 float* ptr = tempStackPtr - (2 * N);
-                pipeline->push_back({ProgramOp::merge_condition_mask, ptr});
+                pipeline->push_back({(ProgramOp)inst.fOp, ptr});
                 break;
             }
             case BuilderOp::push_loop_mask: {
@@ -2838,6 +2839,7 @@ void Program::dump(SkWStream* out) const {
                 break;
             }
             case POp::merge_condition_mask:
+            case POp::merge_inv_condition_mask:
             case POp::add_float:   case POp::add_int:
             case POp::sub_float:   case POp::sub_int:
             case POp::mul_float:   case POp::mul_int:
@@ -3042,6 +3044,10 @@ void Program::dump(SkWStream* out) const {
                 opText = "CondMask = " + opArg1 + " & " + opArg2;
                 break;
 
+            case POp::merge_inv_condition_mask:
+                opText = "CondMask = " + opArg1 + " & ~" + opArg2;
+                break;
+
             case POp::load_loop_mask:
                 opText = "LoopMask = " + opArg1;
                 break;

diff --git a/src/sksl/codegen/SkSLRasterPipelineBuilder.h b/src/sksl/codegen/SkSLRasterPipelineBuilder.h
@@ -562,6 +562,11 @@ class Builder {
         fInstructions.push_back({BuilderOp::merge_condition_mask, {}});
     }
 
+    void merge_inv_condition_mask() {
+        SkASSERT(this->executionMaskWritesAreEnabled());
+        fInstructions.push_back({BuilderOp::merge_inv_condition_mask, {}});
+    }
+
     void push_loop_mask() {
         SkASSERT(this->executionMaskWritesAreEnabled());
         fInstructions.push_back({BuilderOp::push_loop_mask, {}});

diff --git a/src/sksl/codegen/SkSLRasterPipelineCodeGenerator.cpp b/src/sksl/codegen/SkSLRasterPipelineCodeGenerator.cpp
@@ -1882,11 +1882,8 @@ bool Generator::writeIfStatement(const IfStatement& i) {
     }
 
     if (i.ifFalse()) {
-        // Negate the test-condition, then reapply it to the condition-mask.
-        // Then, run the if-false branch.
-        fBuilder.push_constant_u(~0);
-        fBuilder.binary_op(BuilderOp::bitwise_xor_n_ints, /*slots=*/1);
-        fBuilder.merge_condition_mask();
+        // Apply the inverse condition-mask. Then run the if-false branch.
+        fBuilder.merge_inv_condition_mask();
         if (!this->writeStatement(*i.ifFalse())) {
             return unsupported();
         }
@@ -3563,11 +3560,9 @@ bool Generator::pushTernaryExpression(const Expression& test,
             return unsupported();
         }
 
-        // Switch back to the test-expression stack temporarily, and negate the test condition.
+        // Switch back to the test-expression stack and apply the inverted test condition.
         testStack.enter();
-        fBuilder.push_constant_u(~0);
-        fBuilder.binary_op(BuilderOp::bitwise_xor_n_ints, /*slots=*/1);
-        fBuilder.merge_condition_mask();
+        fBuilder.merge_inv_condition_mask();
         testStack.exit();
 
         // Push the false-expression onto the primary stack, immediately after the true-expression.

diff --git a/tests/RasterPipelineBuilderTest.cpp b/tests/RasterPipelineBuilderTest.cpp
@@ -101,6 +101,7 @@ DEF_TEST(RasterPipelineBuilderPushPopMaskRegisters, r) {
     builder.push_loop_mask();              // push into 1
     builder.push_return_mask();            // push into 2
     builder.merge_condition_mask();        // set the condition-mask to 1 & 2
+    builder.merge_inv_condition_mask();    // set the condition-mask to 1 & ~2
     builder.pop_condition_mask();          // pop from 2
     builder.merge_loop_mask();             // mask off the loop-mask against 1
     builder.push_condition_mask();         // push into 2
@@ -121,6 +122,7 @@ R"(store_condition_mask           $0 = CondMask
 store_loop_mask                $1 = LoopMask
 store_return_mask              $2 = RetMask
 merge_condition_mask           CondMask = $1 & $2
+merge_inv_condition_mask       CondMask = $1 & ~$2
 load_condition_mask            CondMask = $2
 merge_loop_mask                LoopMask &= $1
 store_condition_mask           $2 = CondMask

diff --git a/tests/sksl/folding/ShortCircuitBoolFolding.skrp b/tests/sksl/folding/ShortCircuitBoolFolding.skrp
@@ -11,8 +11,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -25,8 +24,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
@@ -37,8 +35,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -50,8 +47,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -62,8 +58,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -75,8 +70,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
@@ -88,8 +82,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
@@ -100,8 +93,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -112,8 +104,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -126,8 +117,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
@@ -138,8 +128,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -151,8 +140,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -163,8 +151,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -176,8 +163,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
@@ -189,8 +175,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
@@ -201,8 +186,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
@@ -220,8 +204,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
@@ -238,8 +221,7 @@ merge_condition_mask           CondMask = $0 & $1
 copy_slot_unmasked             $2 = _1_ok
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _1_ok = Mask($2)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 copy_slot_unmasked             $2 = _2_bad
 add_imm_int                    $2 += 0x00000001
 copy_slot_masked               _2_bad = Mask($2)

diff --git a/tests/sksl/shared/CompileTimeConstantVariables.skrp b/tests/sksl/shared/CompileTimeConstantVariables.skrp
@@ -18,40 +18,37 @@ cmpeq_imm_int                  $1 = equal($1, 0)
 merge_condition_mask           CondMask = $0 & $1
 splat_4_constants              $2..5 = 0x4008F5C3 (2.14)
 copy_4_slots_masked            [main].result = Mask($2..5)
-bitwise_xor_imm_int            $1 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $0 & $1
+merge_inv_condition_mask       CondMask = $0 & ~$1
 store_condition_mask           $2 = CondMask
 copy_slot_unmasked             $3 = integerInput
 cmpeq_imm_int                  $3 = equal($3, 0x00000001)
 merge_condition_mask           CondMask = $2 & $3
 copy_4_uniforms                $4..7 = colorGreen
 copy_4_slots_masked            [main].result = Mask($4..7)
-bitwise_xor_imm_int            $3 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $2 & $3
+merge_inv_condition_mask       CondMask = $2 & ~$3
 store_condition_mask           $4 = CondMask
 copy_slot_unmasked             $5 = integerInput
 cmpeq_imm_int                  $5 = equal($5, 0x00000002)
 merge_condition_mask           CondMask = $4 & $5
 copy_4_slots_unmasked          $6..9 = kConstVec
 copy_4_slots_masked            [main].result = Mask($6..9)
-bitwise_xor_imm_int            $5 ^= 0xFFFFFFFF
-merge_condition_mask           CondMask = $4 & $5
+merge_inv_condition_mask       CondMask = $4 & ~$5
 copy_constant                  $6 = 0x4048F5C3 (3.14)
 copy_uniform                   $7 = colorGreen(0)
 mul_imm_float                  $7 *= 0x4048F5C3 (3.14)
 cmplt_float                    $6 = lessThan($6, $7)
-branch_if_no_active_lanes_eq   branch +4 (label 0 at #47) if no lanes of $6 == 0xFFFFFFFF
+branch_if_no_active_lanes_eq   branch +4 (label 0 at #44) if no lanes of $6 == 0xFFFFFFFF
 splat_4_constants              $7..10 = 0x4048F5C3 (3.14)
 copy_4_slots_masked            [main].result = Mask($7..10)
-jump                           jump +15 (label 1 at #61)
+jump                           jump +15 (label 1 at #58)
 label                          label 0
 copy_uniform                   $7 = colorGreen(0)
 mul_imm_float                  $7 *= 0x4008F5C3 (2.14)
 cmple_imm_float                $7 = lessThanEqual($7, 0x4008F5C3 (2.14))
-branch_if_no_active_lanes_eq   branch +4 (label 2 at #55) if no lanes of $7 == 0xFFFFFFFF
+branch_if_no_active_lanes_eq   branch +4 (label 2 at #52) if no lanes of $7 == 0xFFFFFFFF
 splat_4_constants              $8..11 = 0
 copy_4_slots_masked            [main].result = Mask($8..11)
-jump                           jump +6 (label 3 at #60)
+jump                           jump +6 (label 3 at #57)
 label                          label 0x00000002
 copy_constant                  $8 = 0x3F800000 (1.0)
 splat_2_constants              $9..10 = 0